NVIDIA nForce 790i Ultra SLI - BeHardware
Written by Marc Prieur
Published on March 18, 2008
After a lukewarm nForce 780i which actually was more like the nForce 690, three months later Nvidia is back in the foreground with its nForce 790i. Amongst other things, innovations include native PCI Express 2.0 support as well as the integration of a DDR3 memory controller – a first for NVIDIA.
A new chipWhile the nForce 780i was in some ways just an nForce 680i with an additional chip, the nForce 790i integrates a brand new SPP. This SPP which is NVIDIA’s official name for the northbridge now has native PCI-Express 2.0 support.
On the nForce 780i SLI, you may recall that PCI-E 2.0 was managed by an additional chip, the nForce 200, which was relayed to the SPP via an ’’overclocked’’ PCI-E 1.0. In practice, performances were not equivalent to those offered by the X38’s PCI-E bus. Now there is native PCI-E 2.0 support which, we remind you, is the new specification enabling to attain data speeds of up to 500 MB /s on each lane versus 250 MB /s for the first version. In addition, there are two PCI Express 2.0 x16 ports.
NVIDIA also underlines its advances in PCI-Express support on the SPP that are useful in multi-GPU configurations. First of all, there is GPU-to-GPU Direct Link, which allows cards to directly communicate between each other without having to go through the memory controller. Second, there is Broadcast which means the CPU sends information only once which is then repeated to all GPUs. We will just have to see if these improvements have a notable impact in the case of Quad SLI, for example, or SLI with the recently released 9800 GTX.
PCI-E 2.0 is not the only innovation of the nForce 790i and actually it is the first non-Intel chipset to have DDR3 support. Introduced almost a year ago, this new type of memory differs from DDR2 by voltage that was reduced to 1.5V (versus 1.8V in DDR2 and 2.5V in DDR) as well as with its prefetch. The latter, which went from 2n to 4n bits in the change from DDR to DDR-2, now increases to 8n bits. The internal organization of memory cells has therefore been modified to obtain a doubled speed without increasing their frequency to the detriment of that of buffer inputs / outputs and the external memory bus.
Besides the gain in terms of energy consumption, this type of memory thus adds a noticeable gain in terms of bandwidth. It is officially available in versions ranging from DDR3-800 to DDR3-1600 or the double of DDR2 which varies from DDR2-400 to DDR2-800 and manufacturers now go up to DDR3-2133. The only problem is the very high price. DDR3 on the nForce 790i is matched with EPP 2.0 (2.0 only being related to DDR3 support). You may recall, EPP is a type of evolved SPD which allows memory manufacturers to transmit more information via their intermediaries to the motherboard so that there can be better use of memory depending on the type of EPP profiles they integrate. For example, this can be a profile with the best timings or another with the best frequency.
NVIDIA launches two versions of the nForce 790i, the SLI and Ultra SLI. While the first is officially ‘’limited’’ to DDR3-1333, the second goes further with NVIDIA announcing DDR3-2000 support. This is made possible without overclocking the FSB due to the fact that NVIDIA chipsets offer a flexibility which enables setting the memory and FSB asynchronously (without limits to FSB:DRAM coefficient ratios). The Ultra version was also announced as being more ‘’overclockable’’ although no precise figures were given.
The southbridge remains the same and thus has two Gigabit Ethernet ports, which are possible to combine via DualNet technology as well as six SATA ports configurable in RAID 0, 1, 0+1 or 5. It also manages the last PCI Express x16 port which is therefore a 1.0 type.
The test, PCI-E 2.0
While this is the motherboard that was used in performance tests, we were also able to obtain an ASUSTeK Striker II Extreme at the last minute which we used for evaluating overclocking potential. Based on the nForce 790i Ultra SLI, or the most ‘’overclockable’’ version of the chipset, it is equipped with ‘’Fusion’’ type cooling, in other words, running on passive as well as watercooling.
The testFor this test, we were able to get our hands on two motherboards. The first was a reference nForce 790i SLI model which is sold by EVGA amongst others. Compared to the nForce 780i SLI motherboard, we can see the addition of a chip that adds management of two supplementary SATA, the first being located next to the first PCI-E port and the second being an eSATA port. Cooling was reviewed and is still heatpipe based. It goes around the Socket and cools the chipset as well as the CPU power blocks and DRAM. The optional fan is the same as that on the 780i and is therefore very noisy although we were assured that it wasn’t the definitive version. Finally, in addition to functions managed by the chipset there is a FireWire chip.
