Sign in with
Sign up | Sign in
Your question

7970 crossfire on AMD FX series an incredible waste. Just so you know.

Last response: in Graphics & Displays
Share
March 26, 2013 1:04:23 PM

Just thought I'd post a bit of info on my experience with an 8 core FX chip and 2x / 3x crossfire with 7970s.

I quickly discovered that the 990fx platform and the 8 core FX processors are a lot weaker at gaming than I had expected. Yes I did some research prior to the build, but the weak showing in processor heavy games could have been attributed to lack of threading, which I figured would go away with games that are highly threaded. Boy was I wrong.... But for a different reason. It is a platform bottleneck, not just CPU related. This is only exacerbated when using 2x or 3 cards in crossfire. There is a huge CPU/platform bottleneck on the FX series chips/chipsets That severely cripples multiple GPU configurations.

I just wanted to post this for anyone else that is considering crossfire with 7970s and wanting to play games such as Crysis 3 at high detils and resolutions (eyefinity 3x1080).

In games that are not bottlenecked by CPU and platform limitations the frame rates are good enough, but not nearly as amazing as what can be achieved on an X79 platform with the same 2x 3x 7970 cards.

I suppose for single screen gaming at 1920x1080 one or two 7970 cards would be sufficient on the FX platform. When testing Crysis 3 at very high settings @ 1920x1080 on the FX platform I get 60+ fps with 3x 7970s. Cranking up the res to the eyefinity 5760x1080 crippled the framerates into the 20's. This is just sad, and clearly not the fault of the graphics cards. Monitoring MSI afterburner reveals that each card only ever reaches ~60% utilization. And Crysis 3 in particular handles multiple CPU threading quite well. It is able to load threads across all 8 cores on an FX 8150. So it's not for lack of multi threading on the part of the game engine. Whereas in less CPU / platform dependent games such as Dirt3 all cards will be pegged at 100%, and obviously the CPU is well below 100% utilization in that scenario.

I am expecting substantial gains when moving to the X79 platform I have just received. The 3x 7970s will be paired with an i7-3820 and 32gb of ddr3-1866. My expectations are shaped by a few articles I have read in regards to platform limitations on crossfire and SLI. I'm surprised Tom's hasn't done more research into this topic. The most information I was able to find from TH was only related to running two 7970s and you can see the huge limitations that the 990fx platform puts on just 2 cards.

http://www.tomshardware.com/reviews/fx-8350-core-i7-377...

I had found another article where they had compared 990fx vs. x79 in dual, triple, and quad crossfire / SLI configurations but now I can't seem to find that article. Needless to say that the 990fx becomes the bottleneck after 2 cards.

So if you are someone intending on going for triple or quad crossfire stay away from the AMD platform (ironic...)

If anyone else has a similar experience to share or that has any insight into other articles that pertain to this issue I'd love to hear them. I will post some benchmark comparisons after I have the x79 system up and running.
March 26, 2013 5:43:51 PM

BioShock Infinite pwns the 8350 lol.
March 27, 2013 12:16:45 PM

I have yet to try Bioshock Infinite but I will be since I did get a copy with each of my 7970s, lol.

I have completed my initial testing since moving to the x79 platform. Here are my results for Dirt 3 using the in game benchmark.

On the 990fx with an FX-8150 and 3x 7970s:
72 fps Avg.
55 fps Min

On the x79 with an i7-3820 and 3x 7970s:
142 fps Avg.
122 fps Min

So right off the bat the average frame rate has nearly doubled, up 97% and the minimum is up 121%!!!

Crysis 3 is a different story... While I do see marked improvement, I am still unable to run 5760x1080 at very high settings. All of my test runs have used very high textures, fxaa, 16x AF, motion blur low, and lense flare on. Dropping the graphics setting to High results in very playable framerates. Here's what I was seeing with each platform just using fraps and playing through the same scene.

