 | |||
|
|
|
|
|
 |
NVIDIA GeForce 7900 GX2 Quad SLI Performance Preview May 01, 2006 |
|
 |
|
| |
Summary: Ever since it was first demonstrated at CES in Jan, we've been wondering how fast NVIDIA's Quad SLI performs in today's latest games. Well, today the wait is finally over and we can finally show you benchmarks! Inside you'll find benches of the GeForce 7900 GX2 running in a Quad SLI configuration against the latest cards from ATI and NVIDIA at 1600x1200, 1920x1200, 2048x1536, and 2560x1600. We also dabbled with a little bit of GX2 overclocking for good measure! You won't want to miss these results...
 Introduction | Page:: ( 1 / 13 ) |
At resolutions this high even a mighty GPU like the GeForce 7900 GTX will begin to chug in certain situations in some games, particularly once anti-aliasing is turned on. This is where NVIDIA’s SLI technology comes in.
With SLI, a second graphics card is added to the system, splitting the graphics rendering workload between both graphics cards for improved performance.
[image]
| <% print_image("01"); %> | <% print_image("02"); %> |
While it’s only been on the market for about a year now, NVIDIA’s SLI technology has gained considerable traction, NVIDIA’s sold well over a million nForce4 SLI motherboards to the public – despite their higher price tag – and most games on the market have out-of-the-box support for the technology.
[image]
| <% print_image("03"); %> | <% print_image("04"); %> |
But with the latest games using 3.0 pixel and vertex shaders more extensively, and HDR lighting and other eye candy features like volumetric effects and soft shadows being used more often, even two GeForce cards running in SLI may not be enough. To tackle this challenge, NVIDIA has developed Quad SLI, combining four graphics cards together to theoretically deliver four times the performance of one graphics card. But does Quad SLI really deliver twice the performance of a conventional SLI system? That’s the million dollar question everyone’s wanted to know since Quad SLI was first demonstrated to the public by NVIDIA and Dell at the Consumer Electronics Show at the beginning of this year.
[image]
| <% print_image("05"); %> | <% print_image("06"); %> |
In case you don’t recall, Dell and NVIDIA made a huge splash at CES, showing off Dell’s Renegade XPS 600 PC, a 4.26GHz Pentium D Extreme Edition rig tricked out with four GeForce 7800 GTX 512MB GPUs and a custom paint job. In our Quad SLI: Demystifying the Rumors article we explained how the technology works, what components you’ll need to get a Quad SLI system up and running yourself (as well as clarifying what components aren’t compatible with the tech), and also discussed the supply and availability questions that were brought up in the months following Quad SLI’s debut. The short answer is that Quad SLI is compatible with all nForce4 SLI X16 motherboards, both AMD and Intel-based, but if you want to get your hands on Quad SLI, you’ll have to buy a system from ABS, Alienware, Dell, or one of the other 20+ PC manufacturers NVIDIA lists on their Quad SLI website. For now standalone cards aren’t being sold to the retail market.
| Board analysis | Page:: ( 2 / 13 ) |
[image]
| <% print_image("07"); %> | <% print_image("08"); %> |
What’s in a name?
One of the more confusing aspects of Quad SLI is the actual name of the product. Over the past few months we’ve seen it referred to by multiple product names, ranging from “GeForce 7900 Duo” to the “GeForce 7900 GX2” and “GeForce 7900 Quad”. Judging by the EVGA graphics cards used in our ABS Quad SLI system, it looks like NVIDIA’s board partners are equally confused, as the back of both cards was clearly labeled “e-GeForce 7900GTX QUAD”.
Because of this, NVIDIA ultimately settled on the name GeForce 7900 GX2 for their Quad SLI cards.
[image]
| <% print_image("09"); %> | <% print_image("10"); %> |
Physically, the GeForce 7900 GX2 looks like no other graphics card on the market. The card features an extremely long PCB that’s just over 1 foot in length, making it the longest graphics card we’ve seen in recent memory. At first glance, one GeForce 7900 GX2 card looks like two distinct graphics cards, as NVIDIA literally uses two separate PCBs on one card – each PCB complete with one GPU and its associated memory, as well as the corresponding heatsink/fan unit – but it’s actually one graphics card. If you look closely you’ll see that only the PCB on the bottom has a PCI Express interface; the two PCBs are attached together with six screws.
[image]
| <% print_image("11"); %> | <% print_image("12"); %> |
The heatsink/fan unit on each GeForce 7900 GX2 is a single-slot unit, somewhat similar in design to the cooler used for the GeForce 7800 GTX 256MB but elongated for better performance. An aluminum heatpipe/heatsink combination is used to draw heat off the GPU and its memory modules while a small fan is used to keep everything cool. It’s important to note that while the backplate on the GeForce 7900 GX2 has vented holes on it, suggesting that hot air exhausts out the back of the system’s chassis, this actually isn’t the case, the vents on the back of the GeForce 7900 GX2 are merely decorative, hot air from the card’s cooler doesn’t exit here. The ducted enclosure NVIDIA has employed simply isn’t long enough for hot air to escape through.
| Board analysis (cont’d) | Page:: ( 3 / 13 ) |
[image]
| <% print_image("13"); %> | <% print_image("14"); %> |
As we’ve noted in the past when running two cards in an SLI configuration, one of the drawbacks of NVIDIA’s cooling system is that the primary graphics board on top tends to run about 10 degrees Celsius hotter than the card underneath it. This is because the standard cooling system on NVIDIA’s reference cards relies on drawing in nearby air underneath the graphics card to supply the cooler with fresh air.
To help combat this, for the GeForce 7900 GX2 NVIDIA has reversed the direction of airflow for the uppermost card on the 7900 GX2’s PCB. Rather than taking in air from underneath the board which would be restricted by the second GPU on the PCB, the board up top draws in cool air from above the graphics card. Three vents are drilled into the top of board’s PCB to accomplish this. Air is sucked in through these vents, and then passed across the GPU and memory before exiting the left side of the card’s ducted enclosure (again, air doesn’t exit your system’s case). This helps to keep temperatures down although unfortunately we can’t measure the before/after impact of this change directly.
[image]
| <% print_image("15"); %> | <% print_image("16"); %> | <% print_image("17"); %> |
Another key aspect you’ll notice on the GeForce 7900 GX2 is that each PCB on the card has its own 6-pin power connector. That means you’ll need four of these connectors to run a full Quad SLI configuration.
Technically, while each GeForce 7900 GX2 board ships with 1GB of memory total, keep in mind that each GPU on the GeForce 7900 GX2 board can only access 512MB of memory. NVIDIA’s reference specifications call for the use of 1.4ns memory modules, good for speeds up to 700MHz. Since the modules are only running at 605MHz, that leaves a little bit of headroom for overclocking.
The ABS System
To evaluate the performance of NVIDIA’s Quad SLI platform, ABS sent over one of their pre-release Ultimate M6 Sniper Quad-SLI systems for evaluation. The system was pretty much decked-out, outfitted with an Athlon 64 FX60 dual-core CPU running at 2.6GHz, 2GB of Corsair XMS3200 memory, two GeForce 7900 GX2 cards for Quad SLI, and a Creative Labs Sound Blaster X-Fi Elite Pro sound card. The complete system was housed inside a Thermaltake Armor case while the processor was cooled by a Venus CPU cooler also from Thermaltake.
