Crysis 3 CPU scaling, hyperthreading vs AMD cores

[JohnC]

Looks like 8 of the 12 cores are in use. Six of physical at roughly half and two of the logical as quarter or less. The listed 2600 has 6 of 8 cores in use. All four physical and again very little of the logical.

Isn't it still strange that the physical cores are only being loaded halfway?「(°ヘ°)

Hopefully someone will get Crytek to explain more fully, but I'd bet on plenty of threads and integer math.

I think you're probably right, but I still don't see why the Intel cores only get loaded partway. Seems strange... 「(°ヘ°)

Yes, it most certainly is. Mayhaps we should come up with some conspiracy theory involving AMD, Crytek and large sum of currency? I wonder if renaming "Crysis3.exe" will show the same results...

[Ryu Connor]

Isn't it still strange that the physical cores are only being loaded halfway?「(°ヘ°)

Remember that CPU utilization numbers are not a super accurate representation of resources expended.

Comparing Intel to AMD given the vastly different IPC and clockspeed of their designs is also a pretty serious apple to oranges situation.

[sschaem]

It seem clear that Crytek found HT to cause more problem then it help , and so only run their Cryengine3 on 1 HW thread per core.
The 2 extra workload is most likely OS & driver threads.

It also seem that a 3970k is saturating the GPU.

What this tells us is that the Crytek engine really cares about branching throughput.

And you will see more of that as game engine developers are going to tweak their engine for AMD architecture first and formost.
The same thing will happen with GPU code as GCN will be squeezed to its limits.

Same price chips ~$120

TestA)
i3-2110 - 18 min FPS
fx-6300 - 42 min FPS

Test B)
i3-3220 - 22 min FPS
fx-6300 - 41 min FPS

The GAP is massive in favor of AMD.
This is the reality of next gen game engine and the gap will only widen as future title get even more AMD architecture optimized.

From this, Kaveri might actually crush any and all haswell APU for gaming. specially when paired with a discreet GPU

I expect Intel price to drop prices big time, or Intel will loose the gaming segment.

[auxy]

This is the reality of next gen game engine and the gap will only widen as future title get even more AMD architecture optimized.
From this, Kaveri might actually crush any and all haswell APU for gaming. specially when paired with a discreet GPU
I expect Intel price to drop prices big time, or Intel will loose the gaming segment.

Eheh ... 八(＾□＾;;) Pardon me for saying so, but I think your remarks are ... ah ... a bit ... misguided, and perhaps ... that is ... well, optimistic, at best, really. Or possibly entirely dissociated from reality. ヘ（。□°）ヘ

One single atypical benchmark result does not an entire history of conventional wisdom refute, after all. I'm no Intel fangirl; historically quite the opposite, really, but games need single-threaded FP performance, traditionally, and I see no reason that should have changed. It seems much more likely we need to be poking Crytek to figure out why their game engine runs so weirdly compared to other games. (;￣д￣)ﾉ
^{Ahh, excuse me, Mr. Yerli...}

[flip-mode]

Pardon me for saying so, but I think your remarks are ... ah ... a bit ... misguided, and perhaps ... that is ... well, optimistic, at best, really.

I take back anything I ever said relating to yourself and the term "rude".

[JohnC]

Doom and gloom for Intel! Doom and gloom! They have failed in a single AMD-optimized benchmark = they are done for!
http://www.youtube.com/watch?v=DhS-i-c1XxI&sns=em

[flip-mode]

Doom and gloom for Intel! Doom and gloom! They have failed in a single AMD-optimized benchmark = they are done for!
http://www.youtube.com/watch?v=DhS-i-c1XxI&sns=em

Dude, you just played one of my favorite movie moments of all time. All time.

[JohnC]

Doom and gloom for Intel! Doom and gloom! They have failed in a single AMD-optimized benchmark = they are done for!
http://www.youtube.com/watch?v=DhS-i-c1XxI&sns=em

Dude, you just played one of my favorite movie moments of all time. All time.

Mine too!