On the 990fx with an FX-8150 and 3x 7970s:
High settings:
Min fps 19, Max fps 45fps

Very High settings:
Min fps 13,Max fps 26fps

On the x79 with an i7-3820 and 3x 7970s:
High settings:
Min fps 28fps, Max fps 64fps

Very High settings:
Min fps 20fps, Max fps 4o fps

So while there is a marked improvement, it would seem that even more graphics horsepower is required. What I find really interesting is that when running on the FX-8150 I could get bursts of 120fps when looking up at the sky. On the i7-8120 the frame rate wouldn't jump above 65 or so, almost as if vsync were stuck on. This may be a driver issue, an issue with the game engine, or perhaps a cpu throttling issue? Now here's another weird thing. As I progressed further into the level (the one with the dam, I can't remember the title of the mission, I got to the point where you have to go down into the generator rooms to blow them up or whatever. I was still fiddling with settings as I went through. During the first generator section my frame rate was seemingly locked at 60fps during indoor scenes. I blew up the first generator and made my way to the second. Along the way I changed the AA settings from fxaa to the 2x smaa. This did not result in any kind of performance gain and I actually didn't care for the look of that style of AA. So I went back to fxaa. All of the sudden the game came to life and I was getting steady 60fps with the exception of a few dips into the 40s... Very strange. I'm using the latest catalyst beta 3 driver and a look at afterburner didn't reveal any change in GPU usage. Very strange. Also worth noting is that my fraps runs on the X79 platform do not feel accurate. 20fps according to fraps was very playable, yet the same 20fps on the 990fx was nowhere near playable. I have no idea what would cause this phenomenon.

Also worth noting is that still, at no time during gameplay, does either the CPU or the GPUs hit 100% utilization. I would imagine that there is still some optimization that can be done for the game engine and possibly AMD's drivers. I checked with GPU-Z and 2x of the cads are running @ PCI-e 3.0 x16 and the other at PCI-e 3.0 x8.

So that's been my experience thus far. I guess it's time to get a bigger case and another 7970 to see what it can do.

Let me know your thoughts on all of this!
Related resources
March 27, 2013 4:41:19 PM

You need to post what motherboard yoy are using. they might not be comparable.
March 27, 2013 5:37:20 PM

cia24 said:
You need to post what motherboard yoy are using. they might not be comparable.


Well you're right, they're not really comparable since one is pci-e 2.0 and only dual channel ram and the other is pci-e 3.0 and quad channel... Kinda like comparing a Fiat to a Ferrari.

But for comparisons sake they are the gigabyte 990fxa-ud5 and the gigabyte x79-up4.
March 27, 2013 6:36:40 PM

I wonder if the pcie 3 makes that big of a difference
March 27, 2013 7:18:13 PM

the pcie-3.0 is around 0-3% faster, so basically no difference at all
a b À AMD
March 27, 2013 7:18:50 PM

every1hasaids said:
I. Also worth noting is that my fraps runs on the X79 platform do not feel accurate. 20fps according to fraps was very playable, yet the same 20fps on the 990fx was nowhere near playable. I have no idea what would cause this phenomenon.


look at a site called Tech Report in reference to frame latency tests. frame latency is related to cpu as well as gpu, check out one of their cpu reviews. so the fx chip is delivering the same fps as the i7, but frame latency is uneven, and frames are delivered inconsistantly, making the display output more choppy. That is what is happening.
March 27, 2013 7:41:56 PM

Are you sure it isn't memory bandwidth problems related to high resolutions? I too have 5760x1080 and I think memory bandwidth becomes a bottleneck when I turn on AA.
March 27, 2013 8:52:42 PM

Thank you for that Crowe. True it could very well be the latency, but... There seems to be an issue where fraps is only reading part of the frame rate (I'm guessing it's reading one card perhaps) and when it is reporting 20fps, I'm actually getting 60fps. It is only doing this in Crysis and is only doing it on the X79 platform. I'm guessing there is some kind of compatability issue when using fraps with x79 and 3x 7970s. Or I've installed something incorrectly. Whatever it is, I know what 20fps looks like, regardless of frame latency. It is not displaying 20fps,it's silky smooth.

And Louis, I suppose it could be memory bandwidth that is causing the issues at 5760x1080. It definitely gobbles up the VRAM, during Crysis 3 runs it consumes up to 6.5GB of it!

