 | |||
|
|
|
|
|
 |
NVIDIA GeForce 7800 GTX 512MB Performance Preview November 14, 2005 |
|
 |
|
| |
Summary: With 512MB of GDDR3 memory running at 850MHz (1.7GHz effective) and a 550MHz core clock, the GeForce 7800 GTX 512MB is designed to deliver extreme performance. But does the board deliver? See how the board performs and overclocks as we pit it against five other cards, including ATI's X1800 XT 512MB in all the latest games and benchmarks inside!
 Introduction | Page:: ( 1 / 19 ) |
The evolution of G70
Last June, NVIDIA launched the GeForce 7800 GTX, the company’s first GPU to be built on NVIDIA’s second-generation shader model 3.0 architecture (dubbed CineFX 4.0) and the first product in the GeForce 7800 family. Thanks to its brand new G70 graphics core, the GeForce 7800 GTX delivered considerably more performance than any other graphics card on the market, often outperforming its predecessor, the GeForce 6800 Ultra, by a factor of two.
[image]
| <% print_image("01"); %> | <% print_image("02"); %> |
| <% print_image("03"); %> | <% print_image("04"); %> |
To do this, NVIDIA incorporated a number of improvements into G70 designed to boost performance. G70s vertex shaders were enhanced to speed up geometry processing, while tweaks were made in both the pixel and vertex shaders that made both units better at handling multiply-add (MADD) math operations. These computations are commonly used for lighting (for example, in effects like refraction and reflection, or embossing), normal map calculations (adding depth and height via normal maps to what are actually flat objects), and many other operations.
According to NVIDIA, the enhancements they’ve integrated into G70 increase pipeline efficiency by 50% on a clock-for-clock basis, while the tweaks made to the vertex shading units have sped up triangle setup by over 30%. NVIDIA also improved texture fetching and cut cycles in gamma adjusted rotated grid AA by a similar margin.
The most notable aspect of G70 though was arguably the increased number of pixel and vertex shaders.
NVIDIA integrated 24 pixel pipelines into the GeForce 7800 GTX’s G70 core - eight more pipes than any other previous architecture - while on the vertex processing side, eight units were incorporated into the 7800 GTX, that’s two more vertex units than the six found in the GeForce 6800 Ultra’s NV40 core. When combined with the GeForce 7800 GTX’s high clock speeds, 430MHz on the graphics core and 600MHz on the memory (1.2GHz effective), the GeForce 7800 GTX was without equal, nothing else on the market came close to delivering the features and performance of the GeForce 7800 GTX.
NVIDIA then followed the GeForce 7800 GTX up with another G70-based product, the less expensive GeForce 7800 GT.
The GeForce 7800 GT featured all the innovations found in GeForce 7800 GTX, only the number of pixel and vertex pipelines had been reduced to 20 and 7 respectively. NVIDIA also turned down the clock speeds to 400MHz on the G70 graphics core and 500MHz on the memory (1.0GHz effective). The result was a product that ran 10-20% slower than the GeForce 7800 GTX, but sold for over $100 less than the 7800 GTX, making it more appealing to a larger audience. NVIDIA and their board partners have even partnered with Activision to bundle a copy of Call of Duty 2 with the GeForce 7800 GT for a limited time on select cards, making the GeForce 7800 GT an even more tempting purchase.
After months of delays, ATI finally responded with the RADEON X1800 family, including the RADEON X1800 XT and X1800 XL. Early benchmark results gave the edge to ATI in D3D applications, while NVIDIA reigned supreme in OpenGL titles like DOOM 3. Since then however ATI has released a newer driver that delivers substantially improved OpenGL performance for the X1800 series, in our Quake 4 High-End Graphics Shootout article the RADEON X1800 XT 512MB actually outran NVIDIA’s flagship GeForce 7800 GTX.
[image]
| <% print_image("05"); %> | <% print_image("06"); %> |
To combat this new threat, NVIDIA has commissioned their third graphics product using the proven G70 GPU, the GeForce 7800 GTX 512MB! With significantly boosted clocks and 512MB of GDDR3 memory onboard, the 7800 GTX 512MB is designed to do battle specifically with the RADEON X1800 XT 512MB. Let’s see how this new card performs shall we?
| Specifications, board analysis | Page:: ( 2 / 19 ) |
While the board is still 100% based on NVIDIA’s G70 GPU with 24 pixel pipelines and eight vertex shaders, NVIDIA’s bumped up the clock speeds dramatically. The graphics core is clocked at 550MHz, that’s 120MHz higher than the GeForce 7800 GTX 256MB. This change improves fill-rate from 10.3Mtexels/second to 13.2Mtexels/sec, an improvement of over 20%.
