Computerbase GTX480 review

[Cookie Monster]

Link

This review is pretty much a GTX480 vs HD5870 in a slew of DX11 game benchmarks, showing performance of ingame tessellation and other effects like dof. Also includes DX11 SDK performance numbers and PhysX numbers.

From there performance numbers, DX9 and DX10 games dont benefit much from the GTX480. But in DX11, the GTX480 is 47% faster than the competition.

 

[Lonyo]

About goddamn time.
This is what sites should have done from the start.
It's all very well saying that the GTX480 wins at tessellation and doing full feature tests, but if it wins at tessellation but sucks at everything else, that's bad, if it wins at tessellation and wins at everything else, that makes it an even better future product. But nothing I've seen until now has done feature by feature testing.


Now to read ->
After page 1 I have decided that the website has the greatest way of displaying scores ever, with the mouseover relative percentage scores depending on which card you pick.

 

[Attic]

47% sounds like the wrong number IMO (that will likely stand out to most people who have kept up with gpu info), doesn't sound right at all in comparison to numerous other review sites who have benched DX11 titles like Stalker CoP, Metro 2033 and Dirt 2 and BFBC2.

With the emphasis being put on tessalation evaluations it's important to have some perspective. HardOCP's Metro 2033 gameplay evaluation showed a 15% hit for the 5870 when tessalation was enabled, and a 7.5% hit for the GTX480 when tessalation was enabled. This would be an advantage of 4.5fps with tessalation enabled for every 60fps with it off.

 

[MentalIlness]

I have been wanting to upgrade my rig as well. GPU and CPU. I just upgraded my PSU to a 850watt. And I really don't want to have to upgrade it again to get a Fermi. Damn, that power consumption is too high.

I guess I'll go with ATI this time. Had NV the last three years. Just hope I have good luck.

 

[Lonyo]

Yay for frame rate graphs. All three cards got playable average frame rates (all over 40fs average).
If only every site did their testing like this.

Also it was very interesting to see how low the HD5870 power use was in the Unigine benchmark. 50w lower as a system load than in any other game. Is that because a lot of the GPU isn't doing any work, or because the frame rate is low so the CPU has less work, or both?

Also most of the DX11 tests (synthetics towards the end) also showed the GTX480 having a lead, so it's not just tessellation which favours it (which is nice to know, since I hadn't seen anything which suggests one way or the other).

 

[Cookie Monster]

The difference between the tessellation unit in Cypress vs Fermi is that the former uses a fixed function unit and the one in Fermi is integrated within its GPC. So unlike Fermi, the bottleneck when using tessellation lies within the tessellation unit since the rest of the chip has to wait for it to finish w/e its doing.

@MentalIlness
If you have a 850W PSU, whats stopping to from getting the GTX480? I mean people seem to forget that a PSU's efficency rating drops as you have lighter loads. Plus what is the point in buying such a high rating PSU, when most aren't even breaking the 350W barrier during full load.

 

[VirtualLarry]

The difference between the tessellation unit in Cypress vs Fermi is that the former uses a fixed function unit and the one in Fermi is integrated within its GPC. So unlike Fermi, the bottleneck when using tessellation lies within the tessellation unit since the rest of the chip has to wait for it to finish w/e its doing.

I think that you have that exactly backwards. Fermi has dedicated tessellation hardware, and the 5xxx series has to do tessellation on the shaders.

Edit: Sorry, seems I was wrong. But I do seem to recall reading that one of the cards did tessellation in the shaders. Don't know where I read it though.

 

[Keysplayr]

I think that you have that exactly backwards. Fermi has dedicated tessellation hardware, and the 5xxx series has to do tessellation on the shaders.

AMD GPUs and Nvidia GPUs do not use their shaders for tesselation. AMD GPUs have a dedicated single tesselation unit. Nvidia GPUs have 1 polymorph engine per SM. 480 has 15 SMs, and therefore 15 polymorph engines each with a tesselator.

 

[RussianSensation]

GTX480 walks all over 5870 in DX11 with AA on + tessallation. Looks like SI will have to improve on Tessallation front big time.

