GTA V CPU benches - AMD gets hammered (again)

[Phynaz]

Yep, that hatred of AMD I have, as shown by my first post in this thread about those crappy AMD Pentiums.

 

[TheELF]

you're not thinking hard enough about this to be worth my typing it out. I already gave you enough to be able to understand it.

you said:


and I just gave you a way to increase single threaded performance 15% on the module running the main thread.

There is no one module the main thread is running on,thread migration will juggle the main thread around on all available cores/modules/threads.
If you disable one core per module,so each core that is left will get access to all resources of the module than, maybe.
BUT you will have to actually show us that there is a difference,just saying that there is because there is an effect at cinebench doesn't make it universally applicable.

 

[Hitman928]

What is definitly broken is your ability to understand how a CPU work electricaly speaking, ironicaly you re using your clulessness as an argument.

Out of pity here an hint : (4.7/4.4)^2 = 1.141

This is the power ratio between a 9590 and a 9370, the power numbers i gave are based on Hardware.fr review of the FX9590.

And you, what are your own technical references and credentials..?.

Obviously a ferocious hate for AMD, you should definitly put an end to your constant thread crapping as it s ages that you re bringing absolutely no technical value to this forum, quite the contrary, that s all negativity, contra revenue anyone..?..

What he meant is that the post you were quoting contained sarcasm that you took literally. Anyway, why are you taking the square of the frequency ratio when talking about power?

 

[Phynaz]

No amd just inflates their tdp to make the cpu less attractive. They have too many customers so they are trying to deter a few.

Dude, this is funny :sneaky:

 

[Abwx]

Anyway, why are you taking the square of the frequency ratio when talking about power?

Because GF process has a frequency/voltage curve that is 4.68% more voltage for 10% more frequency in the range that interest us.

When squaring the voltage ratio (1.0468) we get 1.09579 that i round to 1.1, hence the frequency ratio being apparently squared.

 

[Hitman928]

Because GF process has a frequency/voltage curve that is 4.68% more voltage for 10% more frequency in the range that interest us.

When squaring the voltage ratio (1.0468) we get 1.09579 that i round to 1.1, hence the frequency ratio being apparently squared.

Ignoring that your explanation makes no sense, do you actually have access to a GF white paper that shows the voltage curve to which you're referring?

 

[Abwx]

Ignoring that your explanation makes no sense, do you actually have access to a GF white paper that shows the voltage curve to which you're referring?

You better shouldnt ignore that my explanations make no sense, this will spare me some other explanations that you dont need since they would be even more non sensical.

More seriously i dont need a white paper to extract thoses numbers as they are publicly available for whom make the effort to compute the numbers within relevant reviews...

like thoses ones, difference between early Vishera and matured process based chips is clearly noticeable in the frequency limitation, also, do you want an exemple extracted from one of thoses graphs or can you compute it by your own means.?.

October 2012 FX8350

June 2013 FX4350 and FX6350

September 2013 FX9590

September 2014 FX8370E

http://www.hardware.fr/articles/880-5/overclocking-undervolting.html

http://www.hardware.fr/articles/899-3/influence-turbo-undervolting-overclocking.html

http://www.hardware.fr/articles/905-2/am3-220w-turbo-frein-overclocking.html

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

 

[cytg111]

I emulated both a i3 and a pentium with my i7 3770 non k and i can assure you the pentium emulated resulted in freezing,stuttering and erratic unplayable frames while the cpu was maxed out@100%.More cache and faster overall memory do give the emulated pentium a advantage so imagine what the real one would perform like lol.

I found the i3 to be enjoyable enough,i wouldn't mind recommending it myself in a pinch.

Nice, do you have a chance to inspect the actual threads? Maybe even, if realtime, manually lower em' a notch (on the pentium emu of course)

 

[cbn]

Deleted.

 

[jpiniero]

Why do you think this game is not scaling with cores?

Just take a look at the FX-4300 vs FX-8350 in GameGPU.

Or look at the i7-5960X vs i7-4770K (the i7-5960X is still faster despite a lower base clock and turbo) in Game GPU.

There is a benefit above 4 threads, but it's slight. Look at the difference between the 6350 and the 8350, it's not much. The 8350 also has more L3 cache too.

 

[cbn]

There is a benefit above 4 threads, but it's slight. Look at the difference between the 6350 and the 8350, it's not much. The 8350 also has more L3 cache too.

The gains in Game GPU going from FX-4300 (3.8 Ghz base, 4.0 Ghz turbo, 4MB L3 cache) to FX-8350 (4.0 Ghz, 4.2 Ghz turbo, 8MB L3 cache) are pretty good!

PC labs results aren't quite as encouraging, but it could simply be the scene they used.

Still the results show a 15% gain in FPS with 4.7 Ghz FX-6350 over 4.7 GHz FX-4300 (just realize FX-6350 has 8MB L3 cache and FX-4300 has 4MB L3 cache. However, I haven't seen a lot data showing L3 cache on AM3+ adding up to much of a performance gain in gaming). With that mentioned, some of that gain could come from reduction in module penalty.

