NVIDIA nForce 780i SLI - BeHardware
Written by Marc Prieur
Published on February 1er, 2008
While NVIDIA rapidly dominated the chipset market for the AMD platform successively edging out VIA due with the evolution of its nForce since 2001, it’s a different story for the Intel platform. Here it should be pointed out that a bigger adversary is involved although this hasn’t stopped NVIDIA from persevering and with good reason of course. The creator of the GeForce thus comes back with a new high end chipset, the nForce 780i SLI, which replaces an aging nForce 680i SLI launched at the end of 2006.
A last minute delayInitially, the nForce 780i SLI should have been launched at the beginning of November before it was delayed a month and a half. In fact, behind the delay there was a hidden compatibility problem with the Yorkfield, in other words, the 45nm version of the Intel quad core processor, which necessitated a redesign of motherboards.
Note that the nForce 680i SLI is also affected by the same problem although some manufacturers did specify its Yorkfield compatibility in the past. If EVGA, for example, adeptly resolved this situation by offering to exchange the nForce 680 for the nForce 780 and 80 €, this was unfortunately an isolated case.
An nForce 690i SLI ?
The main advantage of the nForce 780i SLI is in terms of PCI-Express which is now in its 2.0 version. You may recall, this now means attaining a data speed of 500 MB /s on each PCI-Express lane versus 250 MB /s for the first version.
This support however is a bit particular because it is done via the addition of a third chip, the nForce 200, to the already present SPP (northbridge) and MCP (southbridge). Connected in PCI-Express to the SPP, the new chip allows managing 32 PCI-Express 2.0 lanes which can be attributed to 2 to 4 graphic ports. In total, the bandwidth available between the nForce 200 and graphic cards is 16 GB /s. On the other hand, this figure is only 7.2 GB /s between the nForce 200 and the SPP as it is interfaced with the latter via 16 PCI-Express 1.0 lanes that are boosted 80% compared to their normal speed.
This doubling of the number of PCI Express lanes on the SPP level enables NVIDIA to offer a triple SLI solution with the nForce 780i which uses 3 ports cabled in x16, one of which is relayed to the MCP. On the 680i, the third port was in x8.
While NVIDIA had to rely on a new chip for PCI Express, this is actually because the SPP and MCP aren’t too different from the ones used on the nForce 680/650i. While there is Yorkfield compatibility due to a modification on the motherboard, at the current time we are still officially limited to FSB1333. This is when Intel will release the QX9770 in January which uses a FSB1600. In terms of memory, there is DDR2, still up to DDR2-1200, where Intel also offers DDR3 on the X38 (however note that DDR3 is actually of very limited interest at this time).
The southbridge remains the same and thus offers two Gigabit Ethernet ports which can be combined via DualNet technology as well as six SATA ports configurable in RAID 0, 1, 0+1 or 5.
In practice, PCI Express 2.0
In practiceFor this test, we were able to get our hands on an EVGA nForce 780i SLI motherboard based on NVIDIA’s reference design, which by the way is very close to that of the nForce 680i SLI. In addition to capabilities inherent to NVIDIA chipsets, there is also the presence of an HD Audio Realtek ALC888S codec and a FireWire Texas Instruments TSB43AB22A controller. On the other hand, the motherboard does not offer an eSATA port. We can see there are both power and reset buttons directly on the motherboard, something very practical for those who like to play with settings and conduct tests outside of the tower casing.
Cooling is assured by a heatpipe based system which relays the MCP, SPP and CPU power stage. It is normally entirely passive but in the case of overclocking an optional fan is included that can be clipped on the main SPP radiator. Unfortunately, it’s quite noisy.
Contrary to what happened with the nForce 680i SLI, the Core 2 Extreme QX9650 functioned perfectly here. On the other hand, the QX9770 did not boot. This is nothing too dramatic given that this CPU isn’t even available yet; however, we can hope the opposite will happen via a bios update even if the FSB1600 is not officially supported by the nForce 7.
