GeForce 8800Ultra Mini Review: G80’s Last Stand?

NVIDIA’s G80 has proven itself time and time again over the last 5 months. With AMD/ATI on the verge of releasing its first real competitor, NVIDIA has pulled out all the stops on what it has perhaps nostalgically called the GeForce 8800Ultra.

Every now and then the ferocious competition in the hardware world produces a product that is completely groundbreaking. Just as often, however, it produces a product that leaves the vast majority scratching their heads. Companies think up new and imaginative SKUs to existing products that offer a little something more than their predecessors. Intel Corp used to be the biggest perpetrator of this mind-boggling exercise. One cannot count the number of Pentium-D SKU’s that came to the market since their introduction up until the launch of the Core 2 family.

Intel is certainly not the only one that does it though. Video card companies do it quite often as well. Although they are certainly getting a bit better than they have been in the past, it is likely that a search of any particular video card will yield three or four results of differing specs and perhaps different names entirely. Granted, these derivatives are not all a result of NVIDIA or ATI’s own prerogative, graphics board partners tend to tweak reference cards to varying degrees and market them with their own monikers as well. Today, however, we will be looking at the most recent SKU variation from NVIDIA itself, the mighty GeForce 8800Ultra.

With what is expected to be a huge launch in the graphics market from AMD/ATI just around the corner, it seems NVIDIA was just itching to play the trump card and beat the R600 to shelves. As far as we know, this 8800Ultra is the end-all graphics board from NVIDIA until the G90-based cards are brought to market. We really don’t know exactly why or even if NVIDIA was determined to put out this “G80’s Last Stand”, considering their current G80 product line has been enjoying what is essentially free-reign over the high-end graphics market for about 5 months now. In any case, the 8800Ultra is here, and we intend to show you just what it can do.

The Card
Glancing at the pictures, the first thing you will notice about the 8800Ultra is the updated cooler design. As opposed to the rather non-impressive dual-slot cooler on the current 8800GTX and 8800GTS series cards, NVIDIA has chosen to go with a more aggressive and extended cooler on the 8800Ultra. The black part you see is some sort of plastic polymer, not the matte black metal we were hoping for. Besides the cooler, however, the card appears to be physically identical the 8800GTX. In fact, we have received reports from NVIDIA officials that the biggest difference between the 8800Ultra and the 8800GTX is software based.

There really is not much else to say about this card, as everything that needs to be said about the features and specifications has already been mention in our initial review of the 8800GTX.
{mospagebreak heading=Introduction&title=Test Setup and Methods}

Test Setup and Methods
As can probably be discerned from the title of this article, this is not going to be a full review of the 8800Ultra. We did not receive a review sample to get the full hands-on experience that we are accustomed to, and our amount of time with the card was severely limited by both geographic and political factors. For that reason, we have really only run conclusive tests in three applications: Futuremark’s 3DMark06, Valve’s Half-Life 2: Episode 1, and Monolith’s F.E.A.R. Nevertheless, we feel that these applications provide a surprisingly good cross-section of the performance that is offered by this new card. Again, it was out of the goodness of a friend’s heart that we were able to get our hands on these cards, and our time with the cards was limited to say the least.

One thing we did get to do, however, was put two of these crazy new cards in SLI mode – something you probably won’t be seeing anywhere else. If the 8800Ultra is truly a more powerful version of the 8800GTX, then by all accounts, two 8800Ultra’s in SLI configuration constitute the most powerful graphics subsystem in existence – at least for now. Here is a look at the platform we used for testing these cards:

Hardware Configuration

Power Supply: Tagan TU96 1100W
Motherboard: ASUS P5N32-E SLI nForce 680i
Processor: Intel Core 2 Extreme QX6800 (Varying MHz, 2933MHz Stock)
Hard Drive: Western Digital 800JD 7200RPM, 80GB with 8MB Cache Buffer
Video: XFX NVIDIA GeForce 8800GTX eXtreme 600MHz core clock 2x NVIDIA GeForce 8800Ultra (612/1080)
Memory: 2048MB (2×1024MB) Corsair XMS2 PC2 8500 (1066MHz @ 5-5-5-15 2T)
Cooling: Vigor Monsoon II
Monitor: Acer X241WSD (24", 1920×1200, 6ms)