[sschaem]

Eheh ... 八(＾□＾;;) Pardon me for saying so, but I think your remarks are ... ah ... a bit ... misguided, and perhaps ... that is ... well, optimistic, at best, really. Or possibly entirely dissociated from reality. ヘ（。□°）ヘ

One single atypical benchmark result does not an entire history of conventional wisdom refute, after all. I'm no Intel fangirl; historically quite the opposite, really, but games need single-threaded FP performance, traditionally, and I see no reason that should have changed. It seems much more likely we need to be poking Crytek to figure out why their game engine runs so weirdly compared to other games. (;￣д￣)ﾉ
^{Ahh, excuse me, Mr. Yerli...}

Take a look at the PS4 HW spec, and the speculation on the next xbox '720' HW spec to be highly similar.
None are based on high frequency/high IPC design, game engine designer are forced to optimize for many thread architectures.

Cryengine is just a prime example today of whats to come with the advance of next gen consoles.

Cryengine is not weird, its just able to spread the workload well across many cores and finally unlock the FX potential.

Also, games have been putting allot of weight on the CPU for floating point operations, but this is also changing. All in favor of AMD architecture.
There is a good reason why tools like z-zip and x264 run faster on a fx-8350 then a i7-3770k, you see the branching & integer compute potential.
Game designers will optimize x86 engine code for this model.

So yes, to date supporting more then 2 core didn't make much sense, but this has changed.

[I.S.T.]

Still, saying they're gonna optimize for "AMD architecture" is incredibly false. Jaguar is a far, far different beast compared to Piledriver/future and current incarnations of Bulldozer...

[JohnC]

Still, saying they're gonna optimize for "AMD architecture" is incredibly false. Jaguar is a far, far different beast compared to Piledriver/future and current incarnations of Bulldozer...

No, no, let him dream! Dream about a new world, a far better world - one in which the eternal dignity of AMD is respected!

[Airmantharp]

Still, saying they're gonna optimize for "AMD architecture" is incredibly false. Jaguar is a far, far different beast compared to Piledriver/future and current incarnations of Bulldozer...

Yes, but it's still a hell of a lot closer than Atom is to Ivy, or the PowerPC CPUs in the PS3 and 360 were to the Xbox's Celeron or other CPUs at the time. Just having eight out-of-order x86 cores is a huge advancement and parallels quite nicely with the PC CPU world.

[cynan]

Looks like 8 of the 12 cores are in use. Six of physical at roughly half and two of the logical as quarter or less. The listed 2600 has 6 of 8 cores in use. All four physical and again very little of the logical.

Isn't it still strange that the physical cores are only being loaded halfway?「(°ヘ°)

Hopefully someone will get Crytek to explain more fully, but I'd bet on plenty of threads and integer math.

I think you're probably right, but I still don't see why the Intel cores only get loaded partway. Seems strange... 「(°ヘ°)

Yes, it most certainly is. Mayhaps we should come up with some conspiracy theory involving AMD, Crytek and large sum of currency? I wonder if renaming "Crysis3.exe" will show the same results...

I have a 3930k and, other than a few benchmarks and stress testing applications (ie, Prime95 maxes out CPU usage), rarely see CPU usage approach anything near 100% - even when encoding media or processing photos with multithreaded applications. It irks me to no end running statistical number crunching (ie, bootstrapping larger data sets), telling the software that I have 12 cores, and yet having to wait while most cores are pegged at ~25% (entering 6 cores instead of 12 doesn't help). I was hoping Windows performance monitor was just not reporting accurately, but I'm not so sure.

And yes, this occurs even with core parking disabled

[I.S.T.]

Still, saying they're gonna optimize for "AMD architecture" is incredibly false. Jaguar is a far, far different beast compared to Piledriver/future and current incarnations of Bulldozer...

Yes, but it's still a hell of a lot closer than Atom is to Ivy, or the PowerPC CPUs in the PS3 and 360 were to the Xbox's Celeron or other CPUs at the time. Just having eight out-of-order x86 cores is a huge advancement and parallels quite nicely with the PC CPU world.

True, very true, but there's still... well, not a world of difference but perhaps a few seas worth?