In order to evaluate the new chipset, we compared it to the X38 by using an ASUSTeK P5E motherboard whose particularity is to combine the X38 and DDR2. Initially we were going to integrate the nForce 680i SLI for comparison but unfortunately our test board gave out in the middle of tests. Anyhow, the first results showed equivalent performances to the nForce 780i SLI in terms of memory, RAID and network.
PCI Express 2In practice, what does PCI Express 2 add to the nForce 7?
To answer this question, we used four PCI-E 2.0 video cards, one after the other, on the first and third PCI Express x16 port as they are in versions 2 and 1, respectively. These were two Radeon HD 3870s and two GeForce 8800 GTs, one a 256, the other a 512 MB.
For tests, after the use of various programs we chose 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’s supplementary memory bandwidth makes the difference.
As you can see, 256 MB cards really tire out quickly. Also, the fact that they are on the PCI-E 2.0 port doesn’t radically change the situation because gains are 15% for NVIDIA and 6% for ATI. In their 512 MB versions, the 8800 GT largely surpasses the HD 3870 while this was the opposite case for the 256 MB cards.
Is this related to the nForce 780i SLI’s rather particular management of PCI-E 2.0?
Without a doubt we can say, yes, given the X38’s performances with 256 MB cards. The gain is ‘’only’’ 5.6% for ATI but for the 8800 GT it attains 21%.
Memory performancesWe 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 rather noticeably affected by the speed of the memory subsystem, something that isn’t always the case. Tests were carried out with a Command Rate of 1T in DDR2-800 4-4-4-12, and with 2T in DDR2-1066 5-5-5-15, FSB1066 and FSB1600 all the while keeping a fixed CPU frequency of 2.4 GHz.
To begin with, here are the results of tests with ScienceMark :
In FSB1066, the two chipsets have rather close performances in terms of bandwidth. On the other hand, the nForce 7 is ahead in memory latency. The transition to FSB1600 is more to the X38’s benefit, notably for bandwidth. So what about more practical performances?
Whether it’s in WinRAR or Crysis, the X38 platform proves to be the fastest and this was independent of the type of memory or FSB. Note that with the X38, DDR2-1066 is faster than DDR2-800 while the nForce 7 is faster in DDR2-800. In WinRAR, the difference is still 10% for the X38 while in Crysis the gap is only 3%.
Serial ATA, RAID, Gigabit
Serial ATA and RAID
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 type of situation, NCQ can be particularly useful because there can be multiple simultaneous commands. 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 in the following configurations : 1 single drive, 2 drives in RAID 0 and 3 drives in RAID 5.
While with one drive performances are close, in RAID 0 and 5 the nForce 7 plateaus out at four simultaneous commands. Beyond this figure, performances stagnate to the point that with 128 commands a single drive has similar results! On the other hand, with the X38 performances continue to improve.
We now move on to file copying. We measure reading and writing speeds, as well as the copying of 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.
With a single drive, the nForce 7 is slightly ahead but it’s the opposite in RAID 0. Performances of the two chipsets in RAID 5 are rather catastrophic in cases where writing is significant. This doesn’t seem to have changed much since the first tests of the nForce 4 when we noticed that RAID 5 « software » wasn’t at ease in this area.
NetworkPioneer in the domain of Gigabit network integration to the chipset with the nForce 3 250 GB, at the time NVIDIA was mostly competing with solutions using the “good old” PCI bus. Since then, motherboards based on rival chipsets more often integrate PCI-Express chips such as the Marvell 88E8056 controller which has a more appropriate bandwidth for this type of network. Nonetheless, besides performances we also wanted to take a look at CPU use. To test speeds, we used the program PCATTCP with a buffer size of 65536.
In terms of speed, the two solutions are close but the advantage still goes to NVIDIA for data transmission. The only thing is that with a 100 MB/s limitation on the Marvell solution, it’s most likely that we are more limited by the sub-system drive than the network. The nForce 7 actually manages to come out the best in terms of processor use which is notably lower. The biggest difference is felt in data transmission. These measurements having been carried out with the Q6600, the Marvell solution consumes roughly 84% of one of the four cores when transmitting at maximum speed!