Software Configuration

Motherboard BIOS: 1002
Operating System: Windows XP Professional with Service Pack 2
Video Driver: NVIDIA ForceWare Version 158.19

So then, the four configurations we used during testing are:

8800GTX: Above setup with a single XFX NVIDIA GeForce 8800GTX eXtreme, all else stock
8800Ultra: Above setup with a single reference NVIDIA GeForce 8800Ultra, all else stock
8800Ultra SLI: Above setup with 2x reference NVIDIA GeForce 8800Ultra’s in SLI configuration, all else stock
8800Ultra SLI OC: Above setup with 2x reference NVIDIA GeForce 8800Ultra’s in SLI configuration, processor overclocked to 3736MHz (14×266).

Our testing consisted of our normal methods; testing each game at each resolution three times. The provided score is the average of those three tests. In several cases, we have actually tested the resolutions more than three times because we thought the numbers were weird for one reason or another. Those scores are reported as the average of all tests, and if anything are marginally more accurate than the other results.

After running tests with a single 8800Ultra, we added another to get scores for SLI. After that, we thought it would be a prime opportunity to explore CPU limitation with the current popular benchmarking applications. When you have a graphics subsystem as capable as two 8800Ultra’s, you can truly begin to see the effects of CPU limitation on games that have been deemed as mostly GPU limited in the past.

It is very very very important to note that the driver used in all tests on the 8800Ultra cards was simply a modified version of the currently available ForceWare version 158.19. NVIDIA will have a special production driver available for the 8800Ultra on May 2nd (today!), and will be following that release with a second in the next week.
{mospagebreak title=Tests: FEAR, HL2 }

Tests
F.E.A.R

The first test we ran on the 8800Ultra’s was F.E.A.R. Since we started using F.E.A.R in our benchmarking suite, it has consistently proven to be a very good indicator of graphics performance at the high-end, and CPU performance at the low-end. This is due mostly to the tremendous amount of graphics and computer options available in the performance settings area of the game. In addition, the built-in cinematic sequence provides a very good sampling of the effects that are prevalent in the single player campaign.

Here you can see at very high resolutions such as 1920×1200 that there is a very strong GPU limitation. The most interesting thing to note about this test is that the 8800Ultra and the 8800GTX are neck and neck. This is likely to be partially the result of the 8800GTX we used being overclocked to 600MHz on the core, as opposed to the stock 575MHz. You will also note that the relative difference between single cards and SLI decreases significantly as the resolution drops down to 1600×1200 and then 1280×1024. This is evidence that F.E.A.R is quite CPU limited at these resolutions with a graphics subsystem capable of delivering never-before-seen performance. Note the 26 FPS increase that was achieved by overclocking the processor at 1280×1024 resolution, and how this increase compares to the difference seen at a higher resolution of 1920×1200.

Half-Life 2: Episode 1

When it really comes down to it, we are just part of the GotFrag.com network. As such, using a benchmark based on Valve’s Source engine is not so much a choice as it is a necessity. Chances are that a good amount of our readers are competitive Counter-Strike: Source players; reading the questions that arise on our forums every so often will support this notion. However, even if we weren’t basically required to run an application based on the Source engine, we still would. Despite being something like three years old already, games based on the Source engine still sport some of the most stunning visuals and outstanding effects that we have ever seen. Like F.E.A.R, Half-Life 2 allows for quite a lot of customization in terms of visual quality. As such, we expect our results to reflect the sensitive nature of the test, as well as be GPU limited on the high-end.