Specifically destined for those who like to tinker, this motherboard offers amongst other things On/Off buttons and an assortment of LEDs which change from green to red depending on the CPU, memory, and chipset voltage and even the FSB frequency : simply frills ? To each his own. A FireWire chip and two eSATA are also present while the audio component is relegated to a PCI-Express card in order to reduce interference that can affect analog outputs.
For the X38, we used an ASUSTeK P5E3 motherboard.
PCI-Express 2.0In practice, what does PCI Express 2 support add? This is what we wanted to determine with the help of Crysis, the latest game from Crytek, and which has the most advanced graphic engine at this time. In fact, in standard resolution, 1680*1050, high quality mode causes problems for 256 MB cards and they are required to use central memory to compensate for this.
It’s only in this type of situation that PCI-E 2.0’s supplementary memory bandwidth makes the difference which is something we verified with a GeForce 8800 GT 256 MB:
Tests were run on the first PCI Express port and the third (the second offers the same speeds as the first). With nForces and the X38, the third port is relayed to the southbridge and in PCI-E 1.0 but in x16 and x4, respectively. Here, we should focus on the gain between the nForce 780 and 790i on the first port. Formerly, we were notably behind the X48 but now we are ahead. This is a sign that shaky support of PCI-E 2.0 with the nForce 780 is a thing of the past. In this type of configuration, the last port on the X38 is more decorative than anything else.
With a card equipped with 512 MB of memory such as the Radeon HD 3870, the impact of PCI Express 2.0 is almost nil. Even the scores of the X38’s 4x PCI-E are satisfactory performances are noticeably lower contrary to those of the nForce.
Memory, RAID 5
Memory performances We were first interested in evaluating the memory controller by using ScienceMark 2 to obtain speeds and latency. The framerate was also measured in Crysis and file compression times were obtained with WinRAR 3.7. These two ‘’practical’’ applications were chosen because they are noticeably affected by the speed of the memory subsystem, something that isn’t always the case. Tests were carried out in FSB1600 (a speed at which the nForce 780i started to show some weakness) and in DDR2-800 4-4-4-12, DDR2-1066 5-5-5-15, DDR3-1066 7-7-7-21 and DDR3-1333 7-7-7-21.
To begin with, here are the results of tests with ScienceMark :
The nForce 790i displayed the lowest latency at 5 to 6ns less than the X38. In terms of bandwidth, the latter has a slight advantage in DDR3-1066 but results are similar in DDR3-1333. Either way, it was the practical performances that interested us the most:
While the 780i showed a certain weakness here, this isn’t the case for the 790i. Actually, we have first rate performances and 1 to 2% ahead of the X38. However, note that there is some resistance with DDR2. The X38 and its DDR2-1066 does as well as DDR3-1333, which however is much more expensive.
In Crysis the X38 is ahead of the nForce 790i, although the lead doesn’t exceed 0.5%. Here again, DDR2 is more than competitive.
As you can see, while NVIDIA’s nForce 780i was a chipset with lower performances than those of the X38, the nForce 790i puts this manufacturer on top again.
RAID 5While the MCP is the same, we still wanted to verify RAID 5 improvements added with drivers with the nForce 790i SLI.
IOMeter is used to simulate the load in a multi-user environment, in this case by using a file server type load comprised of 80% reading and 20% writing, all 100% non sequential. In this test we measured performances, expressed in inputs/outputs per second (IO/s) with 1, 2, 4, 8, 16, 32, 64 and 128 simultaneous commands. The two chipsets were tested with Raptor 150 GB drives.
Just like with the nForce 780i, the nForce 790i plateaus at more than 4 simultaneous commands. Given that this behavior is only present in RAID, it shows a limit related to this mode. Moreover, this is not a factor with the X38. Of course, the nForce 790 is far from being destined to a server environment ; however, we might wonder where this limitation comes from.