[auxy]

I have a 3930k and, other than a few benchmarks and stress testing applications (ie, Prime95 maxes out CPU usage), rarely see CPU usage approach anything near 100% - even when encoding media or processing photos with multithreaded applications. It irks me to no end running statistical number crunching (ie, bootstrapping larger data sets), telling the software that I have 12 cores, and yet having to wait while most cores are pegged at ~25% (entering 6 cores instead of 12 doesn't help). I was hoping Windows performance monitor was just not reporting accurately, but I'm not so sure.

And yes, this occurs even with core parking disabled :P

Have you tried using 6 threads instead of 12? DID YOU EDIT YOUR POST?! I SWEAR THAT WASN'T THERE! ヽ(≧Д≦)ノ
I'm well aware of how CPU usage reporting works (it's reported by sampling over a whole second, but most activities take milliseconds or microseconds to complete), but it still seems like you should be able to do a lot better than 50% per core with almost any given task. (¬_¬) Of course, assuming you aren't GPU-, RAM-, or I/O-limited.

[Airmantharp]

Still, saying they're gonna optimize for "AMD architecture" is incredibly false. Jaguar is a far, far different beast compared to Piledriver/future and current incarnations of Bulldozer...

Yes, but it's still a hell of a lot closer than Atom is to Ivy, or the PowerPC CPUs in the PS3 and 360 were to the Xbox's Celeron or other CPUs at the time. Just having eight out-of-order x86 cores is a huge advancement and parallels quite nicely with the PC CPU world.

True, very true, but there's still... well, not a world of difference but perhaps a few seas worth?

I did go digging, but a quick search didn't highlight just what makes Jaguar so much different, other than the obvious orientation towards low-power devices using a 28nm process. It's no Ivy, but I'm betting it will hold it's own.

[BoilerGamer]

Minium frame rate show the fx-8350 stomping on the i5-3570k, and beating an i7-3770k.

If the fx-8350 and i7-3770k where the same price I would get the i7-3770k, no question.
But we are talking about 180$ vs 340$ difference.

I think CryEngine3 paint the picture pretty well of whats to come, Intel will need to dramatically lower their prices to appeal to gamers.

Intel can do that with ease if they deem it necssary, Do not forget it takes a silicon die twice as large to produce FX-8350(315mm^2) compared to 3770K (160mm^2 with about 1/3 taken up by the IGPU) which is also sold at almost twice the price.

If these benches shows Intel that it it can no longer rely on hyperthreading, it could destory AMD in price wars(massive difference in production cost) or introduce a 6 core mainstream CPU with Broadwell( die shrink can mean either more cores or lower TDP, Ivy was lower TDP because AMD haven't caught up to Intel for ages, if Broadwell were BGA only Intel have even more room to tweak it).

Also, the i7-3820 successor in the IB-E class could easily have 2 more cores with the die shrink, same goes for the higher end IB-Es have room to become 8 cores due to the dieshrink (Intel added two cores during the Bloomfield-> Gulftown die shrink). Intel could have pushed all these out last year, but it was not necessary at the time so they pushed IB-E back a year. Now that AMD finally offered a challenge in gaming performance, I am sure Intel wouldn't hesitates to respond in kind.

[auxy]

I did[n't?] go digging, but a quick search didn't highlight just what makes Jaguar so much different, other than the obvious orientation towards low-power devices using a 28nm process. It's no Ivy, but I'm betting it will hold it's own.

Jaguar is based on the low-power "Bobcat" design used in the E-350/450 series APUs. It's got fairly poor IPC, owing to its low-power roots; Bobcat is somewhat faster than Atom, and Jaguar is supposed to be a 15% improvement on that, so it's a little difficult to say concretely what it can do.

Jaguar has no relation at all to Bulldozer/Piledriver design. It doesn't use modules (having discrete, complete cores, like an Intel design), it doesn't support FMA, and yet it supposedly is a lot more focused on using SIMD/vector processing to achieve its peak performance.

