Roundup: a review of the super GeForce GTX 580s from Asus, EVGA, Gainward, Gigabyte, MSI and Zotac - BeHardware
>> Graphics cards

Written by Damien Triolet

Published on October 8, 2011


Page 1


Armed with a new test protocol, we decided to look at the high-end variants of the GeForce GTX 580 range on offer from NVIDIA's various partners: the Asus GTX 580 Matrix Platinum, EVGA GTX 580 Classified, Gigabyte GTX 580 SOC, MSI GTX 580 Lightning and Zotac GTX 580 AMP≤! Edition take on the reference card with an additional comparison with the Gainward Phantom.

Do better than the reference card
Following the GeForce GTX 480 that was much-criticised for its cooling system, NVIDIA worked hard to make sure that the revision model didn't come in for the same criticism. The cooler they came up with has proven very effective, thanks to the use of a vapour chamber and fairing designed to reduce noise levels.

While an effective reference cooler is good news for users, it makes the task of partners looking to mark themselves out or reduce costs, that much more difficult as itís hard to improve on the reference solution without making the card cost a lot more, or equal it at a lower tariff.

To keep production costs on their high-end variants down, the biggest manufacturers have elaborated more or less generic cooling systems that can be used across as many models as possible so as to benefit from economy of scale and lower R&D costs. Other partners use generic coolers from third-party producers such as Arctic or Zalman.

The cooling system isnít all that matters on a high-end model. You should also have a sturdy power stage to extend overclocking potential and various other settings for more advanced users. Note that when it comes to overclocking, itís still something of a lottery, even on these exclusive models, as their GPUs may or may not have been handpicked.

With the Matrix Platinum, Classified, SOC and Lightning, Asus, EVGA, Gigabyte and MSI are offering highly evolved and customised models. Zotac havenít gone to the same lengths and have settled for a Zalman cooling system for the AMP≤! Edition. We also included the Gainward Phantom, a very popular model though we werenít really convinced by its cooling system. We see it as a low-cost model, less effective than the reference design, which however has been very aesthetically designed. This roundup was also an opportunity to see how the Phantom does in the new test protocol in comparison to the old one.

Page 2
Test protocol 2.0 beta

Test protocol 2.0 beta
After two and a half years of loyal service, we decided to pension off the Sonata 3 and the first version of our full test protocol dedicated to measuring the thermal characteristics of graphics cards. Over this period we used thermal imaging to examine in detail at no fewer than thirty-seven graphics solutions and this has formed a useful database and given us the ability to position the results of new solutions objectively.

Test protocol 2.0 has also been designed to last and is in fact still under development. This report was a good debugging exercise for us and highlighted some areas which need to be improved.

This new test protocol uses a Cooler Master RC-690 II Advanced casing, radically different to the Sonota 3. With many perforations, itís also better equipped cooler wise, with a 140 mm suction fan at the front and 120 mm and 140 mm extraction fans at the back, in place of the single 120 mm fan on the Sonata 3.

Instead of the X48 and Q6600, weíre now on a P67 and Core i7-2600 platform. We went for the Asus Sabertooth P67 motherboard which has an additional slot between the CPU and the first graphics port, which makes it easier to take infrared photos. We did however remove its plastic protection which was stopping us from viewing the temperature of various areas of the PCB. While previously we used the Intel boxed cooler, here we went for a Big Shuriken from Scythe, which is more effective and isnít subject to the size issues you get with tower models, which would have been a problem for taking the photos.

We used a Scythe Kaze Master rheobus to control the speed of the CPU and box fans. The CPU fan was thus fixed at 900 RPM and what we imagined to be 9V for the casing fans, which donít have speed monitoring.

We opted for a high-end power supply: the Seasonic X850. It has the advantage of being passive when the energy consumption is under 200W, namely when our system is at idle. We used an OCZ Vertex 2 64 GB for the system disk and two hard drives, a Hitachi Deskstar 7200 RPM and a Western Digital Raptor 10000 RPM, as secondary drives.

Thanks go to Asus, Cooler Master, Intel and Seasonic for supplying most of the components required for setting this test system up.

We kept our Fluke Ti25 thermal camera for the readings, which gives us an image of the different temperatures that we also took using the sensors on the Asus Sabertooth. However we have replaced our sonometers with the Cirrus Optimus CR152A Class 2 models that allow us to measure sound levels down to 21 dBA, which is as low as the room allows us to go. The noise level readings are not therefore comparable with the old ones, which went down to a minimum of 35 dBA.

The sonometer was mounted on a base and placed 50 cm from the side of the casing, raised up 25 cm from the table on which the casing was placed. The solutions measured at between 21 and 22 dBA can be considered as silent. Up to 25 dBA and the cooling is very discreet. Between 25 and 30 dBA can be seen as discreet. Between 30 and 35 dBA we have down as standard and between 35 and 40 dBA is starting to get noisy. Anything higher than this is noisy and becomes excessive for a computer, though of course, as with the other thresholds, this is subjective and depends on several factors such as the regularity of the noise and the environment.

This time we decided to measure noise levels both during standard usage of the casing and with the graphics card isolated, turning the hard drives and CPU/casing fans off. This allowed us to differentiate between the least noisy cards in much more detail.

The load test has also moved on. We have replaced the 3DMark06/07 Pixel Shader test, in which fps values went much too high, with the first scene from 3DMark 11. We opted for this one because it doesnít use tessellation, a rendering technique that could lead to a drop in energy consumption on cards that are saturated at this level. This scene is however 5% less demanding in terms of energy consumption than that in 3DMark06, which worked the texturing units a good deal more. This load test is therefore now very similar to a very demanding game and a good deal different to tests such as Furmark and OCCT in terms of energy consumption. The CPU was loaded with Prime95 on 4 threads, with the lowest priority.

