GTX480 Vs. HD5870 Crysis benchmark

[Tempered81]

http://www.youtube.com/watch?v=SlEA8E9NomM&fmt=22


Trubritar video review of the two cards on a 4.4ghz 6-core gulftown

 

[Madcatatlas]

Interesting watch. What do you think this means Tempered81?

 

[RussianSensation]

no AA
GTX 480 = 26.4 min / 35.4 avg
5870 = 23.3 min / 36.8 avg

8AA
GTX 480 = 23.8 min / 29.6 avg
5870 = 18.9 min / 32.4 avg

Better mins on the Fermi card.

 

[zagood]

dammit russian...I was just grabbing those.

With max...and FYI res is 19x12.

no AA
GTX 480 = 26.4 min / 35.4 avg / 40.7 max
5870 = 23.3 min / 36.8 avg / 43.02 max

8AA
GTX 480 = 23.8 min / 29.6 avg / 33.43 max
5870 = 18.9 min / 32.4 avg / 37.76 max

 

[Tempered81]

Interesting watch. What do you think this means Tempered81?

Well it means the cards are about equal in crysis, 5870 is 1 fps faster. The 480 is giving better minimum fps.

Crysis benchmark is a good test to show the potential of a gfx card, since all cards scale in this game up to the fastest: 5970. Plus the game is so GPU limited that it uses every resource of the gfx card. (most of the time crysis produces higher temps compared to other games). It's a nice way to test a card at full load.

Something that puzzles me is how great gf100 is at farcry2, dirt2, & hawx, yet it can trail in some games. It's generally 30% faster than a 5870 in those titles. I know ati takes a hit with AF in hawx, and ati sucks with AA on in Farcry2, but we have to give credit to Nvidia drivers for the level of performance the pull on those 3 titles with the new cards.

It's just strange that you don't see that kinda lead across the board, and in some cases, the 480 is in fact losing to the 5870. One more thing about the crysis bench is that it's still one of the most demanding pc titles. You get others like Stalker with Godrays & Metro with dx11, dof & tess that are catching up to the level of suckass performance crysis offers.

 

[RussianSensation]

Tempered81, I hear you on the suckass performance of Crysis. Wait until Crysis 2 hehe. However, I still don't think Crysis is representative of full GF100's performance since it doesn't use tessellation. It is strange though that the NV card is not outperforming 5870 here. Maybe the texture fillrate disavantage is coming into play? (68,000 MTexels/sec for 5870 vs. 42,000 MTexels/sec for 480).

 

[apoppin]

Wait until you overclock both cards
-then the GTX 480 (825/1100) pulls away from the HD 5870 (975/1300)

i have found the same thing in all of my benching with 15 games from 16x10 to 25x16 -
- Overclocking each card results in not a single win for HD 5870 - just one tie

 

[Tempered81]

This guy has the most awesome hardware. Just watched three of his reviews - the GTX480 vs. 5870 crysis benchmark, the 5970 vs. 5870 quadfire review, and the GTX480 quad sli unboxing...

Retail: i7975x, i7980x, 5970 Quadfire, XFX XXX 5870 Quadfire, EVGA GTX480 Quad SLI, EVGA Classified, ASUS p6t6 ws SUPERCOMPUTER

.... That's like > $10,000 USD of hardware alone.

 

[Tempered81]

Yah Mark, lookin forward to your review. When is it coming out?

 

[Nemesis 1]

ATI winner.

about 1.18 on that video look at face of rock cliff. ATI seems to have better detail . looking at left side of both screens the whiteish part much better detail. your right.

 

[apoppin]

Yah Mark, lookin forward to your review. When is it coming out?

Thanks. i am writing the conclusion right now - or i should say the conclusion is dictating itself to me - i need to add a couple of screen shots, proof it and polish it; probably another 4 hours; tonight for sure.

After my ECS motherboard review (tomorrow??), expect GTX 480 vs. HD 5870 at 8xMSAA and i will add Just Cause 2 to my benching suite.

 

[solofly]

I personally don't care about minimum or maximum frames, average is what counts to me...

 

[tviceman]

To me minimums are more important. But regardless, I think Crysis shows a weakness in Fermi - either that or Nvidia has some serious driver optimizations to do. But more than likely Fermi's texture management looks suspect.