[sschaem]

Just based on Crysis3, I dont see Intel lowering prices, even if they easily could, to a level where their CPU become competitive for gaming PCs.

Just looking at Crysis3, AMD seem to have a 30% advantage in price performance at the i3 level.
So for Intel to offer a better gaming performance at this level, the i3-3220 will need to drop below $90
And the i7-3770k will need to drop below $190.

Intel will not murder its profit margin just for gamers. Going from $340 to $190 would cut their margin by ~50%, and would destroy the stock valuation.

[obscureangel]

I know that the game was the AMD evolved, which means that they may work closely and optimized the game for take advantange of Multi-threading, which AMD is actually very good on FX.

I have been investigating why FX sucks, and i found that the games that don't take advantage of Multi-threading, they don't take advantage of the 8 cores of the CPU, and that its the problem.

I had find sometime ago (unfortunally dont know where is the link), the multiplayer benchmark of BF3 where FX8350, performed much better than the i5 3570k because the frostbite supports HT.
Even the i7 3770k recieves a massive performance boost from this games that support HT.

And currently the engines that takes advantage of HT are the frostbite, the metro 2033 engine, source engine, and Cryengine 3
This engines is where AMD FX and I7 takes a good improvement on performance, not sure about far cry 3, didnt read about it.

Just registered to inform you about that
Regards

[Bensam123]

I have a 3930k and, other than a few benchmarks and stress testing applications (ie, Prime95 maxes out CPU usage), rarely see CPU usage approach anything near 100% - even when encoding media or processing photos with multithreaded applications. It irks me to no end running statistical number crunching (ie, bootstrapping larger data sets), telling the software that I have 12 cores, and yet having to wait while most cores are pegged at ~25% (entering 6 cores instead of 12 doesn't help). I was hoping Windows performance monitor was just not reporting accurately, but I'm not so sure.

And yes, this occurs even with core parking disabled

Nothing is made for more then four cores because four cores is mainstream as I said before. People don't design software for hardware that doesn't exist. Servers have plenty of applications that take advantage of more then four cores and they also have plenty of chips that have more then four cores. A hexacore is still a outlier and it will be till they come down in price. Hexacores should've been priced in the 300 dollar range and should've been a meaningful difference between the i5s and i7s. People would actually contemplate buying i7s then.

It's not just about the workload presented with one application, but the overall system workload. W7 and W8 are fairly good at distributing resources and it's entirely possible to use all threads and the entire quad or hexa-core processor based on the system workload, regardless of how well one application is multithreaded. A good example of this is streaming, which I do. It definitely changes your average system workload.

I'm still sorta baffled that AMD didn't drop down any of their server processors to the desktop arena to simply push Intel out with more cores. Although Intel could've started selling 8 core chips due to the two disabled cores on their hexas, it definitely would've changed things. Intel may have been unwilling to do that too due to sacrificing part of their server market. AMD doesn't have such a huge server market and 12 and 16 core chips are just PD's with more cores welded together.

[JohnC]

I had find sometime ago (unfortunally dont know where is the link), the multiplayer benchmark of BF3 where FX8350, performed much better than the i5 3570k because the frostbite supports HT.

That's... not entirely correct, my dear friend



This is not a "multiplayer" benchmark, though, but it's almost impossible to properly do a benchmark of multiplayer portion of such games due to ever-changing number (and location/action) of live players...

[ronch]

I'm seeing quite a bit of discussion on this thread regarding PS4's use of eight Jaguar cores. And it's quite strange why Sony (and perhaps Microsoft too) would pick Jaguar to power their next-gen consoles. Now, I don't know how much faster each Jaguar core used in the PS4 is compared to each of Sony's 'Cell' core in the PS3, or how much faster a Jaguar core used in the Xbox Infinite is (Sony and MS may run their Jag cores at different clocks, or may have some custom design mixed in) compared to one of the Xbox 360's Xenon (or Xenos) CPU cores, but Jaguar obviously isn't gonna break any world records, not now, not 5 years from now. So you have to wonder just what these console makers are thinking.