We now move on to file copying. We measure reading and writing speeds, as well as the copying on the same partition of the different configurations. This is done on a series of files composed of 2 large files totaling 4.4 GB, 2620 files with a total of 2 GB, and finally, 16,046 smaller ones equaling 733 MB. The source or target in reading or writing on the drive is a RAID 0 of two Raptor 74 GB drives.
Like with the nForce 780i and X38 this type of test isn’t to the advantage of RAID 5 configurations. This type of writing with small files isn’t ideal for the software RAID 5 configuration offered by chipsets.
We then measured motherboard consumption with an ammeter that gave readings of the different lines on the ATX block. This allowed us to deduct the processor consumption which arrives via the ATX12V as well as that of other peripherals directly connected to the power supply. Tests were carried out with a Q6600, 8800 GTX and 2x1 GB of DDR3-1066.
Compared to a X38, the reference nForce 790i SLI motherboard had 69.8w in stand-by and 74.5w in load. This is power consumption that is 11.4 and 8.4w higher, respectively, than a X38. It’s a notable difference but still remains reasonable.
Overclocking – CPU Moving on to overclocking the chipset with a Core 2 QX9770, we only give you the frequencies that were validated by four 15 minute sessions of Prime95.
With the nForce 790i SLI, we had to increase chipset voltage starting at FSB1640 to finally attain FSB1800 at 1.5V. The X38 reaches the same limit but we only had to increase power at this last FSB figure. Finally, the nForce 790i Ultra SLI which equips the ASUS motherboard didn’t need a voltage increase to reach the same limit. Does the difference come from the chipset, which is supposed to be better for overclocking, the bios, or motherboard? It’s difficult to say.
Otherwise, we noticed some bizarre behavior on the reference motherboard and ASUSTeK model: with a Q6600 (Stepping B3), we couldn’t go above FSB1400! Beyond this value, the reference version wouldn’t boot and the ASUSTeK deactivated the two last CPU cores. Results obtained with a QX6850 (still a 65nm but with a FSB1333 by default) were better with FSB1760 obtained on the ASUSTeK while it was unstable in FSB1600. With the reference model, we couldn’t go above FSB1560 in these conditions. Given the results obtained with the QX9770, we believe this is only a problem with the bios.
Overclocking – Memory
We end the testing by looking at increases in frequency with four DDR3 modules, here Kingston PC3-14400 CL8s. We made the transition into synchronous mode, FSB1600/DDR3-1600, and then increased frequency with 2 and then 4 modules.
As you can see, the nForce 790i SLI has a few problems in this area because installing the four modules means difficulty in increasing frequency. Otherwise, with the X38 results are the same with 2 or 4 modules and this also goes for the nForce 790i Ultra SLI. Here again, it’s hard to know if the motherboard, bios or chipset is responsible.
ConclusionWhile the nForce 780i was disappointing, this isn’t the case for the nForce 790i which has excellent results. The integration of a new SPP with native PCI-Express 2.0 support proves its worth. Moreover, DDR3 memory management is a plus for the future even if we have to admit that it is having some trouble imposing itself on the market. We did find the lack of DDR2 disappointing, contrary to the X38 and P35, which support both generations of memory.
The new SPP enables NVIDIA to catch up and even surpass Intel in terms of performances while at the same time offering SLI support with 2, 3 and 4 GPUs. Overclocking results are also up to par besides the problem with FSB increases with the Q6600. In theory, this is more involves the bios than the chipset given the results obtained with the QX9770. The chipset and bios are new and so we have to wait for them to mature a little in order to take advantage of their full potential.
Note that the ASUSTeK motherboard based on the nForce 790i Ultra SLI is significantly better in overclocking than the reference model based on the nForce 790i SLI ; however, it’s currently difficult to determine the reason. Is this related to the chipset which is announced as being better for overclocking or quite simply the motherboard and/or bios ?
In the end the nForce 790i Ultra/SLI is very likely to be the chipset of choice for those wanting to compose a ultra high end configuration. However, this is a niche in which this product will probably be confined given the price of DDR3 and that of the first nF790i Ultra SLI motherboards at 300 to 350 €. We can only hope that with time and the arrival of nF790i SLI motherboards in April the price drop under the 200 € mark.
Copyright © 1997-2014 BeHardware. All rights reserved.