 

[MentalIlness]

I used to have a OCZ 450 that was RMA'd. But that was a few years ago.

I now have this one .... http://www.ocztechnology.com/produc...rmance/ocz-gamexstream-850w-power-supply.html

You think it would support Fermi ? It has (4) 12 volt rails....20 amp each. You think it would be enough ?

Internal 120mm fan
Technical specifications
150 x 140 x 86mm
ATX12V v2.2 and EPS12V
OCZ ConnectAll™ universal connectors:
1 x 20+4Pin ATX
1 x 4-pin/8-pin CPU
(supports double CPUs/supplies stable voltage)
2 x PCI-E
6 x 4-pin peripheral
2 x 4-pin floppy
6 x S-ATA
High efficiency
80% @ 115V (Typical load)
83% @ 230V (Typical load)
Overvoltage/Overcurrent/Short-Circuit protection
Active PFC
MTBF:100,000 hours
100~240Vac 10-5A 50/60Hz
+3.3V(36A), +5V(30A), +12V1(18A), +12V2(18A), +12V3(18A), +12V4(18A) Maximum ratings are shown.

EDIT: It has (4) 20 AMP rails. Not 18. My mistake. My PSU is the revised version.

 

[happy medium]

@MentalIlness
If you have a 850W PSU, whats stopping to from getting the GTX480? I mean people seem to forget that a PSU's efficency rating drops as you have lighter loads. Plus what is the point in buying such a high rating PSU, when most aren't even breaking the 350W barrier during full load.



I used to have a OCZ 450 that was RMA'd. But that was a few years ago.

I now have this one .... http://www.ocztechnology.com/produc...rmance/ocz-gamexstream-850w-power-supply.html

You think it would support Fermi ? It has (4) 12 volt rails....20 amp each. You think it would be enough ?

yea use 2 rails for the gtx 480 and you will be fine.

 

[MentalIlness]

Well, If I were to decide on Fermi....I should probably just got with the GTX 470. Although my motherboard right now is the GIGABYTE UD3P and supports Crossfire only.

Would something like HD 5770 Crossfire work as well on this PSU ? And would it be comparable to a GTX 470 or GTX 480 ? Cannot really find any reviews comparing the two setup's.

Makes my head spin thinking about all this stuff.

I apologize for asking these questions in another persons thread.

 

[happy medium]

Well, If I were to decide on Fermi....I should probably just got with the GTX 470. Although my motherboard right now is the GIGABYTE UD3P and supports Crossfire only.

Would something like HD 5770 Crossfire work as well on this PSU ? And would it be comparable to a GTX 470 or GTX 480 ? Cannot really find any reviews comparing the two setup's.

Makes my head spin thinking about all this stuff.

I apologize for asking these questions in another persons thread.

5770 in crossfire = a 5870.

Best bang for your buck is 5770's in crossfire .
5870 is right smack in the middle of the gtx 470 and 480.

Your psu is fine for any card or 2 5770's

 

[Attic]

AMD GPUs and Nvidia GPUs do not use their shaders for tesselation. AMD GPUs have a dedicated single tesselation unit. Nvidia GPUs have 1 polymorph engine per SM. 480 has 15 SMs, and therefore 15 polymorph engines each with a tesselator.

Yes this. Which is a defining difference between the two and attributes to the large performance difference in favor of the GTX480 in synthetic tessalation tests and benchmarks.

My question is what happens when the 15 polymorph engines are busy during high GPU load doing their other tasks for rendering a game? Does this drastically affect or not the tessalation performance and is this why we see synthetic tessalation benchmarks heavily favoring the 480 in extreme tessalation scenarios, whereas in real world gameplay the tessalation performance difference is not as absolute? Synthetic benchmarks show even a 470 generating more frames than a 5970 in some cases.

 

[Keysplayr]

Yes this. Which is a defining difference between the two and attributes to the large performance difference in favor of the GTX480 in synthetic tessalation tests and benchmarks.

