NVIDIA (ZOTAC) GeForce 9800 GX2 - the champ is back!

[HEXUS]

Welcome to the fastest graphics card in the world.

Read more.

[-iceblade^]

finally, a proper review i can read, not this pre release rubbish...

[neonplanet40]

Can i ask why Crysis wasn't in the review? I would have thought that it would be the first game people would want to see results in.

Its a good card but you will pay for it. AMD currenty has an advantage at that point.

[pollaxe]

"Sorry, but the URL you followed to get to this item is invalid."

?

[Tarinder]

Originally Posted by neonplanet40

Can i ask why Crysis wasn't in the review? I would have thought that it would be the first game people would want to see results in.

Its a good card but you will pay for it. AMD currenty has an advantage at that point.

Yup, good question. Let me try to answer it.

Our Crysis testing, under Vista, has shown that whilst it's possible to generate numbers for any given resolution/setting, getting them to a reasonable standard deviation, on three runs, proves to be incredibly difficult.

We need three runs to eliminate any freaky results, but it happens all too often with this game.

For example, three runs might average 21.15, 18.25, 23.15fps, respectively. Run them again and the figures might be 20.12, 16.58, 23.15, for example: there's no consistency.

We've therefore dropped it until we properly explore custom-recorded timedemos, which is what we're doing right now.

[Aez]

hexus graphics card reviews are nice but you run 3 benchmarks, it would be useful to do a lot more.

[neonplanet40]

Originally Posted by Tarinder

Yup, good question. Let me try to answer it.

Our Crysis testing, under Vista, has shown that whilst it's possible to generate numbers for any given resolution/setting, getting them to a reasonable standard deviation, on three runs, proves to be incredibly difficult.

We need three runs to eliminate any freaky results, but it happens all too often with this game.

For example, three runs might average 21.15, 18.25, 23.15fps, respectively. Run them again and the figures might be 20.12, 16.58, 23.15, for example: there's no consistency.

We've therefore dropped it until we properly explore custom-recorded timedemos, which is what we're doing right now.

I see. Thanks for the clarifying Now wheres me wallet

[Hicks12]

anyone think nvidia failed? at practicability(is that a word?) who will actually spend £400 on a gfx card(ultra anyone?) when you can buy say.... 4 9600's or 3 8800GT's or 2 8800GTS for the same price. I dont see the reason to even consider buying it, most of the tests show it can out perform but when most cards are getting say 100fps and that gets 150fps, you wont notice it and there for a waste of £400, the only reason would too make it more future proof but that imo means nothing as new cards come out every 6months(at least) so after the few releases they will be crappy for the price.

[pow1983]

This card is jaw droppingly good.
ATI would have to produce a 3870x4 just to get a little closer to these results.

At least you dont need to buy such a massive wattage Power Supply or even a SLI motherboard to run this BFG! (doom terms).

I hope they can get onto doing a smaller dye like 55nm or even 45nm to help bring the cooling situation under control.

[Crazytom]

Wow. What else is there to say?
For all those people who thought that slapping two GPUs together shouldnt warrant a generation increase (8xxx to 9xxx) this is just a slap in the face. The performance increase shows the completely different league this really is in.

[Will101]

Tarinder - I know this is a new top end product review but it seems decidedly one sided.

You use OCed versions of NVidias comparable current SKUs but reference ATI cards. (To be fair you do point this out in your review)

In all reviews I've seen of the 38** they do well with no AA up to the highest resolutions but are hammered when AA is enabled. I'm don't remember the definitive argument re the need for AA as the res increases to v.high levels but I suspect it is less necessary?

As a pure power tour de force with all setting to the max you've shown the 9800 GX2 to undoubtably shine.

As an intelligent comparison between different manufs top graphics offerings there seems overwhelmingly to me to be a finger on one side of the scales...

Of course you have every right to exercise your right to reply... ;-)

[JARSmith]

I'm James, the Manager of Performance Analysis at HEXUS.

I'd just like to elaborate on these issues, to help better explain the current problems with us using Crysis for benchmarking.

We first investigated using Crysis back in October last year (at the time the 8800 GT was being launched), and ran several cards on the Demo version then available.

After much frustration - and having to drop the maximum resolution we benchmark at, down to 1680x1050 - we managed to get a set of results that were reasonably consistent (<3% variance on two out of the three cards).

However, no matter how many times we ran Crysis, it wasn't possible to get usable results at 1680x1050 from the GeForce 8800GTS 320MB.

Subsequent to this, in our testing of the Radeon HD 3870 (still using the demo at this point), it was discovered not only that there was a high degree of variance between runs, but also complete sets of runs would produce wildly variable sets of results, than from before.

E.g. a set of 3 results from one day of 18.12, 18.25, 21.2, might produce a set of 3 results of 21.15, 23.15, 21.5 the next day, with nothing been changed on the system - the exact same game settings, OS build, and hardware!

