Proof that AMD FX CPUs do not bottleneck high end GPUs

[kipper]

 

And explain this please sweetheart, a 480 and 2500k killing a 7970 and FX in the exact same Beta.

[Nena Trinity]

That's the thing: YOUR games. If it works for games that fully utilize 8 threads, that's just fine.

 

Also, the issue with comparing to your buddies with Intel CPU's is that you failed to mention which CPU's they were using or even the games in general.

 

It'd be so easy to say: "my friend with a Pentium G3258 has more problems with his processor than I do with my FX-8350 in games".

 

People need to start differentiating between 'good-enough/acceptable/etc.' and 'best' within a price category. It's not like FX chips will straight up refuse to run games that a price-equivalent Intel processor would, it's just that a lot of the tested metrics show data in favor of Intel being the BETTER option in many cases, not just simply the only ACCEPTABLE option for a build.

 

People want to buy an FX-6300 for their new budget gaming build? Go ahead, it's your money. Many people here would certainly advise against it, because they could be buying better performance for similar costs. Good-enough isn't good enough for enthusiasts; it may be perfectly fine for an average user, so keep that in mind when seeking advice from enthusiasts. It's like asking biking enthusiasts if a $50 mountain bike from Walmart is a smart buy for the Tour De France, just because it works in getting them across town (I'm terribly sorry for this misleading analogy, but my point still stands).

 

You can buy AMD processors, it's not banned in the forums. But if there are clear reasons given as to why one is advised to buy one processor over the other currently, that does not mean that whoever owns those processors are retarded monkeys that play Smash Bros with a Wiimote stuck up their butt. Everyone wants to validate their purchase, I get it; and in most cases for average users, those purchases are validated, whether it be Intel or AMD, but the difference comes down to what enthusiasts value as BEST.

Everyone I know has i5 or i7 and they are 3rd and 4th generation of them to.

 

Also yes it depends alot on the games!

[SkilledRebuilds]

And explain this please sweetheart, a 480 and 2500k killing a 7970 and FX in the exact same Beta.

"It's a Beta, You cannot compare metrics in a Beta!!!"

[XCalinX]

So a CPU with double the core count of a 2500k at 4.7ghz could not beat a 4 core sandy at 5088mhz?

 

Now let me just point this out to you, that same i5 demolishes all FX systems in BF4 with a GTX 480 (CPU limited)

For God's sake. I AM RUNNING SINGLE CHANNEL RAM WHAT PART OF THAT CAN'T YOU UNDERSTAND?

[incarnate]

For God's sake. I AM RUNNING SINGLE CHANNEL RAM WHAT PART OF THAT CAN'T YOU UNDERSTAND?

Single channel vs dual channel doesn't make that much of a difference in gaming.

[That Norwegian Guy]

it came out worse vs a 2500k with half the core count

 

*Three quarters the "core" count.

 

8350 = 8 Integer units + 4 floating point units = 12 computational units = 4 full cores and 4 crippled cores

2500k = 4 integer cores + 4 floating point cores = 8 computational units = 4 full cores

[i_build_nanosuits]

For God's sake. I AM RUNNING SINGLE CHANNEL RAM WHAT PART OF THAT CAN'T YOU UNDERSTAND?

less than 5% OVERALL MEMORY PERFORMANCE improvement from running single channel DDR3 vs dual channel...so when it comes to actual gaming performance: next to zero benefit from dual channel RAM, i tested this already you can find all my testing in my youtube videos.

[russler]

just bought another 290 so i cant wait to see how an 8350 will cope, dont get me wrong i bought the 8350 when i was just starting pc building and tbh i think intels prices seem extortionate when you dont really know the differences 

[Edgar R. Zakarian]

You run battlefield hardline at 4k... Of course there isn't gonna be a bottleneck since at that resolution the games become gpu dependent. Also not to mention that the price of the 8350 with a 990 fx board and water cooling is as much as an i7.

I dno, I got mine used 8350 990fx motherboard and air cooling asus silent knight 2 for less than 200$ in Denmark.

[david cassar]

I dno, I got mine used 8350 990fx motherboard and air cooling asus silent knight 2 for less than 200$ in Denmark.

In denmark? How is that even possible

[sgloux3470]

The word bottleneck is used so liberally and carelessly here.

Here's the deal. There's always going to be a limiting factor in your PC. Sometimes it's your GPU, sometimes it's your CPU. If you run a low end resolution like 1080p on a high end GPU, the CPU will be your limiting factor. And in that situation yes an AMD FX CPU is more of a limiting factor than an i5 or what have you. But as you tax the high end GPU more, the GPU becomes the limiting factor again.

