Are AMD Current CPUs Not Enough For Solid 60 FPS Anymore?

[cadaveca]

Sucks I'm just on one card now but, I think you can leave you CPU stock and just up the CPU/NB frequency which I think is where the real bottleneck is. When Dirt 2 came out I know positively that I was running my CPU stock with CrossFire and I would get occasional slowdowns. Adjusting the CPU/NB from 2000 to 2400 fixed it.

Yeah, and I get that. However, Dirt2 released long before the signifigant driver change...before this change, Crossfire scaling didn't seem to have this bottleneck.

So, with that in mind, any review out there, that kinda looks at this issue...was done prior to that change in the drivers, and as such, must be verified, before being accepted as fact.

The same is true every month, when a new driver comes out...old reviews may not have accurate data....but because the changes were so great...I truly think that this specific period of time, when the driver was changed, is very important.

Didnt TheMailMan claim solid FPS running at stock?

It's a specific question, with a specific answer.

And as I said, he's running a single card, so the situation is OK. The rest is MailMan trolling, IMHO, because he didn''t like my answer. He ignored that I made the stipulation, that it seems once these 5-D shaders are in greater numbers than 1600, AMD cpus have issues feeding the cards data. Robert, with his 3c4850, is @ 2400 shaders, so runs into the same situation that I describe.

And yes, as I've said before, at stock, you bet Intel runs into the same problem...for some reason not as quickly. And because of this, I largely blame the driver alone...becuase, as erocker suggests here...things weren't so bad before....

I'm NOT singling out AMD cpus here....which is where you seem headed...I really think it has nothing to do with anything other than the VGA driver. BUt agian, I am NOT 100% sure on this.


Robert is looking for a list of these games that have these problems...WITH STOCK CPUS. My testing, so far, says ALL APPS have issues with Crossfire scaling...but not all are affected so much that they drop below 60FPS...hence me telling him that it's giong to take some time before I can identify each one.

Everyone brings up good points that can help overcome this problem, but none of them are actually answering Robert's question. The response "Overclock, and it goes away" isn't a proper answer. The response "I have a single card, and no issues", doesn't apply.


And Robert, please feel free to correct me here if I am wrong in any of this.

 

[CDdude55]

But that's all stuff we already knew.

What is exactly the question that needs to be answered?

 

[cadaveca]

What is exactly the question that needs to be answered?

Which apps, with stock AMD cpus, cannot break 60FPS consistently(ie, 60FPS minimum)...both with single cards, and with dual cards. If there are momentary dips in one small part of the game..I don't think he's too concerned with that. He mentioned Ghostbusters..and a specific level..that becomes unplayable. He's looking for the apps that become unplayable, and to me, specifically, the ones that are unplayable due to cpu @ stock alone.

 

[CDdude55]

Which apps, with stock AMD cpus, cannot break 60FPS consistently(ie, 60FPS minimum)...both with single cards, and with dual cards. If there are momentary dips in one small part of the game..I don't think he's too concerned with that.

But i don't understand, like i said before there is a ton of factors that can contribute to not breaking the 60 frames barrier, which apps and AMD CPU's depends on design, coding, drivers and a TON of other factors. So i guess im not understanding where the testing comes in, i mean of course it's possible to obtain 60 frames with an AMD CPU, but again, even then it comes down to more factors. What are you testing to obtain exactly?

 

[erocker]

I know for a fact that using 10.4's with CrossFire 5850's CPU at stock 2400mhz CPU/NB I would get over 60fps in Dirt 2. No question in my mind.

 

[crazyeyesreaper]

its the fact CD dude that say 10 months ago a 940be + 2 5850s might get 90fps but today on a newer driver same setup you might only get 75fps avg due to the changes amd made in the driver the point here is to pinpoint WHAT apps are most effected what fixes it etc example if a 3ghz 1800mhz nb 940be was doing fine but now that cpu must be 3400mhz with 2400nb to get the same performance it use to have do you follow now CDdude?

were talking that games that use to run fine are all of a sudden more demanding on the system for no reason causing performance loss that was never present untill the driver changed back in i think it was april or march aka crossfire profiles being released on there own ... cant remember for sure

i can honestly say my performance crossfire wise in my old 5850 review was better then on a slower cpu and ram then it is today in the same games

 

[cadaveca]

But i don't understand, like i said before there is a ton of factors that can contribute to not breaking the 60 frames barrier, which apps and AMD CPU's depends on design, coding, drivers and a TON of other factors. So i guess im not understanding where the testing comes in, i mean of course it's possible to obtain 60 frames with an AMD CPU, but again, even then it comes down to more factors. What are you testing to obtain exactly?