As expected, we see that Half-Life 2: Episode 1 is GPU limited at higher resolutions. However, it is evident from these results that the extent of which HL2 is GPU limited is not nearly as defined as with F.E.A.R. Notice that the 8800Ultra SLI graphics setup on the overclocked system still scores significantly higher than the non-overclocked setup. This would indicate that even at a very high resolution like 1920×1440, Half-Life 2 exhibits CPU limitation. This CPU limitation becomes more and more defined as resolution drops. You can see in the chart above that the difference between the overclocked system and the non-overclocked system becomes more distinct as the resolution is decreased. Furthermore, notice that the game appears to be almost entirely CPU limited at 1280×1024 resolution with all graphics cards being tested - a single 8800Ultra is right on par (and even beats, though we wouldn’t really call this a victory so much as a consequence of random variance) with 2 8800Ultra’s in SLI.
{mospagebreak title=3DMark06, Conclusion}

3DMark06

The funny thing about Futuremark’s 3DMark06 graphics performance benchmark is that it actually is quite a terrible test. When it was first released, graphics cards struggled to perform well in the SM3.0 and HDR test, and CPUs offered little help in this task. However, with the launch of graphics cards like the 8800 series and even the X1950XTX from ATI, it has become quite clear that the rigors of 3DMark have been conquered quite handily by today’s video card technology. Evidence of this is found in the HEAVY CPU limitation that is observed on systems that make use of 8800 series graphics cards. A quick visit to Futuremark’s ORB database reveals that the only extraneously high results are generated on computers with extremely heavily overclocked quad-core processors from Intel. Our own tests exhibit this as well.

The ONLY thing that allows 3DMark06 to be a useful benchmark this far into the game is the fact that it is so widely used. There is no other benchmarking standard so commonly referenced, and there is no other score that is hailed with such reverence. Whether the test is good or not is irrelevant thanks to the countless individuals who consider it the holy grail of benchmarks and need to see 3DMark06 results in order to be satisfied with a video card review, no matter how not in-depth said review is. You can see from our results that the most significant performance jump occurs between the overclocked and non-overclocked versions of the 8800Ultra SLI setup. Adding an entire new video card to the system only results in a performance increase of 2242 3DMarks, whereas overclocking the processor by about 800MHz boosts the score more than 2800 3DMarks. To be honest, seeing a score of over 18,000 in 3DMark06 on a system you are working on has some sort of invigorating effect, but the score really does not mean that much. Of course there is still something to be said about increasing graphics horsepower in 3DMark06, otherwise there would be no difference between a single card and SLI, however, the test is just becoming far too CPU limited to be a viable source for determining graphics card performance for much longer.

Final Thoughts and Conclusion
Despite the updated cooler, the 8800Ultra runs pretty hot. Granted, the G80 is quite an efficient little chip, so the maximum temperature we reached was still well under what we would consider unacceptable, but temperatures were certainly not much better than those achieved with the standard 8800GTX cooler. Heat is probably not a big issue for the people who will actually consider buying this card, however. Rumors slate the MSRP of the card somewhere in the neighborhood of $829, which is nearly double that of the similarly performing 8800GTX. That said, the 8800Ultra certainly does outperform the 8800GTX at stock speeds. However, and 8800GTX overclocked to 630MHz on the core will most assuredly take back the crown. But then you have an overclocked 8800GTX against a stock 8800Ultra. So how does the 8800Ultra overclock anyway? We’re so glad you asked.

After testing the overclockability of this card during the brief period of time we had with it, it left us with the impression that the G80 cores used on the 8800Ultras are bin 0 chips. This means that they basically are the crème de la crème of G80 cores – they have more headroom than the rest. We were able to overclock our Ultras to 700MHz on the core with ease, and the memory of course pushed to its rated 2200MHz without a second thought. However, tremendous overclocking ability must be something any prospective buyer would expect after paying $829 for a card that truly does not seem to offer anything significant over the 8800GTX.

We have heard from NVIDIA that this card has an optimized BIOS that allows it to surpass anything the 8800GTX is capable of. We also know that 30-40% performance increases as a result of a driver update are not unheard of. With any luck, we might be able to test these cards with the brand new ForceWare driver that is due out very soon. If performance jumps significantly and the BIOS optimizations prove to be worthwhile, who knows, maybe this card could be worth the money.

Popularity: 5% [?]

Support FPSLabs! Share and Enjoy: