Summary: With their GeForce 7900 GTX and GeForce 7900 GT GPUs, NVIDIA's poised to try and take back the high-end crown from arch-rival ATI. These new GPUs don't come with the stratospheric price tags we've grown accustomed to. In fact, the 7900 GT delivers 7800 GTX-class performance for just $300. Rumors have been swirling for months now over the number of pipelines NVIDIA's G71 GPU contained. 24? 32? 48? Find out inside!
The key to making Moore’s law possible is the routine shrinking of the transistors that lie within today’s latest chips. As the transistors get smaller, manufacturers are able to incorporate more transistors into the same amount of space and thus increase performance. Think of a transistor as a “feature” – you want 24 pixel shaders instead of 16 or 12? That’s going to cost you more transistors.
If the chip manufacturer wants to reduce the manufacturing cost per chip instead of improving performance, smaller transistors can be used to reduce the overall size of the chip, thus allowing the manufacturer to get more chips per silicon wafer.
Of course, another important benefit of smaller transistors is that they require less voltage to operate. This in turn generally helps to combat heat assuming a similar chip design is used (to keep things simple, we’re not going to discuss other mitigating factors such as leakage). As a result, manufacturers can (and sometimes will) often crank up the clock speeds to increase performance.
One of the chief reasons NVIDIA grew so quickly to dominate the graphics semiconductor industry in the late 90s and into 2000 (in 2001 NVIDIA was named the fastest semiconductor company to reach $1 billion in revenue) is because they incorporated the latest manufacturing techniques into their products faster than anyone else. NVIDIA became famous for their 6-month product cycles where their products were replaced every six months by something newer and faster. This was previously unheard of in the graphics industry.
With most of their latest GPUs, NVIDIA uses 110 nanometer transistors. While the 110-nm manufacturing process has been very good to NVIDIA and is used in GPUs ranging from the GeForce 6200 TC all the way up to the latest GeForce 7800s, the process is nearly two years old now and doesn’t contain performance-enhancing features like low-k dielectric. Low-k is important, because as the manufacturing process gets smaller the circuits within the chip are crammed closer together. As clock speeds increase, these circuits can begin to interfere with one another. To prevent this from occurring, NVIDIA’s manufacturing partner, TSMC, uses low-k dielectric material on their smaller 90-nm manufacturing process. This low-k dielectric material is used to encapsulate the copper wires from each other, ensuring better performance (and thus, higher clock speeds) and lowering power requirements. Since the 110-nm process was originally intended to be used in value parts, low-k dielectric wasn’t used in order to help reduce manufacturing costs.
With today’s GeForce 7900 GTX/7900 GT launch, NVIDIA’s moving from TSMC’s larger 110-nm process to 90-nm. But is NVIDIA focusing on improving performance with these new GPUs and the 90-nm process or reducing manufacturing costs? For months now rumors have persisted at sites like the Inquirer that NVIDIA’s latest GPUs would contain more than the 24 pipelines found in the GeForce 7800 GTX. Let’s see what’s new with these GPUs.
While rumors were rampant that the G71 GPU used in the GeForce 7900 GTX and 7900 GT would contain as many as 32 pixel pipelines, G71’s architecture is fundamentally very similar to G70. The following chart summarizes the key differences between the two GPUs:
It’s also important to note that NVIDIA manages to pull all this off with fewer transistors than those that were required for the GeForce 7800’s G70 GPU, 278 million transistors in G71 versus 302 million for G70. NVIDIA is also quick to point out G71’s diminutive (for a high-end GPU) die size of 196mm2. This compares favorably to R580, which boasts a die size nearly twice as large at 352 mm2 and G70 at 334 mm2. With a considerably smaller die, G71 should be even cheaper for NVIDIA to produce than G70, assuming equal yields.
NVIDIA’s PureVideo technology carries over unchanged from previous GeForce GPUs, with three separate video engines allowing 720p, 1080i, and 1080p high definition output and support for a wide range of DVD/video playback features as well as H.264 acceleration.
Like the original GeForce 7800 GTX 256MB, NVIDIA clocks the graphics core and vertex engine independently of each other on the GeForce 7900 GTX and 7900 GT, with the vertex engine running 50MHz faster than the graphics core on the 7900 GTX card, and 20MHz higher than the rest of the graphics core on the GeForce 7900 GT. These margins also remain the same when overclocking via Coolbits or RivaTuner.