See Crazyeye's post...

Also, I don't think Robert cares WHY they don't perform..maybe he just wants to know so he can avoid them(no point in spending money on a gmae that you're not going to enjoy becuase it plays poorly)...maybe the actual cause doesn't matter to him.

You bring up a good point though...maybe lowering an ingame setting can fix the problem..etc...

I know for a fact that using 10.4's with CrossFire 5850's CPU at stock 2400mhz CPU/NB I would get over 60fps in Dirt 2. No question in my mind.

I will take a look without the NB OC...but basically, we could add that app to the list, as he's asking about stock cpu speeds.

It's funny too, because i remember a while ago mention that AMD might increase NB/HTT speed to 2400MHZ for Thuban..but that proved to not be the case. There are some Opterons out now, though, that have 2200mhz stock HTT. Maybe AMD is very aware of the "problem"...if you want to call it that. I said before this wasn't all doom and gloom...but it's an issue for me that means I personally need to spend more $$ on my pc.

 

[pjladyfox]

I'm still going thru this thread so if this has been mentioned earlier I'll try and address it in an update once I'm done.

It's interesting that this came up since I just recently got finished researching out something similar for work. Now these thoughts are all my own and based solely upon the software and tools I was using and should not be considered definitive by any means. Each piece of software is different, due to either design or other factors, and because of this having an absolute set of results is not really possible so I share this with the hopes that it will generate more discussion and maybe help others who enjoy benchmarking hardware.

With that out of the way basically I was trying to identify what factors were affecting some performance issues with a software title I was working on so that I could configure a direct performance comparison between an Intel and AMD CPU. After playing around with things I noticed that the following factors had some influence in performance:

a. CPU speed - This one typically comes into play if the software in question is in a situation where there is a performance mismatch between the CPU and the video card preventing both from working optimally. Now this is a difficult thing to find since each piece of software tends to react differently such as the various differences between say the Unreal and Source engines.

Once you have found performance parity, or basically a situation where even if there is improvement it's not noticable without other tools, then you move onto the next factor.

b. L2 cache size - This one, while you would think would be a no-brainer for many, caught me by surprise. Like a lot of people I figured that if you had enough speed that this would not matter but that turns out not to be the case.

For example, I compared two different Athlon X2 4400+ processors against a stock Core2 Duo E4300. On both systems I used the same RAM and same video card but was seeing a variance of about 3 to 5 percent. After talking with some people at work and at AMD it turns out that it was the amount of L2 that was causing the issue with the older 4400+ having 2MB versus the newer 4400+ only having 1MB compared to the E4300's 2MB.

I then ran into the next issue...

c. L2 cache latency - This one was harder to detect in my case since the variance was only about 1 to 2 percent but I did eventually peg this as the only logical reason for what I was seeing. But, realistically, the earlier factors paid a more larger part than this did and only really seemed to come into play when the software was more RAM intensive than normal such as software like Photoshop or 3D Studio Max which use a LOT of RAM depending upon the operation.

d. L3 cache - This one, like the L2 cache, only really came into play if one of the CPU being compared failed to have any while the other one did. I really did not get to play much with this one since I really did not have that much hardware that directly compared well to begin with and I ran out of time. But, if you're ever picking a CPU this can and should be a factor to consider.

 

[cadaveca]

Thanks for "unvilifying" my posts in the past about cpu cache being important..I had many people(funny, the some of the same people in this thread) say that cache didn't matter.

I also mentioned that L3 cache may be part of the issue here, as increasing NB speed in AMDchips also increases L3 speed too...

 

[pjladyfox]

the problem here *as stated in the other thread* is that you're running crossfire. that adds more CPU load into the mix.

If you had one GPU your performance would end up being better, due to having less CPU power needs on your older/weaker CPU. (remember, me and mailman have 2 more cores than you, and a newer CPU design)

i'm only upto page 3 here and i can already see whats happened. cad knows he's CPU limited and would rather blame AMD, than blame himself for having too slow a CPU to run his video cards.

Mailman is trying to say that he has zero issues whatsoever, even at a stock CPU - because he doesnt have crossfire.

I'm still pouring thru this myself and the more I read the more I'm agreeing with this so I'm glad someone said it already.

Both Crossfire and SLI, at least in my experience, are a LOT more CPU intensive at all levels than many realize. While I give credit to NVIDIA for their ability to evenly scale I think they really did everyone a disservice by doing so since many expect for this to hold true regardless of the CPU being used. I've lost count of how many times I've tried to explain this to someone who goes on a tirade why their rig is not performing like they think it should with a newer piece of software.