There's nothing inherently wrong with the CPU being the limiting factor. In titles that need the Agpu horsepower it will get used. In titles that don't, you don't need the full performance of the high end GPU anyway. (except maybe for frame rate bragging rights.). The best experience involves a frame rate cap with either vsync or gsync anyway.

For 120Hz+ thought it does matter more, but imo high refresh rate is a fad.

[juretrn]

Yes. Overclocked to 4,7ghz it almost matches the performance of a STOCK i5.

FTFY

[ZetZet]

FTFY

well yes, that is what i meant.

[d1screet]

The word bottleneck is used so liberally and carelessly here.

Here's the deal. There's always going to be a limiting factor in your PC. Sometimes it's your GPU, sometimes it's your CPU. If you run a low end resolution like 1080p on a high end GPU, the CPU will be your limiting factor. And in that situation yes an AMD FX CPU is more of a limiting factor than an i5 or what have you. But as you tax the high end GPU more, the GPU becomes the limiting factor again.

There's nothing inherently wrong with the CPU being the limiting factor. In titles that need the Agpu horsepower it will get used. In titles that don't, you don't need the full performance of the high end GPU anyway. (except maybe for frame rate bragging rights.). The best experience involves a frame rate cap with either vsync or gsync anyway.

For 120Hz+ thought it does matter more, but imo high refresh rate is a fad.

> low end resolution like 1080p

Well then, I suppose I'll need to find myself a cheap 1080p monitor at some point

 

A good number of gamers tend to shy away from vsync due to the slight input-delay that it introduces, making controls feel sluggish; but even then, most games HAVE framerate caps defined, it's more that people tend to overstate them for as many frames as possible.

[Prysin]

You know what AMD should do?

 

FX Refresh. GIve us all the FX processors, but make them using Excavator.....

 

To give an idea.

 

FX 6300, 6350, 6370, 8320, 8320e, 8350, 8370, 8370e, 9370 and 8580 is all Piledriver.

Now, Steamroller is the type used in the A10 7850k Kaveri APU and the Athlon X4 860k. Steamroller has around 4-5% better IPC then Piledriver

Excavator (Carizzo APU) has yet to launch, but its IPC is 4-10% better then Steamroller. This means that if AMD isnt giving us ZEN, then atleast give us the more energy efficient Excavator cores and sell them low end like i3 low end... They would be great... all 65-95w parts with up to 6 modules (yeh, excavator is much smaller so they could very likely fit 6 actual modules in the same die area as current FX)..... That would actually be "nice".... Probably even acceptable performance.

 

/brainfart

[othertomperson]

If you want to prove that your CPU does not cause a bottleneck, then first of all use a wide selection of games, and test them all with vsync off at 800x600 and no Anti-aliasing. This will remove any GPU limitation from the equation and your FPS should go through the roof. Effectively this will be the maximum fps that your CPU can provide. If this drops below 60 ever then you can conclusively say that your CPU is a limitation here -- no number Titan Xs would give you 60 fps.

 

This is by far the cheapest way of isolating your CPU's performance -- the alternative is to buy a Titan X.

[Edgar R. Zakarian]

In denmark? How is that even possible

It was all used

[Edgar R. Zakarian]

If you want to prove that your CPU does not cause a bottleneck, then first of all use a wide selection of games, and test them all with vsync off at 800x600 and no Anti-aliasing. This will remove any GPU limitation from the equation and your FPS should go through the roof. Effectively this will be the maximum fps that your CPU can provide. If this drops below 60 ever then you can conclusively say that your CPU is a limitation here -- no number Titan Xs would give you 60 fps.

 

This is by far the cheapest way of isolating your CPU's performance -- the alternative is to buy a Titan X.

Will try this on CS:GO when I get home, thank you!

So if Crossfire (bad optimize) is bottlenecking, then this should show it.

[Umeshwarnath]

i have a challenge for you , test PlanetSide 2 i can't stay at 60FPS with everything on low .. it dips down from time to time ; on the other hand people with i3 CPUs stay at 60 if not over that 95% of the time

 

this is what i'm getting atm with Raptr and my specs

ahaha.....i play ps2 on an 8320 at 4.3ghz on high...i consistently get above 60fps ...Even on large fights with 200+ players i get 50fps...the frame rate is consistent .......it does dip to 45fps once in a while inside biolab(96+ vs 96+).... but  thats due to the gpu and not the cpu. i know for a fact that an i5 would do better .............but .... you will only be lying to yourself if u think the 8350 cant keep up 60fps. ****comments made wrt the game now after all the performance patches. i have no idea how it was a month ago.

[Umeshwarnath]

If you want to prove that your CPU does not cause a bottleneck, then first of all use a wide selection of games, and test them all with vsync off at 800x600 and no Anti-aliasing. This will remove any GPU limitation from the equation and your FPS should go through the roof. Effectively this will be the maximum fps that your CPU can provide. If this drops below 60 ever then you can conclusively say that your CPU is a limitation here -- no number Titan Xs would give you 60 fps.