 

[Lonyo]

Wait until you overclock both cards
-then the GTX 480 (825/1100) pulls away from the HD 5870 (975/1300)

i have found the same thing in all of my benching with 15 games from 16x10 to 25x16 -
- Overclocking each card results in not a single win for HD 5870 - just one tie

Is that with the stock cooler on both? And how do heat/noise scale?

 

[BenSkywalker]

Maybe the texture fillrate disavantage is coming into play?

First time I have seen anyone on this forum notice that, someone pays some attention at least

Tons of foliage, tons of overdraw. High levels of overdraw+high resolution- major texel draw demands. That the 5870 doesn't obliterate the 480 in this bench is the real interesting point from an architectural standpoint. Would be nice if we could get some heavy handed moderation for at least one thread so we could have a discussion about the architectural differences at play, not sure that can happen though.

 

[Lonyo]

To me minimums are more important. But regardless, I think Crysis shows a weakness in Fermi - either that or Nvidia has some serious driver optimizations to do. But more than likely Fermi's texture management looks suspect.

Well.
It could be that old architecture thing.

If SPUs go unused, then the performance of the chip suffers due to underutilization. This design gives AMD a great deal of theoretical computational power, but it is always a challenge to fully exploit it.

ATI has lower minimums (when there is little to no parallel code).
NV has higher minimums (because it's able to work at a higher efficiency more easily)

ATI has higher maximums (because it has a higher optimal performance which Crysis sometimes lets it approach)
NV has lower maximums (peak throughput is lower)

Average is a function of this, depending on how well the code can use ATI's architecture.

 

[extra]

To me minimums are more important. But regardless, I think Crysis shows a weakness in Fermi - either that or Nvidia has some serious driver optimizations to do. But more than likely Fermi's texture management looks suspect.

Meh...I think the minimums are just having to do with the extra ram, but I could be wrong. Either way the performance is close enough between the two as to be "meh". As it seems to be in most games really...

And I dunno...It doesn't look like Fermi has any performance or driver issues or anything like that, looks like Nvidia's drivers are, as usual, pretty solid (as are ATIs now as well)...

I doubt we'll see much more performance from Fermi drivers, because they've had a lot of extra time to work on them and optimize them...The architecture looks really good and looks like it'll have long legs, the problems are solely with power and price. Seems really unappealing at the moment. Could be fantastic though if price went down $100.

 

[apoppin]

Is that with the stock cooler on both? And how do heat/noise scale?

Yes. They are both as loud as hell. The Diamond (reference) HD 5870 gets 975/1300 on 1.35V and the fan is running near max (as is the GTX 480, but i did not raise its voltage; it is hot enough at stock voltage fully overclocked) - but they never got near danger as far as thermals go. The HD 5870 is running cooler - but that is relative.

IF i was seriously considering either O/C - for myself, 24/7 - i'd also consider water cooling either card or else getting a very good air cooler. Stock cooling won't do for me to go further with either card. Remember i am overclocking a "reference" GTX 480 and a reference HD 5870 (the PCS+ won't go over 925/1300 on stock voltage; but then it is quieter than the Diamond when they are both overclocked)

i have to get back to work; i just stopped by. i will try to answer your questions in my writeup tonight.

Tons of foliage, tons of overdraw. High levels of overdraw+high resolution- major texel draw demands. That the 5870 doesn't obliterate the 480 in this bench is the real interesting point from an architectural standpoint.

Run Heaven 2.0 benchmark and there you see really heavy foliage; in the tall grass scene the FPS really drop for both cards - but more so for the Radeon.

This is purely my guess, but i will say "beta release" drivers is the reason for the slightly lower Crysis performance; however, when you O/C both cards, the GeForce pulls ahead further.

Meh...I think the minimums are just having to do with the extra ram

Doubtful, or we would see the HD 5870-2GB (6 pack) version beating the 1 GB version.

 

[ronnn]

Watching that video did suggest that the ati mins are rare and maybe the rumour of just a loading screen problem is true.