We then measured card consumption with an ammeter that gave readings of the different lines on the ATX block. This allowed us to differentiate from processor consumption which arrives via the ATX12V as well as that of other peripherals directly connected to the power supply.
The nForce 780i SLI’s rather high consumption stands out because it shows 65.2 watts for the EVGA nForce 780i SLI motherboard versus 47.9 watts for the ASUSTeK P5E. This is a difference of 17.3 watts.
Overclocking – CPUMoving on to overclocking the chipset with a Core 2 QX9650, we only give you the frequencies that were validated by four 15 minute sessions of Prime95.
With the nForce 7, we were able to attain 440 MHz for the FSB with the chipset’s default voltage of 1.3V and 450 MHz by increasing to 1.4V. On the X38, 450 MHz was obtained with the default voltage of 1.25V, 460 MHz at 1.35V, and then 470 MHz at 1.45V. There is therefore a slight advantage to the X38; however, in both cases these figures are still good even if we would not be able to overclock a potential quad core FSB1600 (and non Extreme Edition) just via the FSB. Note that for the moment and despite these results, the future QX9770 and its FSB1600 are not officially supported by the nForce 780i SLI.
Overclocking – Memory
We end the testing by looking at increases in memory frequency carried out with 2x1 GB of Crucial Ballistix Tracer PC2-6400 configured in 5-5-5-15 at 2.25V and then with 4x1 GB of the same type of memory. Only the frequencies that were validated by four 15 minute sessions of Prime95 are given. In both cases, tests started in DDR2-1066 with FSB1066 and 1333 and then we increased the FSB in order to attain higher memory frequencies while keeping the same ratios
For the nForce 7, results were more deceiving because starting at FSB1066 we could only stabilize memory in DDR2-1080 with two modules. Even in DDR2-1066 the 4 modules were not functioning correctly. In FSB1333, with 2 modules we were able to attain DDR2-1140 while with four modules DDR2-1066 was only stable after the transition from 1.3 to 1.5V on the chipset.
The X38 was more at ease in this domain. Starting at FSB1066/DDR2-1066, we were able to attain DDR2-1200 with 2 and 4 modules, in the latter case there was an increase in chipset voltage to 1.35V starting at DDR2-1180. With settings starting at FSB1333/DDR2-1066, we were able to attain DDR2-1200 with 2 modules and 1.35V and the same voltage was necessary to reach DDR2-1180 with 4 modules.
ConclusionYes, the nForce 780i SLI does allow combining Intel 45nm quad core and SLI, something that wasn’t possible with its predecessor. But actually this is its sole interest because PCI Express 2 support does not offer any significant advantage as long as it is behind that of Intel’s X38.
This isn’t the only area where the X38 is superior. It also shows better performances whether related to the memory interface, RAID 0/5 in multiple simultaneous access or in RAID 5 in writing, even if it remains relatively low. In addition, the X38 proves to be significantly more at ease at managing its 4 DIMM slots at high frequencies. Finally, it’s slightly better in terms of FSB increases while at the same time consuming less than the nForce 7. The presence of a 3rd chip doesn’t help the nForce here.
In addition to SLI, we have to give the nForce 7 credit for its low processor use in Gigabit network transfers although this doesn’t weigh too heavily in the era of quad core CPUs. In the end, the nForce 780i SLI, which is more of an nForce 690i SLI, is therefore somewhat disappointing and limited to those impatient users looking to combine a Yorkfield and SLI.
If this type of configuration interests you, we recommend waiting a few weeks for the arrival of the nForce 790i SLI, which should according to the latest rumors be equipped with a new SPP that has native FSB1600, PCI-E 2.0 and DDR3 support. Given the high price for this type of memory, we hope its management will not be exclusive and that in practice DDR2 will also be supported.
Copyright © 1997-2013 BeHardware. All rights reserved.