Quad SLI 9800GX2 VS. QuadFire 3870X2 in Crysis [Very High]

[Tempered81]

looks like we will wait some mo'


http://www.anandtech.com/video/showdoc.aspx?i=3271&p=4

 

[imported_Shaq]

I wondered about this as every review site had an avg of 45 when on high settings even though the resolutions were different. I thought there was no way this game is CPU limited as it only uses 2 cores lightly. Maybe if it utilized 4 cores, as Crytek stated it would, it would perform faster. So one or more of the high settings is crippling the FPS. Maybe I will run through some benchmarks with low physics as Nvidia stated it was physics accuracy at higher settings that lowered performance.

 

[batmang]

Originally posted by: jaredpace
looks like we will wait some mo'


http://www.anandtech.com/video/showdoc.aspx?i=3271&p=4

It also looks like a repost:
http://forums.anandtech.com/me...=2170025&enterthread=y

 

[lopri]

The review was very interesting and I think Derek was wise in that he didn't say anything conclusive with regard to Crysis. I will try couple things out myself to see how things work. A couple of things from the review:

- If the game is CPU-bound at 'High' and 'Very High' setting, how can we explain the performance hit by enabling AA (or even AF)? I don't think CPU has anything to do with AA?
- 'Platform dependent' doesn't make sense to me because I don't know what exactly it means. Does it mean that a same CPU or GPU magically perform better on NV motherboard? Or does it mean this specific game (Crysis) performs better on NV motherboard? There could be differences in PCIe bus performance, memory controller performance, or even the 'LinkBoost', but generalizing it as 'platform dependent' is not correct or right, IMO. (Although NV would like to use such term to market their 'platform') If anything it could be just that Skulltrail just sucks at 3D.

 

[panfist]

Originally posted by: lopri
The review was very interesting and I think Derek was wise in that he didn't say anything conclusive with regard to Crysis. I will try couple things out myself to see how things work. A couple of things from the review:

- If the game is CPU-bound at 'High' and 'Very High' setting, how can we explain the performance hit by enabling AA? (At any resolution) I don't think CPU has anything to do with AA?
- 'Platform dependent' doesn't make sense to me because I don't know what exactly it means. Does it mean that a same CPU or GPU magically perform better on NV motherboard? Or does it mean this specific game (Crysis) performs better on NV motherboard? There could be differences in PCIe bus performance, memory controller performance, or even the 'LinkBoost', but generalizing it as 'platform dependent' is not correct or right, IMO. (Although NV would like to use such term to market their 'platform') If anything it could be just that Skulltrail just sucks at 3D.

1.) The answer is very simple...with AA disabled the game is CPU bound. With AA enabled the GPU has to work much harder, hence it is GPU bound. It's kind of like saying, "The game is CPU bound at 800x600 but when I increase the resolution to 1600x1200 I get a huge framerate hit." That's because you changed the load on the GPU.

2.) There are many complex parts that go into making a platform. First of all let's define platform: in most cases it means something like CPU + GPU + motherboard chipset + RAM, but practically speaking you're just analyzing the motherboard chipset performance because other people/websites put a lot of work into finding out how good a certain CPU or GPU is independent of the platform. So here are some chipset specific examples...

For instance, do you have x8/x8 SLI or x16/x16? As far as I know, no one has shown a major advantage to x16/x16. Here's a more complicated example. I have an Asus p5n32-e SLI plus, which technically has x16/x16 SLI, but it has x16 lanes through the northbridge and x16 through the southbridge, instead of all of the lanes going through the northbridge. Therefore the clock speed of the bus between the NB and SB will effect performance, also the card on the SB will have some more latency introduced due to the simple fact that the data has to travel through the SB first before it can reach the card itself. These are two platform specific examples for which you can easily look up the specs, but there is a lot more going on "under the hood" that is radically different depending on the chipset.

 

[lopri]

Umm.. did you read the review?

 

[lopri]

Basically the premise here is that Crysis @Very High will always be ~40FPS at any resolution with any GPU (including next or next next gen GPU), unless there are CPUs with faster single-threaded performance.

 

[Sylvanas]

Cmon Nehalem.....But really, I think Crysis is pretty useless with taking advantage of multiple cores. There was an article a while back comparing CPU performance with UT3 and it showed some impressive gains with multicore CPU's- something Cryengine 2 could take note of.

 

[BFG10K]

By changing the settings to a preset value (e.g. high to all very high) you're changing other settings that are not necessarily related graphics that could influence the CPU, settings such as physics and sound.

Crysis is a strange beast because it?s not always clear where the primary bottleneck is.

 

[lopri]

But I don't think there is a big physics/sound variable in Crysis GPU bench? According to the review, Derek is getting same FPS from 800x600 all the way up to 1920x1200. I am not sure why he hasn't tried raising AA or AF.

http://www.anandtech.com/video/showdoc.aspx?i=3271&p=4

 

[Lorne]

I also thought Darek did an excellent job, I also think using super high end equipment as a base platform also is good that we can really find the extent of new hardware limitation and where flaws in hardware or drivers are And Im sure most users out there would like to know its limitations before perchasing new or used in the future before they drop into there system just to find out its the total bottleneck.

All these test also show the game MFG that there or other MFG software scaling differently as newer hardware comesout, This alows designs for future game engine and optimization.

I personaly think the problem is in Crysis itself, Probably not as much bad programming as badly optimised, I belive things like this show where flaws in both programming and the hardware show the flaws hand in hand.

 

[imported_Shaq]

Originally posted by: lopri
But I don't think there is a big physics/sound variable in Crysis GPU bench? According to the review, Derek is getting same FPS from 800x600 all the way up to 1920x1200. I am not sure why he hasn't tried raising AA or AF.

http://www.anandtech.com/video/showdoc.aspx?i=3271&p=4

I have seen some inconsistencies in the benchmark also. When I first ran it on all high @1680 i got 51fps. I reran it several times and could only get 45. Then yesterday I used a mix of high/very high and medium shadows and got 48fps. I noticed very high object detail and very high water dropped framerate quite a bit over the high settings. I OC'd the GX2 from 600/1000 to 713/1100 and gained 1fps when all settings were very high. (36 to 37) It was said that the bottle neck was always graphics but now it seems its only half true.

Now that I reexamine that link, I wonder if he had tried 8800GTX/Ultra SLI and 8800 GTS 512 SLI if it would be the same? It could just be a problem with the GX2's.