Hereís how we organised the tests:

- Ambient temperature reading to keep it between 25 and 26įC
- 45 minutes at idle
- Temperature sensor readings
- Noise level reading
- Casing is opened rapidly for thermal imaging photo
- Casing closed and two hard drives turned off
- 15 minutes at idle
- All fans are turned off and rapid noise level reading
- Hard drives and fans turned back on
- 45 minutes in load
- Noise level reading
- Temperature sensor readings
- 15 minutes in load
- Casing is opened rapidly for thermal imaging photo
- Casing closed and two hard drives turned off
- 15 minutes in load
- All fans turned off and rapid noise reading

This protocol is still at beta stage and we expect to improve it further, perhaps with the addition of a stage with overclocked cards and another at which the speed of graphics card fans is pushed to 100%. We also plan to modify the speed of the original casing fans between the idle and load phases or even replace them entirely in view of how poor they are, not to mention their lack of monitoring. Even with voltage reduced to 9V, they are still very noisy, with a very pronounced mechanical sound. We measured the level of noise linked to the casing at 34.1 dBA, which is much higher than that of the graphics cards alone at idle.

With the aim of highlighting the differences between the old and new test protocols, we compared the reference GeForce GTX 580 as well as the Zotac AMP≤! Edition under both protocols. Note, the reference card expels almost all the hot air from the casing while the Zotac hardly expels any:

At idle, the temperature comparison between the two cards is identical in both protocols. The noise levels are however a good deal higher in the new casing as it's currently configured. While the Sonata 3 wasnít very noisy at idle (remember the dBA values arenít comparable), the RC-690 II Advanced is really annoying, in spite of the fact that its fans run more slowly. When the graphics card was isolated however, the new sonometer allowed us to note the difference in noise levels between the two cards more precisely.

In load, both graphics cards benefit from the new test system: they donít heat up as much and the fan speed can therefore remain at lower levels. However, thereís more of a gain on the Zotac AMP≤!, the new casing being a good deal more efficient in terms of evacuating the hot air that it gives off.

Page 3
Reference GeForce GTX 580 review

The reference GeForce GTX 580
Numerous partners are selling the reference GeForce GTX 580, at base clocks or as overclocked versions (prices as of 26/10/2011):

Asus ENGTX580 2DI 1536MD5: Ä469.90
Club 3D GTX 580 1.5 GB: Ä464.99 to Ä495.20
EVGA GTX 580 1.5 GB: Ä449.70 to Ä477.90
EVGA GTX 580 Superclocked 1.5 GB: Ä439.90 to Ä509.99
Gainward GTX 580 1.5 GB: Ä444.89 to Ä445.00
Gigabyte GV-N580D5-15I-B: Ä438.96 to Ä489.95
Inno 3D GTX 580 1.5 GB: Ä399.99
PNY GTX 580 1.5 GB: Ä399.99 to Ä451.13
Point of View GTX 580 1.5 GB: Ä452.90
Sparkle GTX 580 1.5 GB: Ä399.99
Twintech GTX 580 1.5 GB: Ä399.99 to Ä475.80
Zotac GTX 580 1.5 GB: Ä517.65
Zotac GTX 580 AMP! 1.5 GB: Ä532.51

For the reference GeForce GTX 580, NVIDIA has used a similar PCB to that used for the GeForce GTX 480 although it no longer has holes in it as, in practice, this resulted in no gain in terms of the channelling of cool air. The new PCB has been optimised to be more reliable under the extreme energy consumption conditions of the GeForce GTX 580, which are identical to those for the GeForce GTX 480. Connectivity is also identical with 2 DVI Dual Link outs, a mini-HDMI, 2 SLI connectors and 8+6 pin power connectors.

The most significant development comes in the GeForce GTX 580ís cooling system, which has been optimised in several ways to make less noise. NVIDIA has used a better quality blower and the external part of the radiator which accumulated a lot of heat and radiated it within the casing has been replaced with a more standard model, which is more effective thanks to a vapour chamber, as with the AMD Radeon HD 5970 for example.

Although their cooling systems are similar, the reference GeForce GTX 580 uses a radiator with a vapour chamber (on the right) that is sturdier than that on the GTX 570 (on the left).

NVIDIA is talking a TDP of 244W during gaming, but in reality in the stress tests, it's more like 300W. In an attempt to contain energy consumption, NVIDIA has included components which allow the drivers (but not the GPU itself) to monitor the energy consumption of the GPU or rather the intensity of the current which flows down each of the 12V power suppy channels (PCI Express bus and connectors) thanks to three shunt circuits placed where the current arrives. These shunt circuits measure the fall in voltage of power sent through a very low resistance channel and use this reading to work out the intensity of the current which passes through it.

Such a reading at the source, even if it isn't perfectly precise, gives you the real energy consumption of the card, which can then be protected in its entirety. It doesnít however enable you to limit the maximum intensity available for the GPU, as this is dependent on the efficiency of the power stage. The power stage yield can therefore provide the GPU with a bigger or lesser margin.

In practice, this NVIDIA system, known as OCP, is more a means to avoid using stress tests than maintaining energy consumption in a well-defined thermal envelope as it is piloted by the driver and only active in a handful of applications (Furmark and OCCT).

Six phases are used to power this more demanding GPU, with two additional phases for the Samsung GDDR5 HC04 memory, certified up to 1250 MHz.

Infrared thermography

[ At idle]  [ In load ]

[ At idle ]  [ In load ]

The reference GeForce GTX 580 is relatively well cooled given the GPU energy consumption levels.

Temperature and noise readings

The readings confirm that the card expels most of the hot air out of the casing. This is done quietly at idle but noisily in load, although within an acceptable level for a high end graphics card.

Our opinion
The reference GeForce GTX 580 can be depended on with a high quality and reliable design which will suit a maximum of casings as it sends most of the hot air out of the casing. Although discreet, the card isnít however totally silent at idle and isnít immune from the high noise levels of high-end GPUs in load.