On the memory side, things are even more exciting, as NVIDIA’s reference specifications call for a speed of 850MHz. That’s 250MHz higher than the GeForce 7800 GTX, improving memory bandwidth by 16GB/sec alone, from 38.4GB/sec in the 7800 GTX 256MB, to 54.4GB/sec in the GeForce 7800 GTX 512MB. To put things in perspective, NVIDIA’s current mainstream offering, the GeForce 6600 GT, has 16GB/sec of peak memory bandwidth total. We’ve provided the following chart, which summarizes the changes:
| High-End Graphics Comparison | |||||
| GeForce 7800 GTX 512MB | GeForce 7800 GTX 256MB | RADEON X1800 XT | GeForce 7800 GT | RADEON X1800 XL | |
| Core Clock (MHz) | 550 | 430 | 625 | 400 | 500 |
| Memory Speed (MHz) | 850 | 600 | 750 | 500 | 500 |
| Memory Interface Width | 256-bit | 256-bit | 256-bit | 256-bit | 256-bit |
| Memory Size & Type | 512MB GDDR3 | 256MB GDDR3 | 256MB GDDR3 & 512MB GDDR3 | 256MB GDDR3 | 256MB GDDR3 |
| Memory Bandwidth (GB/sec) | 54.4 | 38.4 | 48 | 32 | 32 |
| Pixel Fill Rate (Gpixels/sec) | 8.8 | 6.88 | 10.0 | 6.4 | 8.0 |
| Texel Fill-Rate (Gtexels/sec) | 13.2 | 10.3 | 10 | 8.0 | 8.0 |
| Pixel Pipelines | 24 | 24 | 16 | 20 | 16 |
| Vertex Units | 8 | 8 | 8 | 7 | 8 |
| |||||
[image]
| <% print_image("07"); %> | <% print_image("08"); %> |
As a result of the new speeds, power consumption is up slightly. Whereas the GeForce 7800 GTX is rated at 100W peak power consumption, the 7800 GTX 512MB is rated at 120W peak. NVIDIA recommends at least a 350W power supply with 22 amps on the 12V rail for single card operation, while NVIDIA recommends a power supply with at least 500W and 30 amps on the 12V rail for SLI. These are the same guidelines NVIDIA recommends for the GeForce 7800 GTX 256MB.
To hit such high speeds, NVIDIA says that they’ve tuned the G70’s fabrication process to speed up certain data paths inside the chip. NVIDIA has also made a few tweaks to the GeForce 7800 GTX 512MB’s PCB design to facilitate the higher clocks. The most notable change you’ll spot though is the GeForce 7800 GTX 512MB’s new cooler.
The reference board up close
As you can see in the pictures, NVIDIA has outfitted the GeForce 7800 GTX 512MB with a completely different cooling unit than its predecessor. This is actually the exact same heatsink/fan unit NVIDIA and PNY use on the high-end Quadro FX 4500 workstation boards. (Leadtek also uses this same cooler on their WinFast PX7800 GTX TDH MyVIVO Extreme card.)
[image]
| <% print_image("09"); %> | <% print_image("10"); %> | <% print_image("11"); %> |
The cooler starts with a copper baseplate, which rests directly over the GeForce 7800 GTX 512MB’s G70 graphics core. Heat is then transferred from the copper plate to an array of long, aluminum-based heat pipes. Inside each of the heat pipes is distilled water. As the GPU begins to heat up, the distilled water in the heat pipe begins to boil, forcing hot vapor to the other end of the heat pipe where it is cooled. From there the vapor condenses back to the liquid phase and returns to the other end of the heat pipe. This cycle is continually in motion, working to keep the graphics core cool.
[image]
| <% print_image("12"); %> | <% print_image("13"); %> | <% print_image("14"); %> |
Helping to keep all this cool is a hefty dual-slot heatsink/fan unit, which is composed of aluminum. The heatsink itself is massive, nearly swallowing up the entire 7800 GTX 512MB board. It has dozens of long, tall fins, further increasing its surface area. NVIDIA then finishes the cooler off with a large cooling fan. Air from the fan passes out the sides of heatsink, including outside the system case.
With such a large fan onboard, you may think that the GeForce 7800 GTX 512MB generates a considerable amount of noise, however the opposite is actually the case. In operation, the fan runs nearly silent, the heatsink/fan unit on our Athlon 64 FX-57 CPU actually ran louder than the GeForce 7800 GTX 512MB! This is easily the quietest fan we’ve ever seen on a high-end NVIDIA reference board, and will give Arctic Cooling’s VGA Silencer series a run for its money in the noise department. But it doesn’t stop there, in addition to the new cooling system, NVIDIA outfits the 7800 GTX 512MB board with 900MHz memory modules, giving overclockers a little bit more headroom in their overclocking endeavors.