Now, about pci-e3.0 vs. 2.0... How many of you have seen an article where they use multiple card configs (3x or 4x in particular) where the additional bandwidth of 3.0 slots isn't beneficial? You can't just read articles that apply to one scenario (the effects of pci-e 2.0 vs. 3.0 when usig a single card) and apply that to all circumstances. The fact is that when running 3x cards you're almost always going to have at least one card running at x8 speeds. 4x cards will have 2 cards running at x8 speeds. x8 on version 3.0 is still plenty of bandwidth to support current cards. x8 on version 2.0... mmm not quite so much. Of course thee aren't any articles out there that cover this issue as it is a fairly niche segment and anyone cramming 3 or 4 cards into a system is more than likely going to be using the x79 platform in order to pair high end CPU and memory capabilities with the high end graphics subsystem. So it's a moot point.
a b À AMD
March 27, 2013 10:01:58 PM

If you are using Lucid MVP virtu, it could very well be misreading your FPS, though normally it over inflates it hugely. Maybe there are different modes with different issues? Maybe you are getting 20 FPS, but the virtu is doing its job in lower latency, despite the low FPS, making it feel smoother than it is.
a b À AMD
March 28, 2013 1:11:26 AM

every1hasaids said:

And Louis, I suppose it could be memory bandwidth that is causing the issues at 5760x1080. It definitely gobbles up the VRAM, during Crysis 3 runs it consumes up to 6.5GB of it!


thats your answer right there, when you run out of vram it goes into system memory so yet it will have an effect. Aim to drop the game settings and/or aa to bring it within the cards 3gb vram limit. Then you can accurately guage the difference between the to platforms a bit more evenly. You can only expect poor performance when you go over a cards vram limit and spill into syystem ram.
March 28, 2013 7:03:13 AM

iam2thecrowe said:
every1hasaids said:

And Louis, I suppose it could be memory bandwidth that is causing the issues at 5760x1080. It definitely gobbles up the VRAM, during Crysis 3 runs it consumes up to 6.5GB of it!


thats your answer right there, when you run out of vram it goes into system memory so yet it will have an effect. Aim to drop the game settings and/or aa to bring it within the cards 3gb vram limit. Then you can accurately guage the difference between the to platforms a bit more evenly. You can only expect poor performance when you go over a cards vram limit and spill into syystem ram.


Well I've sill not gotten a definitive answer on whether the VRAM across multiple cards gets aggregated or not. If it does, which according to to MSI Afterburner's readings it does, then 6.5GB is well within my 9GB of VRAM. If not then I could see how this would cause issues. Although I'm not understanding how that would affect my platform comparisons.
a b À AMD
March 29, 2013 12:29:40 AM

every1hasaids said:
iam2thecrowe said:
every1hasaids said:

And Louis, I suppose it could be memory bandwidth that is causing the issues at 5760x1080. It definitely gobbles up the VRAM, during Crysis 3 runs it consumes up to 6.5GB of it!


thats your answer right there, when you run out of vram it goes into system memory so yet it will have an effect. Aim to drop the game settings and/or aa to bring it within the cards 3gb vram limit. Then you can accurately guage the difference between the to platforms a bit more evenly. You can only expect poor performance when you go over a cards vram limit and spill into syystem ram.


Well I've sill not gotten a definitive answer on whether the VRAM across multiple cards gets aggregated or not. If it does, which according to to MSI Afterburner's readings it does, then 6.5GB is well within my 9GB of VRAM. If not then I could see how this would cause issues. Although I'm not understanding how that would affect my platform comparisons.


no, when in sli or crossfire, the available vram is that of the card which has the lowest vram in the configuration. If you had a 1gb and a 2gb card in crosfire, your useable vram would be 1gb, not 3gb. a 3gb + a 3gb card gives you 3gb useable vram, not 6gb. Each card has to have its own copy of data in its vram, data can not be shared/split across the cards. if it were that way it would have a severe impact of performance as that data has to travel between cards and over the pcie bus constantly. It would be fine if the technology had each card render 1/3 of the screen, but thats not how it works, each card renders a frame aternately, so each card must have its own copy of the data for the frame its processing, it can not share with the other cards.
March 29, 2013 2:35:37 PM

This has motivated me to try and see a difference in Crysis 3 at the settings I've been running, sans one of the cards. I'd be curious to see if 2x 7970 3gb perform the same as 3x 7970 3gb cards since obviously Crysis 3 is a huge VRAM hog. I'll let you know how it goes. I may have to ttrade in my 3x 7970s for 2x or 3x 6gb versions.
!