Page 4
Asus GeForce GTX 580 Matrix Platinum review

Asus GTX 580 Matrix Platinum
Asus is marketing an extreme GeForce GTX 580: the Matrix Platinum. A Republic of Gamers card, it's looking to make space for itself among overclockers with a voluminous cooling system and numerous overclocking features. Of course it has factory overclocking, although this is down on what you might think youíd get from this sort of a monster: 816 MHz for the GPU against 772 MHz for the reference card. An increase of 6% which isnít however matched by the memory, which remains at 1002 MHz (like on the reference card). Note that Asus only offers a 1.5 GB model, which isnít a problem given the fact that the 3 GB versions donít really bring anything extra to the table.

Asus GTX 580 Matrix Platinum: Ä519.00 to Ä539.96

The card

Very big, the Matrix is bigger than the reference GTX 580 in height, width and depth: 3 slots in place of 2, 29.5 cm long in place of 27 cm and an extra 2.5 cm on the top of the card. This last part isnít however because of the cooling system and is only there for aesthetic reasons: it reinforces the massive feel of the card and houses the colour diodes for the Matrix logo. The diodes change in colour as the GPU load increases or decreases.

To cool the GF110 GPU, Asus uses an extreme version of its Direct CU II cooling system. Five large heatpipes run over its large base and are then cooled with two large radiators. A metallic frame ensures the rigidity of the card and supports the two large 10 cm fans. A cover closes the card and channels the flow of air towards the radiators.

There's an aluminium radiator for the sensitive components on the power stage at the level of the PCB. Although the flow of air from one of the two fans passes across it, in load this air will be hot as it will previously have passed through the radiator that evacuates the GPU heat. The Samsung HC04 GDDR5 memory modules, which are identical to those used on the reference card, are open to the air, which isnít a problem as the GDDR5 doesnít heat up too much and is positioned in the flow of air.

Thereís a perforated metallic plate at the back of the card but it isnít in contact with the PCB or other components and isnít part of the cooling system but rather there for aesthetic reasons and to protect the back of the PCB.

Of course it was out of the question to base this extreme model on the reference PCB and Asus has developed an entirely customised PCB with numerous improvements. The power stage benefits from this of course with no fewer than sixteen phases for the GPU in addition to the two phases for the memory. Moreover Asus has used a NEC TOKIN Proadlizer, a very sophisticated capacitor which facilitates the implementation of a compact power supply circuit that doesnít produce too much noise. In other words, this fashionable component has replaced a series of capacitors to simplify what is a high quality design. It's ideally placed just behind the GPU so as to be as close as possible to the source thatís supplied with power. Two 8-pin PCI Express connectors are required to supply this power stage.

Asus says that it has put a double protection system in place with the addition of fuses between the OCP circuits and the power stage. In practice, according to what we could see on the PCB, Asus has simply replaced the shunt circuits with fuses with known resistance close to that used by NVIDIA. This means that this isnít a double protection system but rather a strengthened version of OCP, based as it is on the same components.

These fuses represent relatively high values: 2 x 12A for each 8-pin PCIE connector and 12A for the power supply via the bus. This makes for a total of 720W! This is of course far beyond what is authorised by the PCI Express spec, especially in terms of the motherboard as a PCI Express bus isnít supposed to supply more than 5.5A. Moreover this card is already over the 5.5A limit without overclocking and can be by a lot more if you push the voltage and frequency up. Some motherboards may not respond well to thisÖ In any case Asus has given itself a generous margin so that the GTX 580 Matrix doesnít limit extreme overclocking potential, as long, of course, as you happen on a card that is generously endowed here.

In terms of bonuses, Asus provides for better connectivity than on the reference card with the addition of a DisplayPort connection. A Safe Mode button allows you to return to your original clocks if youíve been too ambitions when modifying the clocks in the bios.

For advanced users and bench table overclocking enthusiasts, Asus has provided voltage read points to meaure the voltage of ins (3.3V and 12V) as well as GPU, memory and PLL voltages. These points are pierced, with a marking on either side of the PCB for ease of use.

There's a large red button for forcing the fan up to full speed. On our test system, moving up from 40% to 100% brought the GPU temperature down from 80įC to 72 įC, at pretty annoying noise levels however. There are two other buttons under this red button and these allow you to increase the GPU voltage progressively (in 0.0125V increments), with a set of diodes marked from 1 to 10 which allow you to identify the level you're at.

Finally, for pro overclockers Asus says, connection spaces you can fill with neutral resistance or conduction pencil have been provided to turn the OCP off, allow modification of the memory and PLL voltage and double the power stage controller clock to improve stability in the case of high overclockjing.

In terms of software, Asus supplies GPU Tweak, based on Rivatuner and GPU-Z. Very thorough, this software allows you to modify clocks and voltages, offers full monitoring and bios updates. For ROG cards, you can also modify memory timings as well as the clocks straight in the bios.

Infrared thermography

[ At idle ]  [ In load ]

[ At idle ]  [ In load ]

The Matrix Platinum is very well cooled both at idle and in load.

Temperature and noise readings

The readings confirm the decent performance of the Matrix Platinum with a significantly better cooled GPU than on the reference card. Thereís a margin to play with as the cooling system fans run at 43%. Noise levels are very low at idle.

Our opinion
The GeForce GTX 580 Matrix Platinum is probably the most advanced GeForce GTX 580, thanks to a very high performance cooling system and a PCB that has been designed to handle high overclocking and satisfy the most advanced users who are the people most likely to be prepared to shell out for it.

Our main reservation about this model is that it doesn't respect the maximum PCI Express bus energy consumption limit and this could be a problem with certain motherboards if you overclock the card. Its triple slot format will also prevent it from being used in mini-PCs as well as most SLI systems.