Don't get me wrong here I'm not bashing on any particular company but I do find it funny sometimes where when NVIDIA does comparisons to Crossfire in most cases I've seen it's never using the same platform as what AMD uses. Of course, the reason for this is so that they can quietly avoid the CPU bandwith issue and continue to sell their bridge chip to Intel which of course does not have the problem seeing how the platform used does not have performance parity with the AMD system. But I'm getting off on another tangent that really is not pertinent to the discussion at hand.

Thanks for "unvilifying" my posts in the past about cpu cache being important..I had many people(funny, the some of the same people in this thread) say that cache didn't matter.

I also mentioned that L3 cache may be part of the issue here, as increasing NB speed in AMD chips also increases L3 speed too...

I'm still pouring thru this thread but I must admit I'm surprised this did not really get focused on. Still, pretty interesting thread so far so I'm glad it did not get locked early on.

Robert(the OP) highlighted this too, in GhostBusters...his vgas ahve enough grunt to push that game maxed out, but certai nscenarios lead to really poor performance, just simply due to the number of vgas he has(of course, the investigation revolves around finding out exactly when these limitations are exposed, and it may just tunr out to be bad game programming..and not the number of vgas...that's the question that needs to be answered.).

I must have glossed over this or missed it but one thing I think needs to be considered here is that the base game itself really was not designed to scale beyond a single video card and a dual-core CPU. With that in mind any artificial (read: SLI or Crossfire profile) performance improvement with this title will focus more on the CPU with it being affected more by L2 and L3 cache size than CPU speed. To rule this out a good comparison would be the following:

Intel Core i5 750 2.67GHz, L2 4 x 256kb, L3 8MB 95w (or a Core2Quad Q9550 which would be a better match I think)

vs

AMD Athlon II X4 640 3.0GHz, L2 4 x 512kb, L3 n/a 95w

vs

AMD Phenom II 945 3.0GHz, L2 4 x 512kb, L3 6MB 95w

I would be willing to bet that where you would see performance issues on the Athlon II that the Phenom II nor Core i5 would suffer from mainly due to the addition of the L3 cache which would help any multi-GPU configuration and hinder those without it. I know the Core i5 in there will most likely have better performance than either of the AMD ones but I'm not really sure to be honest.

 

[dir_d]

I'm still pouring thru this thread but I must admit I'm surprised this did not really get focused on. Still, pretty interesting thread so far so I'm glad it did not get locked early on.

I tried to focus on it but they went another way with the discussion. I strongly feel that the CPU-NB is the key, specially since now there is conclusive data by Anandtech.

 

[cadaveca]

I tried to focus on it but they went another way with the discussion. I strongly feel that the CPU-NB is the key, specially since now there is conclusive data by Anandtech.

Oh, I agree 100%(as does erocker, clearly). The only point I was trying to make when you mentioned that previously was that the situation may not be exactly as Anandtech reports...because that review was so old...it might actually be worse!

 

[crazyeyesreaper]

indeed ^ it IS worse cadaveca i re went through my old 5850 crossfire review back when i was using RELEASE DRIVERS and performance today is WORSE then it was back then...


940BE at 3.4ghz 2000nb 4gigs DDR2 800mhz 5-5-5-15 2T back then gets better performance then
965BE at 3.6ghz 2400nb 4gigs DDR3 1333mhz 7-7-7-20 1T
performance is about 7-9% less then it use to be for me granted i WONT notice it as i use Vsync at all times but when i did the review i ran with it off and if i run those same games now yea..... the difference is there and its extremely noticeable

Phenom II 940 + 5850 crossfire mini review
example 1680x1050 VERY HIGH in crysis DX10 mode 4xAA i got 37fps average on the 940be + DDR2 now at same settings im only around 33fps with a 965be DDR3

 

[TheMailMan78]

I know for a fact that using 10.4's with CrossFire 5850's CPU at stock 2400mhz CPU/NB I would get over 60fps in Dirt 2. No question in my mind.

With that being said I guess its safe to say its basically a driver problem and nothing more? I mean look at it this way.
1. CPU is the same. Even faster if you OC.
2. GPU is the same.
3. RAM is the same.
4. Driver is different.

Hmmmm I wonder what the problem is.
This isn't brain surgery.