I'm siding with predictions here that games will increasingly be multi-threaded. Given the graphics muscle each of the next-gen consoles offer, Jaguar obviously isn't gonna cut it with just a few cores driving a beefy GPU to run a high-power title. Given this, game developers are obviously being encouraged (or forced) to use as many cores as they can (whether they like it or not), not just to drive the performance of their particular game for either console, but to push better coding practices and techniques in general to use more and more cores, which will obviously help PC gaming as well.

Now, helping PC gaming obviously isn't on Sony's agenda, but it obvious is for MS, which has a vested interest in PCs in general. So why would Sony use Jaguar just like MS (purportedly) will? It's x86 compatibility. Perhaps the game devs have been crying out to console makers to make it easier for them to port games between platforms, and going x86 all the way will obviously help make everyone's lives easier. Game titles, especially the big ones, are by no means cheap to develop, and tediously porting each title just so it could run on a particular console isn't the most sensible thing in the world. And people couldn't care less too about which CPU ISA is used inside their console or PC. As long as games run fine, they're ok with it. The odd man out here would be Nintendo which is still relying on PowerPC. If Wii U doesn't sell well, I'm not sure game devs will be tempted to port their games for the Wii U especially if they're getting enough sales from the other systems. Then again, that's just my opinion.

Moving forward, I think we'll see more and more titles pushing the FX chips out there. Intel may well counter this trend by adding more cores to negate the FX's advantage. Given their process lead it's probably gonna be easier for Intel to add more cores than AMD. As AMD and everyone knows, they can't fend Intel off forever so they're starting to play in adjacent markets now. Who knows, 20 years from now AMD may be selling Cola.

Looks like AMD is again paving the way for Intel to push technology further. It happened with AMD64, HyperTransport, multi-core x86 CPUs, FMAx, etc. that in the end, will still put AMD in second place. What AMD can do, Intel can do too, with far more R&D and marketing money. It's the reality, folks. The best we can do is buy AMD products but it can only help so much.

[Airmantharp]

I don't think it's the ISA that challenges developers as much as the part you highlighted about coding practices- it's how they structure the engine that matters. The ISA will factor into some decisions no doubt, but the biggest challenge has been to force developers to really break their workloads up.

[I.S.T.]

I'm seeing quite a bit of discussion on this thread regarding PS4's use of eight Jaguar cores. And it's quite strange why Sony (and perhaps Microsoft too) would pick Jaguar to power their next-gen consoles. Now, I don't know how much faster each Jaguar core used in the PS4 is compared to each of Sony's 'Cell' core in the PS3, or how much faster a Jaguar core used in the Xbox Infinite is (Sony and MS may run their Jag cores at different clocks, or may have some custom design mixed in) compared to one of the Xbox 360's Xenon (or Xenos) CPU cores, but Jaguar obviously isn't gonna break any world records, not now, not 5 years from now. So you have to wonder just what these console makers are thinking.

I'm siding with predictions here that games will increasingly be multi-threaded. Given the graphics muscle each of the next-gen consoles offer, Jaguar obviously isn't gonna cut it with just a few cores driving a beefy GPU to run a high-power title. Given this, game developers are obviously being encouraged (or forced) to use as many cores as they can (whether they like it or not), not just to drive the performance of their particular game for either console, but to push better coding practices and techniques in general to use more and more cores, which will obviously help PC gaming as well.

Now, helping PC gaming obviously isn't on Sony's agenda, but it obvious is for MS, which has a vested interest in PCs in general. So why would Sony use Jaguar just like MS (purportedly) will? It's x86 compatibility. Perhaps the game devs have been crying out to console makers to make it easier for them to port games between platforms, and going x86 all the way will obviously help make everyone's lives easier. Game titles, especially the big ones, are by no means cheap to develop, and tediously porting each title just so it could run on a particular console isn't the most sensible thing in the world. And people couldn't care less too about which CPU ISA is used inside their console or PC. As long as games run fine, they're ok with it. The odd man out here would be Nintendo which is still relying on PowerPC. If Wii U doesn't sell well, I'm not sure game devs will be tempted to port their games for the Wii U especially if they're getting enough sales from the other systems. Then again, that's just my opinion.