Page 5
EVGA GTX 580 Classified review

EVGA GTX 580 Classified
EVGA launched its high-end GeForce GTX 580 some time after the competition as part of its Classified range. The large PCB has been entirely revised to give as big a margin as possible for overclocking, in addition to a significant factory overclocking: 855 MHz for the GPU and 1053 MHz for the GDDR5 memory compared to 772 and 1002 MHz on the reference card. These are gains of 11 and 5% respectively. The GTX 580 Classified exists in 1.5 and 3 GB versions and you can also choose between air cooling and watercooling:

EVGA GTX 580 Classified 1.5 GB: 490Ä
EVGA GTX 580 Classified 3 GB: 560Ä
EVGA GTX 580 Classified Hydro Copper 1.5 GB: 690Ä
EVGA GTX 580 Classified Hydro Copper 3 GB: 730Ä

EVGA sent us the standard 3 GB version.

The card

The EVGA GTX 580 Classified wonít go unnoticed Ė itís a big mutha! Its PCB is even bigger than that on the MSI Lightning and measures 28 cm in length and 12.5 cm in height. This is a format that allows EVGA to use a radial type blower (8 cm in diameter) from which a casing extends across the whole card to direct the flow of air out of the system. This very basic plastic casing, coupled with the size of the card mean the card is somewhat lacking, aesthetically speaking.

As with the reference card, the core of the cooling system consists of a vapour chamber fixed to an aluminium radiator. A small extension has been added to increase cooling capacity. There's a flattened heatpipe running over this. Note however that the aluminium fins that make up the extension arenít as densely packed in as in the main part because the extension widens out towards the top end. We wonder how effective the extension actually is, especially in view of how the hot air (roughly 20% of the total) is expelled inside the casing, in the direction of the SLI connectors.

A metal plate covers the whole of the PCB and is in contact with the Hynix T2C memory modules as well as with the sensitive power stage components. It also ensures that the huge PCB is kept rigid and serves as a support for the blower.

Although the GTX 580 Classified is huge, this is partly because EVGA wanted to make space for a sturdier power stage: it has announced that it can handle energy consumption of up to 1000W, in comparison to just 400W for the reference card. To recap, the Asus GTX 580 Matrix uses 720W fuses. EVGA would therefore seem to be going for a belt and braces approach here.

There are fourteen phases for the GPU, accompanied by three phases for the memory. EVGA says that it has looked in depth at the quality of the signal sent to the GPU, so as to facilitate overclocking, using NEC TOKIN Proadizers and high-end inductors.

This power stage requires no less than three PCI Express connectors: 2 x 8 pins + 1 x 6 pins. To facilitate installation with a maximum of power supplies, EVGA supplies an 8-pin to 2 x 6 pin adaptor and a 6-pin to 2x molex adaptor. You will however need a correctly sized power supply, with at least two 6-pin connectors and one 8-pin connector.

Note that with its Classified model, EVGA has made an effort to reduce energy consumption via the PCI Express bus 12V supply. In spite of significant overclocking, then, EVGA has managed to make to do with 4.3A where all the other GTX 580s, with the exception of the MSI Lightning, exceed the 5.5A security limit.

EVGA has implemented OCP as defined by NVIDIA but extended it so as to take the fourth 12V power source into account.

Finally EVGA supplies a strip of electrical (but not thermal of course) insulation to be placed on the sensitive power stage components should you be using an alternative cooling system, so as to avoid any short-circuits. Moreover, thereís an extreme cooling kit (LN2) that can be purchased separately for cooling the power stage.

The GTX 580 Classified is compatible with the EVBot system (connector on left) which allows you to modify clocks and voltages without going via the system. It doesnít have an HDMI connector however Ė a DVI adaptor is supplied for this purpose.

EVGA has provided GPU voltage read points, memory and PLL for advanced users as well as diodes (on the left on top photo) to show that these voltages are functioning properly. Thereís a switch for moving over to an optimised bios to avoid cold bug when using the LN2. This bios also turns OCP off and allows you to increase the GPU voltage up to 1.8V.

EVGA supplies Precision, based on Rivatuner to facilitate overclocking, but Precision doesnít support modification of voltages. Youíll need the E-LEET software for that. Unfortunately there's no official support for the GTX 580 Classified here yet, though there is a beta version available. EVGA did assure us that this was a priority and that the situation would be resolved rapidly but seems to be dragging its feet somewhat.

Infrared thermography

[ At idle ]  [ In load ]

[ At idle ]  [ In load ]

In spite of significant factory overclocking, the EVGA GTX 580 Classified power stage doesnít overheat and the GPU temperature is a good deal lower than that on the reference card.

Temperature and noise readings

The readings confirm that the GTX 580 Classified is well cooled, with a lot of the hot air being expelled out of the casing. They do however highlight the major drawback with this model: noise. The GTX 580 Classified was the only GTX 580 we tested that was noisier in idle than our test system (34.1 dBA). In load it breaks all noise records and is even louder than the MSI GTX 580 Lightning.

Unlike the Lightning however, the GPU temperature of 87įC at 66% of maximum blower rotation speed (3450 RPM) shows that this isnít a calibration issue where GPU temperature in load has been favoured over noise criteria but unfortunately simply a feature of the cooler used here. At idle however you will be able to bring noise levels down and let the GPU temperature increase slightly above normal.

Our opinion
Although EVGA has done some great work on the power stage (more robust and with higher performance so it can handle advanced overclocking), unfortunately the cooler designed for this model isn't as good. Unless youíre deaf or very interested in the GTX 580 Classifiedís overclocking potential, there are better solutions available.

For watercooling enthusiasts however, the Hydro Copper option for the GTX 580 Classified can be purchased for "just" Ä200 extra and won't suffer from the same issue. Is this the price you have to pay for silence?

Page 6
Gainward GTX 580 Phantom review

Gainward GTX 580 Phantom
Two customised GeForce GTX 580s are available from Gainward, with the Phantom their top-end model. Both cards are based on the same PCB, which is also used on some of the Gainward GTX 570s, and offer identical, minimal factory overclocking: 783 MHz for the GPU instead of 772 MHz, which is a 1% increase. They only differ by their cooling systems.

The cooling system used here isn't particularly efficient but has been designed to look good at reduced cost. Although this model isn't therefore in the same category as the other cards in this report, we wanted to take a look at what it was capable of in comparison to the reference model, especially as we have a new test system.

Gainward is marketing both 1.5 GB and 3 GB versions, with the only difference being the memory modules used.

We tested the 1.5 GB version.

The card

Gainward has come up with an atypical design for its GeForce GTX Phantoms, with the fans placed under a large radiator. While this may look good visually, this design doesn't have as robust a finish as the reference cardís cooler. Gainward has opted for a rather large cooler with a huge radiator, which extends down the entire length of the card. Six nickel plated copper heatpipes run across it and are fixed to a wide aluminium base. The manufacture is pretty standard here, with no direct contact.

An aluminium plate covers the whole of the PCB, is in contact with the sensitive components and includes a radiator part for the power stage. All this takes up three slots though the Phantom (27 cm) isnít any longer than the reference card.

There are two DVI outs, an HDMI out and a DisplayPort out on the customised PCB. Only a DVI to VGA adaptor is supplied in the bundle, in addition to the small manual, a CD for the drivers and a double molex to 8-pin PCI Express power supply cable convertor.

The six phases for the GPU on the reference model have been kept for this design, in addition to the two phases for the Samsung HC04 GDDR5 memory. The components are totally different however and seem to have been chosen to reduce costs. Gainward has retained the original version of OCP however.

Infrared thermography

[ At idle ]  [ In load ]

[ At idle ]  [ In load ]

Although no problems came up at idle, in load the Gainward GTX 580 Phantom proves rather inefficient. The power stage heats up considerably and the card has an impact on the temperature of the various other components in the casing.

Temperature and noise readings

The readings confirm this and show that noise levels in load are also very high.

Our opinion
The Gainward Phantom is on sale at the same price as the reference GeForce GTX 580 and didnít convince us due to high temperatures and noise. Though it might look very good, itís a very big card and manufacturing quality is down. We then prefer the reference card, which is more efficient.

Page 7
Gigabyte GTX 580 SOC review

Gigabyte GTX 580 SOC
Gigabyteís high-end GeForce GTX 580 is part of the Super OverClock range and comes with a sturdier cooler and PCB than the other Gigabyte models. Gigabyte says that it has hand picked the GPUs for this SOC model so as to be able to offer both significant factory overclocking and as high an overclocking potential as possible: 855 MHz for the GPU against 772 MHz on the reference card, which is a gain of 11%. The memory clock is also up to 1025 MHz, from 1002, a difference of just 2%.

Like Asus, Gigabyte hasnít bothered with a 3 GB variant.

The card

The Gigabyte GTX 580 SOC is the smallest card in this report, except for the reference model. Although, at 27.5 cm, itís half a centimetre longer than the reference card, and a heatpipe and cover stick out 1 cm on top, the Gigabyte model takes up just two slots.

While the cooler is relatively small, a lot of effort has gone into designing it. Thereís a very big vapour chamber fixed to a first radiator on the base. Two big heatpipes lead up from there to a second smaller radiator. A metal frame is used to give the whole structure some rigidity and support the three 75 mm fans. Thereís also a metal cover over it all.

Contrary to what you might think at first if you look at the photos of the card in profile, the card wasn't delivered bent but with the radiators and casing deliberately shaped and the orientation of the fans carefully chosen to give maximum effectiveness. The metal plate fixed to the vapour chamber is in contact with the Hynix T2C GDDR5 chips and a small radiator is fixed to the sensitive power stage components, which are positioned in the flow of air.

The overall look is not one of the strong points of this SOC model but the metal casing gives a solid feel to the card. Perhaps Gigabyte should use another colour for its high-end PCBs!

The PCB has been entirely revised, with twelve phases for the GPU power supply and two phases for the memory. This rather complex power stage occupies both sides of the PCB. Thus, only a third of the MOSFETs are on the front, attached to a small radiator, with the other two tiers at the back of the PCB, with no radiator. Gigabyte says that it has only used superior quality components and there are no fewer than five NEC TOKIN Proadlizers at the back, supposedly to simplify the designÖ

The GTX 580 SOCís power supply system goes very much hand in hand with the OC Guru software that must also be installed. Unless you do this, the twelve phases for the GPU will be constantly in use, even at idle, and this increases energy consumption. With OC Guru installed, just one phase is used at idle, with more kicking in gradually as the load increases. A diode system shows you which GPU phases are being used.

Although Gigabyte has gone for the reference OCP system, the design draws a little too much on the PCI Express bus. With a 5.77A reading, this is the only model weíve tested that goes beyond the 5.5A security limit in 3DMark. In Furmark this value jumps to 6.88A and this will be an issue on many motherboards, especially if you overclock it further.

As a little bonus, Gigabyte has provided a double bios with an XTREME button to switch over to the second bios for extreme overclockers: the clocks are the same but itís optimised to prevent cold bug with the LN2. Although it appears to be very poorly positioned, this isnít a problem when it's used for what it was designed to be used for. It could however have been better placed so that it could also serve as a secondary bios for a standard overclocking mode.

We donít however know what was going through the minds of the Gigabyte engineers when they positioned the voltage read points just beside this button. In addition to the fact that there are no markings on the PCB, these points are also rather inaccessible. The Gigabyte engineers will probably tell us that it would have been ridiculous to detail the function of each (from left to right: GPU voltage, GPUIO, memory, PCIE and earth) given that it would have been impossible to read them here. A little more reflection might have been of valueÖ

Infrared thermography

[ At idle ]  [ In load ]

[ At idle ]  [ In load ]

The Gigabyte GTX 580 SOC tends to heat up a lot in load and this is because most of the MOSFETs are at the back of the PCB, with no radiator or flow of air, unlike at the front.

Temperature and noise readings

Above and beyond the fact that the power stage heats up, the readings show that there are no particular temperature issues with this card. It is moreover quieter than the reference model at idle and not as noisy in load. Note however that to get these results, the fans run at almost maximum speed.

Our opinion
Relatively compact and with significant factory overclocking, the Gigabyte GeForce GTX 580 SOC seemed well placed in comparison to the other cards in the report. Unfortunately it also has a couple of design errors: the first concerns the positioning of the advanced features which are difficult to access and the second, which is much more important for most users, concerns the power stage. The MOSFETs at the back of the PCB heat up significantly as thereís no cooling here. This must call the cardís lifespan into question if youíre intending to do any overclocking. This could well lead to more generalised overheating as the cooling system doesnít have much room for manoeuvre.

In addition, we're not all that enthusiastic about the fact that the card exceeds the specs in terms of the energy drawn from the PCI Express bus, nor that you must install software to make the power stage more efficient, particularly in idle. This SOC model was, then, a real disappointment that will only satisfy factory overclocking enthusiasts who donít want to take things any further.

Page 8
MSI GTX 580 Lightning review

MSI GTX 580 Lightning
MSI has also brought out a high end GeForce GTX 580, the Lightning, which is based on an imposing PCB that has been fully revised with a view to major overclocking. The GPU is clocked at 832 MHz by default, which is a gain of 8% on the reference card (772 MHz). The memory is clocked at 1050 MHz, up from 1002, which is a 5% increase which, while still modest, is the biggest memory overclocking of the cards tested.

MSI is also marketing an Xtreme Edition variant, which apart from the colour, differs with respect to the video memory (3 GB) and bios support for anti-dust technology which runs the fans in the other direction for 30 seconds on start-up of your PC. The PCB, cooler and clocks are all identical to the Lightning. We tested the standard model.

MSI GTX 580 Lightning 1.5 GB: Ä484 to Ä539.99
MSI GTX 580 Lightning Xtreme Edition 3 GB: Ä579 to Ä619.99

The card

The first thing you notice with the GTX 580 Lightning is how big it is. Its PCB is 1 cm longer than the one on the reference card and, more unusually, 1.5 cm higher to give us 28 cm x 11.5 cm. Add to that the cooling system cover, which sticks out 1 cm from the back of the card and a heatpipe which extends 1 cm above the card and you have a veritable monster, even if it only takes up two slots. However the SLI connectors are at their usual height so as to guarantee compatibility with other GTX 580 models.

The cooling system rests on an aluminium base from which five nickel plated copper heatpipes lead (two of them are very big). They run up to an enormous radiator, which is cooled by two 9 cm fans. The design of the blades for these fans has been studied to maximise yield. An aluminium casing covers this system.

A metal plate covers a large part of the PCB and is in contact with the sensitive power stage components as well as the Samsung HC04 GDDR5 memory modules. This plate also ensures the rigidity of the whole card, which is important given its size.

If MSI has gone for such a big PCB, it isnít only so it can add a DisplayPort connector but rather to give some muscle to the power stage, which is made up exclusively of better quality components. The power stage is supplied by two 8-pin connectors and has twelve phases for the GPU and three phases for the memory. A set of diodes at the back of the PCB allows you to observe the number of phases being used in real time.

At the back of the PCB, MSI has also succumbed to the NEC TOKIN Proadlizer fashion, with four of these components, which reduces the number of capacitors required to produce as clean a signal as possible.

A real advantage of this MSI design is that it only partially uses the 12V source supplied by the PCI Express bus, which prevents damage to your motherboard, even with the most extreme overclocking. During a very high load, in Furmark, the GTX 580 Lightning only consumed 0.55A via the PCI Express 12V source, compared to over 5A for all the other GTX 580s, with most of them even exceeding the 5.5A limit fixed by the security standard. MSI has used the reference card OCP system.

MSI has also implemented voltage read points for the GPU, memory and PLL. These are mounted on small connectors on which run the small cables (supplied) that facilitate the readings.

Apart from these read points, a series of small switches at the back of the PCB offers various advanced features: turning OCP off, increasing PLL voltage, adjusting the PWM clock between 260 and 310 MHz to stabilise certain extreme overclocking tasks, increasing GPU voltage and, lastly, increasing memory voltage. On the right thereís also a switch to move over to a second bios which is pre-loaded with a bios to avoid cold bug during LN2 overclocking and which can also serve as a backup bios or a bios with higher clocks. There are also two other switches in the middle of the PCB which arenít documented but which have been designed to facilitate overclocking with the LN2.

MSI supplies the well-known Afterburner software, which makes it easier to do your overclocking or modify the GPU, memory and PLL voltages.

Infrared thermography

[ At idle ]  [ In load ]

[ At idle ]  [ In load ]

The MSI Lightning is the best-cooled card overall in this report. In load however, it sends more hot air towards the bays and hard drives than most other cards.

Temperature and noise readings

While there are no temperature problems, the readings do highlight a double record in terms of noise levels. First of all the MSI GeForce GTX 580 Lightning is the only GTX 580 that is truly silent at idle. Next, it's the noisiest card in load. With a rotation speed of 100%, it seems clear that MSI has calibrated its cooling system to keep the GPU temperature low at any cost.

Our opinion
While the atypical shape of the MSI GTX 580 Lightning will pose problems for some small casings - it does, however, allow you to set up SLI and triple SLI systems - the numerous features on offer will be certain to delight extreme overclocking enthusiasts.

We advise Ďstandardí users to recalibrate fan speed. It would probably have been worth MSI offering a bios that was better calibrated for noise containment with the current bios being used as an optional one.