 

[crazyeyesreaper]

its the driver true mailman but 10.3 10.4 10.5 10.6 10.7 10.8 all have the SAME issue so 6months of the same PERFORMANCE breaking issue and its kinda shhhhh dont say anything or AMD will freak sorta thing

 

[TheMailMan78]

its the driver true mailman but 10.3 10.4 10.5 10.6 10.7 10.8 all have the SAME issue

Not true. 10.4 did not have the issue. You want 60 FPS with crossifre? Use those. Problem solved.

 

[cadaveca]

Not true. 10.4 did not have the issue. You want 60 FPS with crossifre? Use those. Problem solved.

Not true. Second card is useless, even with 10.4. But, single card performance is very high with 10.4 in comparison to other drivers, so it's not as evident.

10.4 is probably the best driver for this issue, but 10.8 is good too..single card performance is lower, but Crossfire performance is a fair bit better than 10.4

For example, 10.4..I can run BFBC2 with one card, no problem. add second card..enable 4xAA...performance TANKS. 10.8, 4xAA works now...but FPS is still the same as single card.

It's really weird, TBH.

 

[crazyeyesreaper]

uh wrong mailman im still using 10.4 and performance is worse here then the 9.11 hotfix drivers that fixed xfire long ago problem is 9.11 is before the updatable crossfire profiles so i cant roll back either otherwise BC2 runs like shit so its pretty much a no win situation right now

 

[TheMailMan78]

Well according to erocker his was fine.

 

[crazyeyesreaper]

and erocker was back then at 4ghz + with a 2600 or so NB thats a far cry from STOCK

im seeing WORSE performance today from a 965be with 1333mhz 7 7 7 20 1T on 10.4 cpu at 3.6 nb at 2.4


then i did 10months ago on a 940BE at 3.4ghz 2200nb and in the review i posted stock and oc speed made ZERO impact on crossfire performance so in 10months it took a 600mhz cpu clock speed increase and 600mhz NB speed bump to get the same performance i had back then on the same games

 

[TheMailMan78]

and erocker was back then at 4ghz + with a 2600 or so NB thats a far cry from STOCK

im seeing WORSE performance today from a 965be with 1333mhz 7 7 7 20 1T on 10.4 cpu at 3.6 nb at 2.4


then i did 10months ago on a 940BE at 3.4ghz 2200nb and in the review i posted stock and oc speed made ZERO impact on crossfire performance so in 10months it took a 600mhz cpu clock speed increase and 600mhz NB speed bump to get the same performance i had back then on the same games

Nope.....

I know for a fact that using 10.4's with CrossFire 5850's CPU at stock 2400mhz CPU/NB I would get over 60fps in Dirt 2. No question in my mind.

 

[cadaveca]

Well according to erocker his was fine.

Maybe i notice it, and he doesn't, because I use 5870's rather than his 5850's. I don't know. But given the testing so far, that would make sense...more shaders=bigger bottleneck. I have 10% more shaders...might make for 10% more bottleneck...I don't know yet.


Like I mean really...adding a second card should enable 4xAA...fortunately, with 10.8 it does...but all previous drivers failed.

And still, with stock CPU..BFBC2 is more often than not 45-60FPS, with 2560x1600. But it's one of those apps that plays fine @ 45FPS...

 

[TheMailMan78]

Maybe something was not set up correctly then and now bum drivers have added to the problem?

 

[Nick89]

I call BS. My 940 BE OCed to 3.4 with my 480 plays all my games EXCEPTIONALLY well.

 

[cadaveca]

Maybe something was not set up correctly then and now bum drivers have added to the problem?

You could be right. And that's a big reason why I haven't really mentioned this before...I try to complain only about problems that can be replicated, and not just by the two systems I have here...

I don't know what the hell is going on, and I've said this from the start. It seems to be the driver, and seems related to the number of shaders....but only seems...

For all I know, it could be because the "master" card in my system is in the bottom slot...could be some weird bios problem...I have two rigs exactly the same same, save the PSU, so any of the components could be the cause, I suppose(except the PSU).

I think i have eliminated OS issues, driver install problems, and other such oddities...and other people that have completely differnt configs(Intel vs my AMD), but use the same drivers, have the same problem...but it's very possible I overlooked soemthing...

That's why I'm doing the testing, to try to isolate the problem. But even that seems wasteful, given that much older drivers, from last year, seem to not have this issue...so it's so very easy to just say "It's the driver"...

I call BS. My 940 BE OCed to 3.4 with my 480 plays all my games EXCEPTIONALLY well.

You're overclocked. And you're using a single card. :shadedshu So of course you do. Multiple cards, and stock cpus, please.