Initially I put this down to problems with the Demo, however once the full game was released the problems persisted.

Therefore these issues, in addition to the games inability to properly scale on multi-GPU platforms - mainly on AMD GPU based systems, and eventually much later on fixed in the 1.1 patch - led us to abandon the use of Crysis in our testing.

The hopes of being able to use Crysis were once again dashed, when even upon the release of the 1.1 patch - and also with the assistance of the excellent Crysis Benchmark Tool: Crymod Modding Portal | Finished Assets | Benchmarking Tool (released) - the inconsistencies remained, albeit to a lesser extent as it now only seemed to affect some cards, rather than all of them.

I doubt I need to tell anyone on here that these kind of variances would result in us being forced to make inaccurate, and misleading, judgements of a card’s performance, and this is something that neither myself or any other HEXUS member of staff would be willing to accept.

Still I've noticed the recent release of yet another Crysis patch, which I hope this time, in conjunction with making our own time-demo to benchmark, will result in consistent and therefore trustworthy and relevant comparisons.

Aez, to answer your query:

This is something we've always been looking at doing since the move to Vista, so expect to see a couple more games being benchmarked soon - once I've found a couple more that meet our exacting standards.

[Tarinder]

Originally Posted by Will101

Tarinder - I know this is a new top end product review but it seems decidedly one sided.

You use OCed versions of NVidias comparable current SKUs but reference ATI cards. (To be fair you do point this out in your review)

In all reviews I've seen of the 38** they do well with no AA up to the highest resolutions but are hammered when AA is enabled. I'm don't remember the definitive argument re the need for AA as the res increases to v.high levels but I suspect it is less necessary?

As a pure power tour de force with all setting to the max you've shown the 9800 GX2 to undoubtably shine.

As an intelligent comparison between different manufs top graphics offerings there seems overwhelmingly to me to be a finger on one side of the scales...

Of course you have every right to exercise your right to reply... ;-)

Will101,

All intelligent points.

Anyone seriously contemplating spending £300+ on a video card wants the absolute fastest thing possible, and the point of this card was to be the fastest around. It achieves that aim with consummate ease, as evinced by our benchmark results. It doesn't win by five per cent or 10 per cent. The lead is often 50 per cent, especially at the highest I.Q. settings. Superlative I.Q. is where these cards should excel at, because why would you pay £150+ and still expect to see jaggies at 1,920x1,200?

I personally run a Dell 3007WFP-HC and want to play games at the panel's native resolution of 2,560x1,600. The GeForce 9800 GX2 is the closest that any single card comes to letting me do that with all the eye-candy turned up to maximum.

The card isn't cheap, but it doesn't need to be when it has a significant performance lead over the next-best card. Our HEXUS.bang4buck graph shows that paying the extra, over and above a Radeon HD 3870 X2, brings more-than linear performance gains.

Yes, you're right, overclocked versions of the GeForce 8800 GT and GT 512 have been used, but they're included because we've reviewed them in the recent past. Indeed, practically every NVIDIA partner has released a pre-overclocked SKU based on the G92 core, compared with just a few from the ATI camp.

Does that make the ATI Radeon HD 3870 X2 a bad card? The answer is no, not at all. Quoting from my own Radeon HD 3870 X2 review:

"Is it worth the £279 or so? The answer is a tentative yes, because it achieves top-dog status, in our set of benchmarks at least, from brute power rather than the elegant and well funded developer-relation support that NVIDIA currently enjoys."

The bottom line is that NVIDIA wanted to grab back the mantle of the world's fastest graphics card. The GeForce 9800 GX2 does this, easily.

I'd love the card to be priced at £199, sure, but NVIDIA's pricing, whilst exorbitantly expensive for most, makes sense for the few people that want, and more to the point, can afford, the fastest.

Think of it along the same lines as Intel's Core 2 Quad CPUs vs. AMD's Phenoms. Intel has a performance lead at the very high-end and knows it. The QX9770 is priced at £999 because AMD has nothing that can compete in that space. As much as we malign Intel's pricing, there's little left to argue with respect to performance.

An aside, I believe that ATI has better cards in the £90-£130 range, encompassing the Radeon HD 3850 and 3870 SKUs. I also believe that the best card for around £270 is the Radeon HD 3870 X2. The best card at £399, though, is clearly the GeForce 9800 GX2.

[Anastie]

Good review and even better answers from those at Hexus, thanks for that. Always appreciate it when those involved keep in touch.
Cheers.

[-iceblade^]

very good that we can actually speak to the reviewers. top notch, this site... Also good, solid arguments put back by the reviewers, though if possible, i would consider trying to remove the discreptancy by using standard reference cards when performing a review...

[dhyta23]

Nice to see information about this.