Page 9
Zotac GTX 580 AMP≤! Edition review

Zotac GTX 580 AMP≤! Edition
Zotac has based its GeForce GTX 580 AMP≤! Edition on the AMP! Edition, using the same PCB and factory overclocking, but the Zalman VF3000F cooler. The GPU is clocked at 815 MHz, which represents a gain of 6% in comparison to the reference model (772 MHz). The memory is up to 1025 MHz from 1002 MHz, namely a modest gain of 2%. Still on the subject of this AMP≤! Editionís memory, the card is only available in a 3 GB version.

Zotac GTX 580 AMP≤! Edition. Ä500

The card

The VF3000F cooler means the Zotac occupies 3 slots. This cooler has a copper base from which five heatpipes run over numerous fins. These fins ensure the rigidity of the card as a whole, though it isnít as sturdy as it could be. Two 85 mm fans are used for the cooling and they're covered with a metal casing.

A metal plate covers a large part of the PCB and is in contact with some of the sensitive power stage components as well as the Hynix T2C GDDR5 memory modules. Unfortunately this plate stops just before an important section of power stage components (the MOSFET drivers), in contrast to the reference modelís cooling system. Small pads placed between this plate and the cooler prevent the cooler from getting squashed when moving the card around.

While Zotac makes its own PCB, this is simply a replica of the reference model PCB, which does inspire a certain confidence with its black matte coating and rounded edges. Zotac has however replaced the aluminium polymer type capacitors with better quality tantalum capacitors (in yellow).

Infrared thermography

[ At idle ]  [ In load ]

[ At idle ]  [ In load ]

In load, the power stage heats up a lot, which is at least in part due to the lack of passive cooling of some sensitive power stage components.

Temperature and noise readings

The Zotac GeForce GTX AMP≤! Edition proved to be very discreet at idle and the quietest of the GeForce GTX 580s in load, at the same time keeping fan speed in reserve.

Our opinion
The Zotac GeForce GTX 580 AMP≤! Edition will delight users looking for maximum performance with noise levels that are as low as possible. Among its faults are the fact that the triple slot format will pose a problem in SLI and that the poorly designed Zalman radiator has been used for the power stage. We advise you to spare it any additional overclocking.

Page 10
Summary of results

Summary of results
Here weíve brought together all the temperature readings taken during the tests. We have highlighted the particularly interesting results:

At idle, all the cards do well and the GPUs are well cooled, with however a small advantage going to the Asus Matrix Platinum.

In load, we noted a few differences, the most important of which are linked to the fact that only the reference GTX 580 and the EVGA Classified expel most of the hot air out of the casing, mainly benefitting the CPU, with the hard drives benefitting from the entry of cool air into the new test casing. One of them does however get close to 50 įC, and this, when it's in simple rotation, with the Gainward Phantom leading to the biggest increase in the temperature of all the components in our system.

Page 11
Temperature and noise levels

GPU temperatures
We have shown the various GPU temperatures at idle and in load.

We have displayed the readings at idle for the card on its own and the readings in load for the whole system, which doesnít change anything for the noisiest cards, which totally cover the noise levels of the casing (34.1 dBA).

While almost all the manufacturers do a little bit better, even a lot better (MSI and Zotac) than the reference model at idle, the EVGA GTX 580 Classified is the exception to the rule.

In load, there is more of a difference between the cards and the Zotac card clearly does best. Asus and Gigabyte do better than the reference card, while the Gainward is noisier, as is the MSI, which suffers here from a poorly calibrated fan, and the EVGA cooler, which probably needs to be sent back to the drawing board.

Page 12
Energy consumption

Energy consumption
We measured the energy consumption of the different cards, keeping in mind that thereís often a variation between identical cards of the same model, due, among other things, to leaks in current and different GPU voltages. Moreover, some run at higher clocks as they are factory overclocked.

We measured energy consumption directly on the graphics cards themselves.

At idle, we observed an average energy consumption of 32W for all the GeForce GTX 580s. Note however that this climbed to 40W for the Gigabyte SOC model when the OC Guru software wasnít installed, as the card's power stage doesn't go into idle when it's not running. Note the clear advantage for the Asus Matrix Platinum which probably comes from the fact that the power stage is particularly optimised to this effect. This is moreover also why the cardís GPU temperature is the lowest.

In load, the difference between the cards increases and the situation becomes more complicated. It should be noted that thereís a relationship between GPU temperature and energy consumption as higher temperatures increase current leakage. A well-cooled GPU will consume a little bit less while a badly cooled GPU can find itself in a vicious circle. The same goes for the power stage, which can see its yield fall when it heats up. These reasons probably explain in part why the energy consumption of the Classified, Phantom and SOC cards explodes when you switch from 3DMark to Furmark, which is moreover tested in a version which doesnít launch OCP.

On the other hand, the MSI Lightning is the most economical GeForce GTX 580 in load.

Page 13
Infrared thermography: Graphics cards

Infrared thermography: graphic cards

[ Reference GeForce GTX 580 ]
[ Asus GTX 580 Matrix Platinum ]
[ EVGA GTX 580 Classified ]
[ Gainward GTX 580 Phantom ]
[ Gigabyte GTX 580 SOC ]
[ MSI GTX 580 Lightning ]
[ Zotac GTX 580 AMP≤! Edition ]

At idle, all the cards are well cooled with a small advantage for the Asus Matrix Platinum, which benefits from reduced energy consumption in this state.

[ Reference GeForce GTX 580 ]
[ Asus GTX 580 Matrix Platinum ]
[ EVGA GTX 580 Classified ]
[ Gainward GTX 580 Phantom ]
[ Gigabyte GTX 580 SOC ]
[ MSI GTX 580 Lightning ]
[ Zotac GTX 580 AMP≤! Edition ]

In load, you can see how well the Asus, EVGA and MSI cards do and on the other hand, how the power stage heats up on the Gainward, Gigabyte and Zotac.