In terms of pricing, based on our conversations with NVIDIA’s board partners, we expect prices on the GeForce 7900 GT to start at about $299, with 7900 GTX boards retailing for $499. XFX and EVGA gave us even more specifics:
As you can see, both EVGA and XFX have planned to produce “stock” 7900 GT and 7900 GTX boards that follow NVIDIA’s reference speeds to the letter, and a line of factory overclocked cards that sell for a slight premium. While we don’t have the final clocks for EVGA’s boards, we do know that XFX is shooting for very high clock speeds with their highest-end GeForce 7900 GTX board, the XFX GeForce 7900 GTX XXX Edition 512MB board, with a core clock speed of 700MHz while the memory runs at 900MHz. For $50 less you can get the XFX GeForce 7900 GTX EXTREME 512MB which runs at 690MHz core/875MHz memory and at $525 the XFX GeForce 7900 GTX EXTREME 512MB
runs at 675MHz core/850MHz memory.
Among the GeForce 7900 GT boards, XFX’s most ambitious card is the XFX GeForce 7900 GT XXX (560MHz) 256MB, this card ships with a 560MHz graphics core and 825MHz memory for $350. Two 7900 GT Extreme boards are planned, one running at 550MHz core/815MHz memory, and a second SKU with a 520MHz core and 750MHz memory.
If you’re familiar with the board design of the GeForce 7800 GTX 512MB, NVIDIA’s GeForce 7900 GTX should look very familiar to you. That’s because NVIDIA’s using the exact same PCB, and has only made a few minute changes to the board’s design. Looking over the two reference boards side-by-side, you’d hardly notice a difference between the two cards, but there are a few subtle differences. Let’s quickly go over the basics first.
NVIDIA uses a combination of a copper base plate (which rests directly over the GPU) with a large aluminum heatsink, and four heat pipes to transfer heat off the GPU and memory modules. Topping it all off is a large fan, which supplies the cooler unit with the air from within your case. The fan itself is quite large, in fact it’s slightly larger than the 80mm case fans many of you are probably using in your system’s case right now. Fortunately though it doesn’t run loud, in fact it’s practically silent in use, even when overclocking. The fan sucks in the air from within your case, using it to help keep everything cool. From there the air from the fan passes out both sides of the heatsink, including outside your system case.
One key difference between the board design of the new 7900 cards in comparison to the 7800s is their new 7-pin HDTV-out mini-din connector. According to NVIDIA, with the new 7-pin min-din connector, you can plug an S-Video cable directly into the connector, or use a dongle for component or composite connections. With the old 9-pin connector used on previous GeForce cards required a dongle to use S-Video, component, and composite outputs.
In terms of power requirements, NVIDIA’s guidelines are pretty similar for the GeForce 7900 series as they were for the 7800s. NVIDIA recommend a 350W-400W power supply (22-26A on the 12V rail) for the GeForce 7900 GTX, while a 300W-350W power supply is recommend for the 7800 GT (20-23A on the 12V rail). For SLI rigs NVIDIA is even more specific:
Just to provide a frame of reference, NVIDIA’s “mid-range SLI PC” consists of an Athlon 64 4000+ with one hard drive, an nForce4 SLI motherboard with 1GB RAM, two optical drives, and one PCI sound card, while the “high-end SLI PC” contained an FX-57 CPU, nForce4 X16 SLI motherboard with 2GB RAM, two hard drives running RAID 0, two optical drives, and a PCI sound card. Needless to say that the power requirements for the 7900 GT/7900 GTX aren’t anything extraordinary for your typical hardware enthusiast nowadays.
The 7900 GT board
Since it runs at considerably lower clocks than the 7900 GTX, the 7900 GT gets by with a simple single-slot copper cooler. In fact, this is the smallest heatsink/fan unit NVIDIA’s used on a high-end card since the GeForce4 Ti 4600 over four years ago. The PCB itself is smaller than the GeForce 7800 GTX 256MB PCB as well, in fact it’s even shorter than the GeForce 7800 GT’s PCB. Due to this fact, and the board’s lower operating temperatures (and noise levels in comparison to the 7800 GTX 256MB), we’d feel very comfortable putting the GeForce 7900 GT in a SFF system. In fact, the GeForce 7900 GT has even sweeter SFF potential than the GeForce 6800 GT or Radeon X800 XL because of this.