Moving forward, I think we'll see more and more titles pushing the FX chips out there. Intel may well counter this trend by adding more cores to negate the FX's advantage. Given their process lead it's probably gonna be easier for Intel to add more cores than AMD. As AMD and everyone knows, they can't fend Intel off forever so they're starting to play in adjacent markets now. Who knows, 20 years from now AMD may be selling Cola.

Looks like AMD is again paving the way for Intel to push technology further. It happened with AMD64, HyperTransport, multi-core x86 CPUs, FMAx, etc. that in the end, will still put AMD in second place. What AMD can do, Intel can do too, with far more R&D and marketing money. It's the reality, folks. The best we can do is buy AMD products but it can only help so much.

Well, Intel already was planning FMA3 when AMD announced FMA4... Hell, they semi-worked together on the whole thing.

As for why Sony picked Jaguar? Simple. Higher performance compared to Cell in most if not all things at a lower power draw and, more importantly, lower space. Jaguar cores are incredibly small, and this is why the PS4 will be running an APU instead of two separate chips. It lowers manufacturing costs out of the gate.

[ronch]

I don't think it's the ISA that challenges developers as much as the part you highlighted about coding practices- it's how they structure the engine that matters. The ISA will factor into some decisions no doubt, but the biggest challenge has been to force developers to really break their workloads up.

To recap, I said this:

I'm siding with predictions here that games will increasingly be multi-threaded. Given the graphics muscle each of the next-gen consoles offer, Jaguar obviously isn't gonna cut it with just a few cores driving a beefy GPU to run a high-power title. Given this, game developers are obviously being encouraged (or forced) to use as many cores as they can (whether they like it or not), not just to drive the performance of their particular game for either console, but to push better coding practices and techniques in general to use more and more cores, which will obviously help PC gaming as well.

I was talking about programmers being increasingly forced to use more and more cores. Nothing to do with ISA in this regard. The only time I talk about ISAs is when I said Nintendo is the odd man out here, being the only major gaming platform to use PoewrPC. The rest (PC, Xbox Infinity and PS4) all use x86. This obviously gives devs more work to make their games run on Wii U.

[Meadows]

Given the graphics muscle each of the next-gen consoles offer, Jaguar obviously isn't gonna cut it with just a few cores driving a beefy GPU

And this is based on what bad dream you were having last night?

[ronch]

@Meadows : Given early indications that the PS4's GPU will be something like a high end GCN-class part, and performance estimates putting Jaguar just 10% faster than Bobcat, I stand by my reckoning that unless you use all eight of those Jaguat cores to drive that GPU, the GPU will just end up begging to be fed. Or are you ok with pairing a nice HD7870 with an E-350-based board?

[Meadows]

@Meadows : Given early indications that the PS4's GPU will be something like a high end GCN-class part, and performance estimates putting Jaguar just 10% faster than Bobcat, I stand by my reckoning that unless you use all eight of those Jaguat cores to drive that GPU, the GPU will just end up begging to be fed. Or are you ok with pairing a nice HD7870 with an E-350-based board?

We're talking about 720p-1080p gaming at framerates between 30 and 60. The CPU is sufficient, so I fail to see the issue. The developers can finally add detail and effects without compromising framerate, and the extra CPU cores should result in more in-game physics and better AI.

[Chrispy_]

I wouldn't be so sure about that. The E-350 I had was disappointingly slow - slow enough at 1280x720 to be occasionally CPU-limited in even quite old games, despite the fact that it had all the graphics on minimum and I know that a similarly specced HD4550 used to fly along. I think it was most noticable at a retro-LAN where I dug it out for service playing UT2003, Quake3, Starcraft Brood War and a bit of WC3 tower defense.

It harks back to the fact that Atom + ION netbooks/nettops were always held back by the Atom, rather than the GPU; Whilst clearly superior to Atom, Bobcat/Brazos cores weren't the order-of-magnitude faster like the big boys (Nehalem or Stars cores) - 60-80% higher IPC than an Atom or something like that IIRC.