[image]
| <% print_image("15"); %> | <% print_image("16"); %> | <% print_image("17"); %> |
Some of NVIDIA’s board partners will no doubt use this to provide factory overclocking. We’ve already received XFX’s GeForce 7800 GTX 512MB card, and sure enough it was clocked higher than stock, sporting a core clock frequency of 580MHz (30MHz over default) while its memory was running at 865MHz (15MHz over stock).
| Performance impact of 512MB memory | Page:: ( 3 / 19 ) |
Bigger is better?
For most people, everything has to be bigger, better, and faster. This theory applies to practically everything. From super sizing your fries at McDonald’s to opting for a shiny new big-screen TV.
It’s not surprising then that this same bigger is better mentality also applies to computers. We always want bigger hard drives, crave on faster and faster GHz+ processors, and insist on more and more RAM. While many of these upgrades are well worth the money, some of them play on this belief, and just aren’t cost-effective.
We wanted to see where the GeForce 7800 GTX 512MB’s performance really came from – did the additional 256MB of memory actually make a difference, or are the board’s enhanced clock speeds more responsible for the board’s added performance?
To answer this question we underclocked our GeForce 7800 GTX board to the same speeds as NVIDIA’s GeForce 7800 GTX 256MB: 430MHz on the core, 600MHz on the memory. Let’s take a look at the results:
F.E.A.R.
Quake 4
Looking over our first batch of results, it’s hard to make a case for the additional memory found on the GeForce 7800 GTX 512MB. In fact, the extra memory adds a little bit of latency, actually hampering performance slightly. Fortunately, things did brighten up for the GeForce 7800 GTX 512MB once we booted up Half-Life 2 and Call of Duty 2…
| 512MB Memory Scaling (cont'd) | Page:: ( 4 / 19 ) |
Half-Life 2
Call of Duty 2
In Half-Life 2 with 4xAA/16xAF, the additional memory found on the GeForce 7800 GTX 512MB improved performance by 2% at 1280x1024 and 1600x1200, and 7% at 2048x1536. Meanwhile, in Call of Duty 2, the results were even better, we saw a 9% improvement at 1600x1200 and 2048x1536.
Based on our CoD 2 results last week with the X800 XL 512MB, we expected even higher figures for the GeForce 7800 GTX 512MB in Call of Duty 2. Since that test we’ve changed the testbed configuration and just as importantly, enabled the “Optimize for SLI” setting in the CoD 2 graphics menu. Since publishing our first round of CoD 2 test results, we’ve found that this setting boosts performance for non-SLI configurations and have been using it ever since, including the results above. That likely accounts for part of the difference.
| Test Systems | Page:: ( 5 / 19 ) |
System Setup
AMD Athlon 64 FX-57
ASUS A8N32-SLI Deluxe
2GB OCZ DDR400 SDRAM
ATI RADEON X1800 XT
ATI RADEON X1800 XL
CATALYST 5.11
NVIDIA GeForce 7800 GTX 512MB
NVIDIA GeForce 7800 GTX
NVIDIA GeForce 7800 GT
NVIDIA GeForce 6800 GT
Driver version ForceWare 81.89
250GB Maxtor Hard Drive Maxline III SATA Hard Drive w/16MB Cache
Windows XP Professional SP1
DirectX 9.0c
Benchmarks
IL-2 Sturmovik: Forgotten Battles
Call of Duty 2
Serious Sam 2
Half-Life 2
F.E.A.R.
Battlefield 2
Quake 4
| Half-Life 2 4xAA/16xAF | Page:: ( 6 / 19 ) |
Half-Life 2 – Direct3D
| Battlefield 2 | Page:: ( 7 / 19 ) |
Battlefield 2 – Direct3D
| Quake 4 | Page:: ( 8 / 19 ) |
Quake 4 – OpenGL
| IL-2 4xAA/16xAF | Page:: ( 9 / 19 ) |
IL-2: FB – OpenGL
| F.E.A.R. Performance | Page:: ( 10 / 19 ) |
F.E.A.R. – Direct3D
| F.E.A.R. Performance 1600x1200 | ||
| Card | Min FPS | Max FPS |
| GeForce 7800 GTX 512MB | 43 | 137 |
| GeForce 7800 GTX | 34 | 101 |
| GeForce 7800 GT | 25 | 80 |
| GeForce 6800 GT | 17 | 65 |
| RADEON X1800 XT | 26 | 78 |
| RADEON X1800 XL | 19 | 62 |
| ||
| F.E.A.R. 4xAA/16xAF | Page:: ( 11 / 19 ) |
F.E.A.R. – Direct3D
| F.E.A.R. Performance 1280x1024 | ||
| Card | Min FPS | Max FPS |
| GeForce 7800 GTX 512MB | 32 | 107 |
| GeForce 7800 GTX | 23 | 87 |
| GeForce 7800 GT | 19 | 74 |
| GeForce 6800 GT | 12 | 56 |
| RADEON X1800 XT | 20 | 90 |
| RADEON X1800 XL | 18 | 71 |
| ||
| Call of Duty 2 | Page:: ( 12 / 19 ) |
Call of Duty 2 – Direct3D