 

This is by far the cheapest way of isolating your CPU's performance -- the alternative is to buy a Titan X.

but that is not how you play your games right? or do you? in a real world scenario where you would be playing high--ultra settings@1080p ,you wanna know if a given cpu is enough to drive a given gpu whilst keeping consistent framerate for a good experience. also i am not trying to say the 8350 would do a better job than the i5 because  i know it doesnt . but if one doesnt have the money for the intel cpu or  just wants an amd cpu....everyone mustnt pounce on the person saying they are wrong or foolish and treat them like an outcast.

[linuxfan66]

Can he test Arkham knight as a benchmark[emoji57]

[othertomperson]

but that is not how you play your games right? or do you? in a real world scenario where you would be playing high--ultra settings@1080p ,you wanna know if a given cpu is enough to drive a given gpu whilst keeping consistent framerate for a good experience. also i am not trying to say the 8350 would do a better job than the i5 because  i know it doesnt . but if one doesnt have the money for the intel cpu or  just wants an amd cpu....everyone mustnt pounce on the person saying they are wrong or foolish and treat them like an outcast.

 

It doesn't matter if that's how you play the games or not: you're benchmarking it, not playing it. Real World benchmarks are true for that particular system and that particular system only. You said that "AMD FX CPUs do not bottleneck high end GPUs". This is how you prove that -- by finding out what the limit of the CPU actually is. Then you can compare with an i3 (which is about the same price and the CPU people are actually comparing it against) to see which just flat out handles it better without relying on a mid-range GPU to make the CPU look better than it actually is -- which is what "real world" approach relies on.

 

If your CPU can provide 60 fps with the GPU out of the equation then by all means claim that pairing this CPU with a top end GPU such as the 980 Ti is a good idea. Until you do that no amount of real-world fudging is going to help your case.

 

Ultimately the reasoning comes down to this: if you ever get framedrops at 800x600 without AA then that 980 Ti was a waste of money and you may as well have gotten a 960 or 280X.

 

 

Will try this on CS:GO when I get home, thank you!

So if Crossfire (bad optimize) is bottlenecking, then this should show it.

 

CS:GO seems pretty CPU bound anyway -- which isn't saying much, that game seems to run on literally anything.

 

I would also think that Crossfire bottlenecking would have the opposite effect. If a game doesn't support SLI/CF effectively you either get massive microstuttering, or it just doesn't use the second GPU at all, effectively giving you half the GPU you could theoretically have.

 

If I were testing CF scaling I would do the opposite. I would want to effectively remove the CPU from the equation. I would do this by running the game at 4K with all of the motion effects and depth of field -- but I would be keeping an eye on your vram usage. If you go over the vram of either card that would effect your result as well. I would test the game with a single card, test with two and see if there's any difference (positive or negative).

 

I would also make sure that, again, the CPU itself is capable of running the game at a low resolution effectively. If your CPU is still bottlenecking you in spite of you trying to make the game as GPU-bound as possible your results aren't going to tell you anything.

[Sam Z Man]

*Three quarters the "core" count.

 

8350 = 8 Integer units + 4 floating point units = 12 computational units = 4 full cores and 4 crippled cores

 

They are not full cores and crippled cores, every core is equal they just have very low IPC compared to Intel.

[Faceman]

They are not full cores and crippled cores, every core is equal they just have very low IPC compared to Intel.

Not in the slightest.

 

Cores Vs. Modules

When you look at the Bulldozer Block Diagram, what do you see?

I see 4 modules, and 8 integer cores. 

 

Lets define what a core is:  A CPU core is a collection of components in a processor used to process code. Its not just the logical processor (the integer core), not just the FPU, not just the decoder, not just the l1 cache not just the l2 cache but all of these components combined is what forms a core.

 

Modules and cores are very different. This module design is a very old microarchitecture design(1996) and it doesn't work the same way a true core does. The issue with this design is that resources are shared, namely the floating point unit.  The floating point unit is very important for gaming.  So while it does have 8 integer cores/threads, those cores have to share resources for certain tasks, and this causes one core to wait until the other is finished.  Because of this, it is not a true 8 core processor.

 

A true core 8 core processor is when there is no situation you can present short of disabling cores that will cause even one thread out of eight to have to wait to be processed. Not a single situation. You can't say the same thing about the FX processors. If both threads on a module need the FPU one of the threads will have to wait.

[Sam Z Man]

snip

 

I know this, i was referring to the fact that he said that it had 2 different types of cores in each module, when in reality each module has 2 identical processing units in it.