 

[AzN]

no AA
GTX 480 = 26.4 min / 35.4 avg
5870 = 23.3 min / 36.8 avg

8AA
GTX 480 = 23.8 min / 29.6 avg
5870 = 18.9 min / 32.4 avg

Better mins on the Fermi card.

It's got more bandwidth. Better minimum is expected.

 

[Qbah]

Doubtful, or we would see the HD 5870-2GB (6 pack) version beating the 1 GB version.

From the AT HD5870 E6 2GB review:

That's a 40% increase in minimum framerates between the 1GB and 2GB models (11.6 vs 16.3). And since the average on the E6 is just a tiny bit higher, the minimum pops up during the test very rarely.

So the minimum here is not a consistent occurrence and is not really a show stopper (though it is obviously always nice to have higher minimums). Both models play almost the same and the rare dip on the 2GB model is less noticeable. And the only difference is memory size.

 

[Patrick Wolf]

That video was for Crysis, not Warhead?

 

[apoppin]

From the AT HD5870 E6 2GB review:

That's a 40% increase in minimum framerates between the 1GB and 2GB models (11.6 vs 16.3). And since the average on the E6 is just a tiny bit higher, the minimum pops up during the test very rarely.

So the minimum here is not a consistent occurrence and is not really a show stopper (though it is obviously always nice to have higher minimums). Both models play almost the same and the rare dip on the 2GB model is less noticeable. And the only difference is memory size.

First of all, you are showing us Warhead; not Crysis.

Secondly, you don't see driver issues - especially with the first chart? Look at 5970 and CF.

And in the second one, four tenths of a frame rate, shows no perceptible difference; the same card will often vary by that much (little) on different days.

 

[Qbah]

First of all, you are showing us Warhead; not Crysis.

Secondly, you don't see driver issues - especially with the first chart? Look at 5970 and CF.

And in the second one, four tenths of a frame rate, shows no perceptible difference; the same card will often vary by that much (little) on different days.

Both games use the same engine. And run pretty much the same. I know I saw no difference in speed between Crysis High and Warhead Gamer on my HD4870. Hell, Warhead looked a bit worse, with texture pop-up and texture streaming visible.

And the "issues" are with CrossFire configurations only. However, it looks like every CF setup suffers - maybe due to higher CPU load in such a setup? Would surely explain it (and it's not the first time we can observe lower minimums with a dual card/GPU setup). So is it really an issue? Perhaps a downside of the technology in this particular engine.

However. since we're talking about single cards here, how does it matter? It doesn't (not to mention the possible explanation above). We have two cards that have everything but the RAM size identical. I can draw conclusions from that. I'm sure you can too. So why the silly talk?

Look at the averages. Even with the extremely low minimum on the CF setups it happens for a split second only (possibly at the same time the single cards hit their minimums). The very small difference in average for 1GB/2GB means that the minimum values are so rare and insignificant that they do not impact perceived performance one bit. Does the 2GB model get higher minimums? Yes. How often does a minimum frame happen? Once per test. So you get 30 average and once during the test the game drops to 10FPS for a split second... Big deal, I forget about it as soon as I kill the next baddie or reload my gun (note, I don't find 30 FPS acceptable - it's just an example).

But it doesn't change the fact that for this split second the 2GB card dipped to 16FPS where the 1GB card dipped to 11. Both have near identical averages so a minimum value is (if nice to know) kinda useless as it happens once... I like to know it but it sure wouldn't stop me from getting the card that gives more FPS on average. Unless there are portions of the test run that run with extremely high FPS and skew the test. Hence graphs are very very nice (hello HardOCP).

No idea what more you want to see, but this is proof enough for me Crysis benefits from more than 1GB RAM. Now, this benefit is so miniscule it's not worth a price premium IMO (as the cards score the same averages). But it's there. People like to argue about things like that and this here test shows there is some benefit. Which can have a value of useless to "I'm in no matter the cost" - it really depends on the person.

Now a card with 70 average and 40 minimum is better than a card with 90 average and 15 minimum in my eyes. Since 70/40 is perfectly playable and getting 90 average doesn't offer anything more. But the 15 minimum will annoy... And getting annoyed is the last thing you want to be when playing a relaxing game.

 

[MagickMan]

Apoppin, your 5870 is worse than I thought, that may be the worst voltage for 975MHz that I've seen.