My question is what happens when the 15 polymorph engines are busy during high GPU load doing their other tasks for rendering a game? Does this drastically affect or not the tessalation performance and is this why we see synthetic tessalation benchmarks heavily favoring the 480 in extreme tessalation scenarios, whereas in real world gameplay the tessalation performance difference is not as absolute? Synthetic benchmarks show even a 470 generating more frames than a 5970 in some cases.

What it "could" mean is, as time goes on, more devs "might" make more usage, or should I say heavier usage of tesselation. Why wouldn't they? In the Heaven benchmarks, there is a quite shocking difference in the image quality going from non tesselated to tesselated. This is an extreme example, but does provide what "could" come down the pike. I can guaranty you that ATI's next series will have improved tesselation performance, however which way they decide to accomplish it. Yes, the Unigine test is synthetic, but it's still a game "engine" that most likely will be used in some games coming up. Don't forget, all that extra detail generated still has to be rendered. And I'm sure we both know that it would be ridiculous to think that for example a 8600GT with the most incredible tesselator in the world, second to none, could run that Unigine bench (this is hypothetical, I'm not saying 8600GT has a tesselator, just used as an example.) You need the horsepower to render all the extra detail as the geometry is enormously increased.

Synthetic, yes. But "could" be a sign of things coming fairly soon, especially if devs choose to use the Unigine Heaven engine in their games.

 

[imaheadcase]

That site is so much easier to view the graphs than anandtech. Can't quite put finger on why, but its just seems easier on the eyes. That was some pretty through testing as well they did.

 

[Attic]

snip

Nice, ya I would love to see the kind of detail in the Heaven and Stone Giant demos get inside of some games real soon. The detail and immersion factor in those demos due to tessalation is incredible. I'd hope future games offer a user attributable level of tessalation as well. I'm not sold on whether or not a dedicated tessalation like AMD in the 5xxx unit or nVidia's implementatin in the 400 series is going to win out in the future.

I'm sure the 6xxx series will improve AMD's tessalation performance, but is it going to be something that offers better performance in games and still loses out in synthetic benchmarks to the 400 series due to the fact that synthetic tessalation benchmarks can take advantage of nvidia's 15 PM engines due to those 15 engines not needing to process the rest of the game world, wheras on ATI the dedicated tessalation unit is free from having to process everything else rendering a 3d scene. Or am I looking at the hardware the wrong way?, i'm assuming as the Fermi GPU becomes more loaded that the tessalation performance will drop as long as the majority of the load isn't doing tessalation.

 

[Attic]

That site is so much easier to view the graphs than anandtech. Can't quite put finger on why, but its just seems easier on the eyes. That was some pretty through testing as well they did.

werd, the charts in the review are just well done.

 

[ViRGE]

What it "could" mean is, as time goes on, more devs "might" make more usage, or should I say heavier usage of tesselation. Why wouldn't they? In the Heaven benchmarks, there is a quite shocking difference in the image quality going from non tesselated to tesselated. This is an extreme example, but does provide what "could" come down the pike. I can guaranty you that ATI's next series will have improved tesselation performance, however which way they decide to accomplish it. Yes, the Unigine test is synthetic, but it's still a game "engine" that most likely will be used in some games coming up. Don't forget, all that extra detail generated still has to be rendered. And I'm sure we both know that it would be ridiculous to think that for example a 8600GT with the most incredible tesselator in the world, second to none, could run that Unigine bench (this is hypothetical, I'm not saying 8600GT has a tesselator, just used as an example.) You need the horsepower to render all the extra detail as the geometry is enormously increased.

Synthetic, yes. But "could" be a sign of things coming fairly soon, especially if devs choose to use the Unigine Heaven engine in their games.

The biggest factor is really going to be how lazy devs are. Most development houses already do their character modeling at very high detail levels for deriving bump/parallax maps and such, so high quality displacement maps are just one more thing you can easily derive. The penalty is that all of those displacement maps will eat up more disk space and more memory, so devs may not want to do that if they're on a space budget.

I think the AT article put this best: tessellation is shockingly easy to take advantage of as long as NVIDIA can overcome developers' apathy. In fact the only real downside is that only character models are like this; environment models aren't usually built at great detail, so devs would have to put some real work in to them if they wanted to use GF100's greater tessellation abilities on the environment.