Quad SLI is compatible with all of NVIDIA’s nForce4 SLI X16 motherboards; both AMD and Intel platforms are 100% compatible with Quad SLI. We’ve been told that Quad SLI requires in the neighborhood of an 850-1000W power supply (although NVIDIA’s working on getting this figure reduced), which is part of the reason why NVIDIA’s launching it with system builders first.
Quite frankly NVIDIA felt it was safer to launch Quad SLI in a controlled environment with system builders first, as OEMs can ensure that their shipping Quad SLI systems have proper cooling and power to adequately run four cards in a Quad SLI configuration safely. With an increasing number of board partners providing lifetime warranties on their retail graphics cards, this makes sense considering the ramifications.
We’ve received conflicting information on whether Quad SLI will ever make its way to retail via NVIDIA’s board partners, most likely it’s going to boil down to whether or not enough demand is there and if NVIDIA and their partners feel confident enough that Quad SLI could be done at the retail level without creating a tech support nightmare where lots of perfectly good cards are returned or damaged because end users systems aren’t setup properly to handle it.
The Quad SLI cards that are shipping now are based on NVIDIA’s G71 GPU rather than G70, although the cards are clocked differently than the GeForce 7900 GT/GTX. While NVIDIA’s Quad SLI boards ship with 24 pixel shaders, 8 vertex shaders, and 16 ROPs just like a 7900 GT or 7900 GTX, NVIDIA clocks their Quad SLI boards at 500MHz on the graphics core while the memory runs at 600MHz. That’s about 50MHz faster than a GeForce 7900 GT on the graphics core, but 60MHz slower on the memory than the 7900 GT. This means that more or less a single Quad SLI card should run roughly about as fast as two GeForce 7900 GTs running in SLI mode.
NVIDIA says they’ve selected these clocks in order to “enable maximum performance within the power and thermal envelope that we are allocating.” In other words, if they cranked up the clocks higher, the power requirements and thermals such a card would require would have been too high for what they were targeting.
Each Quad SLI card ships with 1GB of memory (512MB per GPU) for a total of 2GB in a full Quad SLI configuration. At this time NVIDIA has no plans to transfer Quad SLI to any other GPUs, for now its G71 technology only.
With NVIDIA unleashing a slew of new GeForce 7 products today, ATI’s decided to respond by reducing prices on their Radeon X1900 line and bringing back a GPU that’s been resurrected from the dead: R520. This chart sums things up nicely:
As you can see in the table above, on the high-end ATI’s slashed their prices for the Radeon X1900 XT/XTX, going from $549/$649 to $479 and $549 respectively. Meanwhile, the X1900 CrossFire board also sees a slight reduction, going from $599 at launch to $549 as of today. Going further down the X1K lineup, we see that the X1800 CrossFire 512MB board is now priced at $349, while the Radeon X1800 finally makes its grand introduction today at $329. If you recall the original R520 launch, ATI initially announced two X1800 XT SKUs, a 256MB Radeon X1800 XT priced at $500 and a 512MB Radeon X1800 XT card retailing for $600. Both cards boasted the same clock speeds.
Ultimately though only one X1800 XT SKU made it to market in a timely fashion, the more expensive 512MB board. 256MB X1800 XT boards never hit the retail channel en masse, in fact a quick scan of PriceGrabber reveals no 256MB X1800 XT boards on the market while PriceWatch has 256MB X1800 XT listings starting at $375. ATI states that their board partners have had the option of producing 256MB boards all along but that none of them have chosen to do until just recently, hence the “available” reference in the availability column. We should add that the same applies to the Radeon X1800 XL 512MB listed above – ATI’s board partners have had this option available to them, but have only recently decided to opt in on it. It’s here where things begin to get a little less clear.
ATI has begun shipping Radeon X1800 GTO GPUs to their board partners now, with retail availability expected by the end of this month. ATI’s X1800 GTO GPU is based off the exact same R520 graphics core as the Radeon X1800 XT/XL, only it has had some of its functionality disabled – the chip sports 12 pixel shaders, 8 vertex shaders, and is built with the same 512-bit (internal) ring bus memory controller with a 256-bit external memory interface that is used on elder X1800/X1900 cards. ATI also maintains 2x dual-link DVI connectors, while the chip runs at 500/500.