P.S. The FX-8350 and FX-6350 have the same L3 cache at 8MB. The FX-4300 is only Vishera with 4MB L3 cache.

 

[desprado]

What is definitly broken is your ability to understand how a CPU work electricaly speaking, ironicaly you re using your clulessness as an argument.

Out of pity here an hint : (4.7/4.4)^2 = 1.141

Insulting other members is not allowed, or trolling.
Markfw900

This is the power ratio between a 9590 and a 9370, the power numbers i gave are based on Hardware.fr review of the FX9590.

And you, what are your own technical references and credentials..?.

Obviously a ferocious hate for AMD, you should definitly put an end to your constant thread crapping as it s ages that you re bringing absolutely no technical value to this forum, quite the contrary, that s all negativity, contra revenue anyone..?..

i dont understand why u always troll on topics based on AMD negative performance?

 

[soccerballtux]

There is no one module the main thread is running on,thread migration will juggle the main thread around on all available cores/modules/threads.
If you disable one core per module,so each core that is left will get access to all resources of the module than, maybe.
BUT you will have to actually show us that there is a difference,just saying that there is because there is an effect at cinebench doesn't make it universally applicable.

thread migration doesn't happen in WoW because they locked it down. Considering this is a console game, it might not happen here either.

Additionally you can use Process Lasso to do the same thing-- lock it down and prevent the thread juggling.

Why wouldn't it be universally applicable? This has come up in plenty of other reviews across the net, I just explicitly tested for it. I DID _SHOW_ you in CB.

You're spreading FUD without providing a reasonable excuse for saying it doesn't apply across the board. You can go download some benchmarks and prove me wrong easily enough yourself; Cinebench is a well known test and is good enough for my stamp of approval aka 'this will happen across the board'. My stamp of approval is more valuable than your FUD because I have good reasons for it.

 

[soccerballtux]

What he meant is that the post you were quoting contained sarcasm that you took literally. Anyway, why are you taking the square of the frequency ratio when talking about power?

power consumption increases with the cube of the voltage, and also scales with frequency.

 

[Dave2150]

Updated chart with oc sandy and more cpus.Old sandy is still pretty good and you can oc it to 5Ghz then most likely it will be faster than OC 4790k

I suggest looking at your own graph again.

The 4790k at STOCK (4Ghz) is faser than a 2500/2600 at 4.5Ghz.

Clock the 4790k to 4.6-5Ghz (yes, some 4790k's do clock this high) and will will still beat the 2500/2600 even if they are clocked to 5Ghz.

 

[Hitman928]

You better shouldnt ignore that my explanations make no sense, this will spare me some other explanations that you dont need since they would be even more non sensical.

More seriously i dont need a white paper to extract thoses numbers as they are publicly available for whom make the effort to compute the numbers within relevant reviews...

like thoses ones, difference between early Vishera and matured process based chips is clearly noticeable in the frequency limitation, also, do you want an exemple extracted from one of thoses graphs or can you compute it by your own means.?.

October 2012 FX8350

June 2013 FX4350 and FX6350

September 2013 FX9590

September 2014 FX8370E

http://www.hardware.fr/articles/880-5/overclocking-undervolting.html

http://www.hardware.fr/articles/899-3/influence-turbo-undervolting-overclocking.html

http://www.hardware.fr/articles/905-2/am3-220w-turbo-frein-overclocking.html

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

You've shown nothing that backs up your original statement. All you've done is taken a bunch of data points and tried to fit them to your own personal theory within a limited scope. No thanks.

power consumption increases with the cube of the voltage, and also scales with frequency.

Based upon a more or less first order analysis (I'm using this term loosely) digital cmos designs have a squared voltage to power relationship, assuming all else is equal. The frequency to power relationship is linear. There are other factors that come into play and it gets really complicated with the insane amount of transistors in modern designs, modern power saving techniques (at the hardware level), the increasing impact of static power dissipation, etc. However, no one should argue that power increases by the square of the frequency, it makes no sense even within the limited scope that Abwx was using.

 

[SPBHM]

I suggest looking at your own graph again.

The 4790k at STOCK (4Ghz) is faser than a 2500/2600 at 4.5Ghz.

Clock the 4790k to 4.6-5Ghz (yes, some 4790k's do clock this high) and will will still beat the 2500/2600 even if they are clocked to 5Ghz.

stock 4790K is 4.2GHz if you leave turbo enabled and have high load.

 

[greatnoob]

Requoting so /thread:

Just a heads up, an open-world game like this likely uses worker threads which means any one core can be utilised to its max.

The more worker threads -> more cores being used and/or more load on a core. This means any worker thread can be run on any core. So there could be 10 worker threads running on just one core rather than 2 on each core (it all depends on the work dispatcher and OS). No use trying to find a trend when it doesn't exist!

 

[TheELF]

thread migration doesn't happen in WoW because they locked it down. Considering this is a console game, it might not happen here either.