| Call of Duty 2 4xAA/8xAF | Page:: ( 13 / 19 ) |
Call of Duty 2 – Direct3D
| Serious Sam 2 4xAA/16xAF | Page:: ( 14 / 19 ) |
Serious Sam 2 – OpenGL
| Serious Sam 2 Performance 1600x1200 | ||
| Card | Min FPS | Max FPS |
| GeForce 7800 GTX 512MB | 38.3 | 96.9 |
| GeForce 7800 GTX | 30 | 68.2 |
| GeForce 7800 GT | 25.4 | 55.2 |
| GeForce 6800 GT | 17.9 | 37.6 |
| RADEON X1800 XT | 27.5 | 64.5 |
| RADEON X1800 XL | 24.9 | 62.4 |
| ||
| Overclocking | Page:: ( 15 / 19 ) |
F.E.A.R. 4xAA/16xAF
Quake 4 4xAA/16xAF
| SLI Gaming: Quake 4 | Page:: ( 16 / 19 ) |
Quake 4 – OpenGL
| SLI Gaming: FEAR | Page:: ( 17 / 19 ) |
F.E.A.R. – Direct3D
| SLI Gaming: IL-2 | Page:: ( 18 / 19 ) |
IL-2 – OpenGL
| Conclusion | Page:: ( 19 / 19 ) |
Call of Duty 2 performance improves by 21% at 1600x1200 with 4xAA/8xAF and 24% at 2048x1536, while we saw a whopping 31% improvement in Half-Life 2 at 2048x1536 with 4xAA/16xAF over the GeForce 7800 GTX 256MB. In benchmark after benchmark the GeForce 7800 GTX 512MB consistently finishes about 20-30% faster than the GeForce 7800 GTX 256MB, and sweeps all tests against its intended competition, the RADEON X1800 XT 512MB. Here the GeForce 7800 GTX 512MB typically comes out ahead by 8-12%, although there are exceptions such as IL-2 Sturmovik, where the margins are wider, and two cases in CoD 2 where the boards are separated by 5% and 6%. NVIDIA can now rest easily, with the GeForce 7800 GTX 512MB they’ve got the performance crown in OpenGL once again.
All this performance does come at a pretty high cost though. The GeForce 7800 GTX 512MB officially carries an MSRP of $650. This is high enough to make the GeForce 7800 GTX 256MB look like a charity case, and nearly twice the price of what many GeForce 7800 GT boards are selling for at retail.
In NVIDIA’s defense, their previous card launches of late haven’t remained at MSRP for very long. GeForce 6800 GS boards for example were selling for as low as $200 by the end of launch day last Monday. Hopefully history will repeat itself again for the GeForce 7800 GTX 512MB, boards should be available for purchase starting today.
On another note, we’re personally a little surprised NVIDIA decided to designate this board as the GeForce 7800 GTX 512MB. Why not call it GeForce 7800 Ultra 512MB? Clearly this board delivers “Ultra” caliber performance in comparison to its predecessor the 7800 GTX 256MB, but more importantly, by using the Ultra moniker, NVIDIA would leave room for a 256MB SKU with the improved speeds found on today’s board. If ATI is ever able to get their availability problems resolved with the X1800 XT, the 256MB RADEON X1800 XT board could be quite a competitor to the GeForce 7800 GTX 256MB thanks to its high clock speeds. Our 256MB vs 512MB results with the GeForce 7800 GTX today suggest that 512MB of memory may not play as big of a role in performance with today’s latest games as we’d thought. As of this writing, the only X1800 XT SKUs at retail is the 512MB board like the card we tested today, but ATI also announced a $499 X1800 XT 256MB SKU with the same clock speeds as the 512MB board.
If the X1800 XT 256MB board ever hits retail en masse and performs similarly to its 512MB sibling, the GeForce 7800 GTX could quickly become yesterday’s news. But if NVIDIA had left the door open by announcing today’s GeForce 7800 GTX 512MB as an Ultra product, they would have had room to announce a 256MB board boasting the same 550/850MHz speeds. We have a feeling this would be a pretty sweet graphics card, especially if the price was reduced just right. Say around $500 or so.
Of course, if you want to play devil’s advocate, you could also argue that by not using the “Ultra” designation for today’s GeForce 7800 GTX 512MB, NVIDIA could have even more in the works for the 7800 series. We’ll just have to wait and see how everything plays out.
In any case, the GeForce 7800 GTX 512MB is here today, and for the time being at least, it’s the new king of the hill in desktop graphics performance. We can’t wait to see what NVIDIA’s board partners do with this board. We have a feeling it could be a race among all of them to hit even higher clock speeds than what we’ve presented to you today. Sounds like fun!