The Radeon X1800 GTO also supports ATI’s CrossFire technology although right now support is tied to the Radeon X1800 CrossFire Master card, just like the Radeon X1800 XL. This means that you’ll have to fork over $249 for a Radeon X1800 GTO card and then another $349 for a Radeon X1800 CrossFire 512MB card only for the more expensive X1800 CrossFire board to run with half of its memory disabled and only 12 functional pixel shaders. Remember, according to the CrossFire FAQ:
ATI realizes this is a far from practical solution, particularly for the more price conscious mainstream price segment, which is why they’re working on integrating support for peer-to-peer X1800 GTO dongle-less CrossFire support into an upcoming Catalyst driver (just like the Radeon X1600 and X1300 cards do), but ATI couldn’t provide an ETA on when this feature would be integrated into the Catalyst driver suite.
3DMark 06– Direct3D
3DMark 06– Direct3D
Battlefield 2 – Direct3D
Quake 4 – OpenGL
LOMAC – Direct D
Pacific Fighters – OpenGL
F.E.A.R. – Direct3D
Call of Duty 2 – Direct3D
Serious Sam 2 – Direct3D
Far Cry – Direct3D
Bars marked in yellow denote single-card configurations. The blue bars are all SLI systems running with various levels of AA.
Between the two GPUs, the slam dunk product in our opinion is NVIDIA’s GeForce 7900 GT. For $300 you’re basically getting a graphics card that delivers performance that’s a little greater than the GeForce 7800 GTX (typically the 7900 GT runs between 5-8% faster than a 7800 GTX), but in a smaller, cooler-running package. With the GeForce 7900 GT, NVIDIA’s basically established a new level of performance at the $300 price point that ATI currently doesn’t match: looking over the latest PriceGrabber and PriceWatch prices, the closest equivalent ATI and their board partners currently have to the GeForce 7900 GT is the X1800 XL, which currently sells for about $310. As you probably know by now, the X1800 XL was positioned against the GeForce 7800 GT, not a card that performs like the GeForce 7900 GT, there’s just no comparison.
ATI’s working on getting 256MB Radeon X1800 XT boards to market via their board partners and if they are successful in doing this a 256MB X1800 XT board could give the GeForce 7900 GT some competition based on our benchmarks with a 512MB Radeon X1800 XT, but that’s not what the market looks like today. In addition, ATI currently lists a $329 MSRP for the X1800 XT 256MB, pricing it slightly higher than today’s $299.99 GeForce 7900 GT boards. We’ll just have to see how the 256MB X1800 XT performs and what the final price looks like once boards are available, but in any case the GeForce 7900 GT is well positioned to take on this card if it eventually ships.
In the case of the GeForce 7900 GTX, NVIDIA cranks up the clock speeds significantly to deliver a part that outruns the so-called “GeForce 7800 Ultra” of yesteryear, the GeForce 7800 GTX 512MB. This is a pretty nice achievement, especially considering the GeForce 7900 GTX’s $500 price tag, but it doesn’t have the clear cut victory that the GeForce 7900 GT currently enjoys. ATI’s Radeon X1900 XT boards are currently selling for as low as $509 online, and based on our benchmarks, deliver very competitive performance in games like F.E.A.R., Call of Duty 2, Battlefield 2, and Far Cry with HDR. (NVIDIA continues to dominate for those of you who are into flight sims though.) Thanks to its quad-heat pipe cooling solution, the GeForce 7900 GTX runs cooler and quieter than the X1900 XT/XTX boards while NVIDIA’s SLI is clearly more robust than CrossFire but ATI’s got a very competitive part in this segment of the market. Right now there really is no decisive winner here, your final decision will most likely boil down to what types of games you play on your PC.
Based on our conversations with some of NVIDIA’s board partners though, they’ve got some pretty aggressive plans for ratcheting up the stock speeds of the GeForce 7900 GTX on their overclocked line of boards. Perhaps one of these boards may manage to grab a few of the benchmarks that the X1900 XTX holds today…
|© Copyright 2003 FS Media, Inc.|