It might it might not,thats why anyone should show some proof with their claims.

Additionally you can use Process Lasso to do the same thing-- lock it down and prevent the thread juggling.

You can lock down the whole game to only one core,thats not really good,if you know a way to lock down individual threads to certain cores than please do share.
Using affinity will not lock down the threads,they will still be jumping around but only on the cores that you alow,that is the same thing that I said you should do.

Why wouldn't it be universally applicable? This has come up in plenty of other reviews across the net, I just explicitly tested for it. I DID _SHOW_ you in CB.

I have nothing against cb,but in cb every core gets used 100% and you gain 15% ,games only use one core 100% so there is no way of a game having the same amount of gain.
I only said that you should test it and show how much difference it really makes.

You're spreading FUD without providing a reasonable excuse for saying it doesn't apply across the board. You can go download some benchmarks and prove me wrong easily enough yourself; Cinebench is a well known test and is good enough for my stamp of approval aka 'this will happen across the board'. My stamp of approval is more valuable than your FUD because I have good reasons for it.

How am I spreading FUD?
Do you really believe that every program/game uses the exact same number of commands or even command sets?
FXs are constantly beating intels at cb and still in games it's the complete reverse.
I don't have an FX so I can't test it on this or any other game,so I'm asking someone who has one to show us how much you could really gain from doing something like this.
If it is 15% then kudos.

 

[Erenhardt]

Purepc.pl updated their tests with CPU graphs:
http://www.purepc.pl/karty_graficzne/jakie_sa_wymagania_sprzetowe_gta_v_test_wydajnosci?page=0,8

Its good to see fx8300 gives some benefit from the extra 2 threads compared to fx6300.

 

[TurtleCrusher]

Purepc.pl updated their tests with CPU graphs:
http://www.purepc.pl/karty_graficzne/jakie_sa_wymagania_sprzetowe_gta_v_test_wydajnosci?page=0,8

Its good to see fx8300 gives some benefit from the extra 2 threads compared to fx6300.

FX-8350 and 7870 get minimums of 58fps and average 62fps at 1920x1080 on high with FXAA. Perfect for me at 1680x1050!

 

[skipsneeky2]

Nice, do you have a chance to inspect the actual threads? Maybe even, if realtime, manually lower em' a notch (on the pentium emu of course)

Had Rivatuner OSD running,the cores were certainly pegged on the dual core.Had hell of a high load in both a emulated i3 and i5.Upper 90's usage.Maybe if i had a k series cpu with a overclock it would have been lower.

A spot where i would spawn and get like 80 fps,i would get about 29 with the emulated Pentium.Maybe a oced G3258 could do a bit better but i wouldn't dive into this game with one expecting much.The constant stutter and freezing pretty much ruins the experience.

 

[frozentundra123456]

Purepc.pl updated their tests with CPU graphs:
http://www.purepc.pl/karty_graficzne/jakie_sa_wymagania_sprzetowe_gta_v_test_wydajnosci?page=0,8

Its good to see fx8300 gives some benefit from the extra 2 threads compared to fx6300.

Getting an i5 gets a lot more benefit. Stock it is still 15% faster than an overclocked 8350 and will use a lot less power.

 

[2is]

When looking at the CPu usage the 2600K is loaded at 72.5% for the main cores while the 8350 is at 61%, the 20% higher loading of the former translate in 20% higher Fps, how surprising that a CPU that is underutilised will perform not as well as a one wich is better maxed out....

Problem with having weak cores, even if you have 8 of them is not all tasks can be split across multiple cores. So if you have a single thread being bottlenecked by a slow core, then all the other threads running on the other cores are just waiting for CPU to finish processing the main thread.

This is why the "just wait until games are optimized for AMD" theory that some people push around here since, well, seems like forever now, never has and never will come to fruition. More cores is great, but not at the expense of how powerful each core is.

 

[soccerballtux]

It might it might not,thats why anyone should show some proof with their claims.

that's how the WoW engine works, go look it up yourself

You can lock down the whole game to only one core,thats not really good,if you know a way to lock down individual threads to certain cores than please do share.
Using affinity will not lock down the threads,they will still be jumping around but only on the cores that you alow,that is the same thing that I said you should do.

Like I said, Process Lasso

I have nothing against cb,but in cb every core gets used 100% and you gain 15% ,games only use one core 100% so there is no way of a game having the same amount of gain.
I only said that you should test it and show how much difference it really makes.

wtf?

How am I spreading FUD?

already explained this

[quoteI don't have an FX so I can't test it on this or any other game,so I'm asking someone who has one to show us how much you could really gain from doing something like this.
If it is 15% then kudos.

like I said, CB is my go-to bench for testing total throughput because it efficiently scales across multiple cores. Because of the 15% improvement in CB, and knowing that games are less intense than CB, we can expect a significant improvement elsewhere, too. I just can't be bothered to test it. The 15% just confirms what I saw in other benchmarks elsewhere.