| Test Systems | Page:: ( 4 / 13 ) |
System Setup
AMD Athlon 64 FX-57
ASUS A8R32-MVP Deluxe (CrossFire Xpress 3200)
ASUS A8N32 Deluxe (nForce4 SLI X16)
2GB OCZ DDR400 SDRAM
ATI Radeon X1900 XT 512MB
Catalyst 6.4
EVGA GeForce 7900 GX2 1GB
Driver version ForceWare 87.24
NVIDIA GeForce 7900 GTX 512MB
Driver version ForceWare 84.43
250GB Maxtor Hard Drive Maxline III SATA Hard Drive w/16MB Cache
Windows XP Professional SP1
DirectX 9.0c
Benchmarks
Call of Duty 2 1.01
Half-Life 2 Lost Coast
Far Cry 1.33 (1.4 patch for ATI cards)
F.E.A.R. 1.03
Quake 4 1.0.4
Elder Scrolls IV: Oblivion
| HDR: HL2 Lost Coast | Page:: ( 5 / 13 ) |
Half-Life 2: Lost Coast – Direct3D
| Quake 4 | Page:: ( 6 / 13 ) |
Quake 4 – OpenGL
| F.E.A.R. Performance | Page:: ( 7 / 13 ) |
F.E.A.R. – Direct3D
| Call of Duty 2 | Page:: ( 8 / 13 ) |
Call of Duty 2 – Direct3D
| Oblivion Mountains HDR | Page:: ( 9 / 13 ) |
Oblivion – Direct3D
| Oblivion Foliage HDR | Page:: ( 10 / 13 ) |
Oblivion – Direct3D
| Far Cry HDR | Page:: ( 11 / 13 ) |
Far Cry – Direct3D
| Overclocking | Page:: ( 12 / 13 ) |
FEAR
| Conclusion | Page:: ( 13 / 13 ) |
Now that we’ve had some time to test it, we can’t help but wonder if perhaps Quad SLI is a technology that was announced a little too early. As it stands right now, the Quad SLI solution implemented in NVIDIA’s GeForce 7900 GX2 is still a little rough around the edges. We saw more than our fair share of BSODs, crashes, and other glitches with Quad SLI. In fact, testing the system was actually a somewhat frustrating experience, at times we saw so many lockups and other errors that we were forced to reboot the system after each run. Oblivion in particular seemed to give the Quad SLI system the most frequent problems.
Now with that being said, fortunately there is a bright side – we’re pretty confident that these errors are driver-related, and have nothing to do with the hardware or the Quad SLI concept in general. Once we swapped out the GeForce 7900 GX2 Quad SLI setup with a pair of GeForce 7900 GTX cards and NVIDIA’s “beta” ForceWare 84.43 driver, the instability instantly went away. The GeForce 7900 GX2 also seemed to run more stable when one card was used, although admittedly it wasn’t 100% stable here either.
If you recall SLI’s initial launch in late 2004 NVIDIA ran into similar problems back then as well, and we all saw how that situation played out. NVIDIA’s driver team tackled the issues head-on and got them resolved, and even opened up new features such as SLI AA, SLI Profiles, and cross-vendor SLI support. By the end of last year SLI was quite robust.
The question now is how quickly can NVIDIA solve Quad SLI’s problems? That’s a question we may not know the answer to for many more months, hence the reason why we think NVIDIA would have been better off holding off on launching Quad SLI until it was more mature. Unfortunately, with their flashy CES demonstration with Dell, NVIDIA let the genie out of the bottle too early, putting unnecessary pressure on themselves to deliver in a reasonable amount of time (at CES Dell said the first systems would be available in Q2’06) and with good performance. NVIDIA should have waited to first announce Quad SLI later in the year; E3 would have been the perfect venue actually. It would have given NVIDIA more time to develop a clear message – there was an awful lot of confusion out there right after E3 – and more importantly, given NVIDIA’s driver team more time to produce a more stable product. As it stands today we have a strong feeling that anyone who spends $6,000+ on a Quad SLI system is going to end up sorely disappointed. Fortunately it looks like none of the system vendors have shipped Quad SLI systems to retail just yet, giving NVIDIA more time to perfect their driver. Who knows, maybe in a few weeks they can have everything resolved?
There are reports that NVIDIA’s already hard at work on a second-generation GeForce 7900 GX2 card that features a shorter PCB that should fit better in most system cases. This enhanced Quad SLI card may also hit the retail market as a standalone card (so you don’t have to buy a complete PC to experience Quad SLI) under the GeForce 7950 GX2 name. If these reports are true, this could be the Quad SLI card that truly lives up to the potential of the Quad SLI concept.