Page 14
Infrared thermography systems

Infrared thermography: systems

[ Reference GeForce GTX 580 ]
[ Asus GTX 580 Matrix Platinum ]
[ EVGA GTX 580 Classified ]
[ Gainward GTX 580 Phantom ]
[ Gigabyte GTX 580 SOC ]
[ MSI GTX 580 Lightning ]
[ Zotac GTX 580 AMP≤! Edition ]

None of the cards poses any temperature problem at idle.

[ Reference GeForce GTX 580 ]
[ Asus GTX 580 Matrix Platinum ]
[ EVGA GTX 580 Classified ]
[ Gainward GTX 580 Phantom ]
[ Gigabyte GTX 580 SOC ]
[ MSI GTX 580 Lightning ]
[ Zotac GTX 580 AMP≤! Edition ]

In load you can see that the Gainward Phantom is the GTX 580 which has most of an impact on the system as a whole.

Page 15
Overclocking and performance

Obviously each individual card of any model can react differently when overclocked but we tried to push all the cards in our possession so as to be able to observe any generalities. Note on this subject that the reference, Asus, EVGA, MSI and Zotac cards were supplied to us by their respective manufacturers, while the Gainward and Gigabyte cards came from in-store stocks.

In an open test system, we used Furmark to load the cards to a maximum to ensure stability by increasing the clock in 25 MHz increments, without changing the original GPU voltage. Next we looped Crysis Warhead to validate the overclocking as, generally speaking, this exercise is more demanding than Furmark.

Note that GDDR5 memory has different mechanisms for the detection of errors that mean that although the memory remains stable when overclocked significantly, it does lose a lot of time to repair the associated errors, either by returning the corrupted data or recalibrating frequencies. Given that the Fermi GeForces struggle with a high memory clock, we didnít overclock it.

Here are the clocks that we obtained:

Note first of all that the GPU voltage isnít fixed and can vary from one sample to another. This gives NVIDIA more flexibility when it comes to validating its GPUs at a given spec.

The GeForce GTX 580s generally clock up to 850 or 875 MHz without too much difficulty. Only the EVGA Classified and Gigabyte SOC went up to 925 and 900 MHz respectively without any difficulty or modification of the GPU voltage. Given their high base clocks, these only represent 8 and 5% clock increases respectively. Although, also overclocked to a relatively high level the MSI Lightning wouldnít go beyond 850 MHz and remain stable, which translates to a small gain of 2%.

The cards with reference clocks, or clocks close to those of the reference card are often those with the highest overclocking margin, the factory overclocking eating into it even when GPUs are handpicked.

For information, we observed the performance of all the cards at their original clocks and when overclocked in Crysis 2, at 1920x1080 DirectX 11 Extreme:

In comparison to the clocks of the reference card, factory overclocking allows for a gain of between 1 and 9%, with the EVGA and Gigabyte models out in the lead here of course. Once we had overclocked all the cards, the gains on the reference card were more closely grouped at a gain of between 7 and 12%.

Page 16

At the end of this report, we can give you some clear purchasing advice, but above all, remember the results of our recent piece on the impact of the 3 GB GeForce GTX 580 models, or lack of it rather and the fact that they make no real difference for most usage situations. If you were paying the same for your 3 GB model then we'd say, 'Why not?í, but paying extra or opting for a lower quality card so as to be able to afford a 3GB GeForce GTX 580 model seems pointless to us.

If youíre looking for a "budget" GeForce GTX 580, we wouldnít advise you to go for the Gainward Phantom, but rather the reference model, which is of better quality, more efficient when it comes to noise and temperature levels and can give better value for money, depending on the manufacturer.

The priority for this report however was to highlight the differences between the high-end GeForce GTX 580s. Asus, EVGA and MSI did best with their Matrix Platinum, Classified and Lightning models, all of which have big cooling systems, very sturdy power stages and numerous options to target the most advanced users and even the pro overclocker niche. These models are of course more expensive, but give you a maximum margin for overclocking as long as the GPU itself cooperates (not all samples have the same potential).

These three GeForce GTX 580s are however very big and the Asus Matrix Platinum, with its three slots, will make setting up an SLI system impossible in most situations while the original bios of the MSI Lightning prioritises GPU temperature in its calibration of fan speed, at great cost to your ears.

The EVGA Classified suffers from having a substandard cooler. The standard version of the card is far too noisy both at idle and in load, leaving just the expensive watercooling model: the Hydro Copper.

We thought weíd found the ideal compromise between size and overclocking potential in the Gigabyte GTX 580 SOC but unfortunately a badly conceived design means the power stage heats up too much. Thanks to a significant factory overclocking, this model could suit those looking for a high performance GeForce GTX 580 without having to overclock it.

Finally, we advise those of you looking for a quiet card to take a look at the Zotac GeForce GTX 580 AMP≤! Edition, making sure you take note of the little "≤" which makes all the difference... if you manage to find one that is, as it looks as if Zotac has now stopped distributing this model. Its Zalman VF3000F cooler is the GeForce GTX 580 cooling system which best contains the noise of its high consumption GPU. Once again, the cooling for this modelís power stage suffers from a poorly conceived design, which makes the card unsuitable for overclocking.

To conclude, weíd like to say something about a problem that is unfortunately all too common when it comes to the energy consumption of high-end graphics cards: going over limits defined by the security spec for power supply via the PCI Express bus. This can damage some motherboards either instantly or over time, especially if they arenít high-end versions and are put through overclocking sessions or stress tests. While exceeding the official limit via the power supply connectors isnít a huge problem as powerful power supplies are more than capable of taking the load on, we think that it's important for manufacturers to make sure that their products won't damage others and that the spec of the power supply via the bus is respected. Only EVGA and MSI have made the effort to make sure of this with the GeForce GTX 580 Classified and the Lightning.

Copyright © 1997-2015 BeHardware. All rights reserved.