Jaguar CPU in XBox and PS4 *spawn*

I think a game CPU intensive as the last patched Prey is much more usefull in benchmarking overall system performances... Pure calc power doesnt speak about GPU/CPU/RAM latencies of a single RAM pool system as todays consoles

 

Westmere is the same microarchitecture as Nehalem. At Anandtech, a 3.46Ghz Westmere scores 117, a 2.66Ghz Nehalem will score around 90 or 33.8/Ghz vs 21/GHz for Jaguar cores, 62% higher IPC. If you take a different benchmark, like Kraken, you get close to twice the IPC for the 920.

Cinebench is almost completely immune to main memory bandwidth. Increasing clock doesn't change the IPC.

Cheers

 

Many people say what CPUs in Xbox One and PS4 is weak, but really can anyone tell why. Because all games work, and work very good. And also some years ago many peoples said what Xbox 360 have weak CPU, and PS3 have even weaker CPU, because it have only one main core, but games also worked very good.

 

They are weak compared to modern CPUs. Compared to last gen In-order CPUs there are fast, even if they don't have the peak giga-bollocks of the PPUs/SPUs of last gen.

Games made for consoles work well, because you architect your code to take advantage of the many cores and rely less on single thread performance.

Games made for PC without any considerations for the console might suck when ported because of the much lower single thread performance. The best example of this is PUBG where the game struggles to even hit 30 fps.

Cheers

 

But bassicaly that's it. Other games work very good. And thank you for explanation.

 

Did you even read what i posted?
Also got data for your kraken? because the best i can find (kraken 1.0) is 6627ms for 920 and 8990 for 5150.

i wasn't only talking about Cinebench

 

Anandtech, same place you got the Cinebench results:
Westmere/I7 990X 3.46GHz: 1338, Kabini/5150, 1.6GHz: 5664. Westmere 86% higher IPC.

I can't find anything where the 5150 is close to a 920 in IPC, let alone outright performance.

Cheers

 

It's worth pointing out that the consoles have 2 cores disabled for games since launch, and as you add more and more cores to a games workload you get dimishing returns. A game can only be multithreaded to a point at least currently.

Some games still lean on just a couple of cores even now.

In the base consoles at least there's 0 competition between them and i5 750. X1X might be the exception, with the fastest clocks and ipc improvements.

 

That's not how single-thread scores work..

Single-threaded score wouldn't double with increased core count at ISO clocks.

 

Lol, thanks I must have experienced a senior moment

 

If you take into account clocks (2.66 base and 2.93 single threaded turbo), there do seem to be some examples where i7 920 has < 50% IPC advantage over 5150.

https://www.anandtech.com/bench/product/1224?vs=47

After dividing by clock speed (gHz, base or turbo depending on bench) you get:

z(Win7) 3D Particle Movement: Single Threaded (assuming 2.93)
Score (Higher is Better)
5150: 25.95
i7 920: 31.03
x1.2

z(Win7) 3D Particle Movement: MultiThreaded
Score (Higher is Better)
5150: 85.00
i7 920: 182.11
x2.14

z(Win7) Cinebench R10 - Single Threaded Benchmark (assuming 2.93)
Score in CBMarks - Higher is Better
5150:1001
i7 920: 1312
x1.31

z(Win7) Cinebench R10 - Multi-Threaded Benchmark
Score in CBMarks - Higher is Better
5150: 3443
i7 920: 6094
x1.77

z(Win 7) x264 HD Benchmark - 1st pass - v3.03 (I'm assuming this is multithreaded so 2.66)
Frames per Second - Higher is Better
5150: 18.74
i7 920: 24.47
x1.31

z(Win 7) x264 HD Benchmark - 2nd pass - v3.03
Frames per Second - Higher is Better
5150: 5.50
i7 920: 9.70
x1.76

z(Win7) 7-zip Benchmark (Win7)
32MB Dictionary - Total MIPS - Higher is Better
5150: 2957
i7 920: 6121
x2.07

Not being able to use hyperthreading hurts the Intel cores in the single threaded tests, and there are examples of the 920 being as little as 20~30% faster when scaled for clock speed (I know it won't be truly linear scaling with frequency, but close enough).

With 8 threads the 920 simply leaves the 5150 for dust. With two complete quad modules in the consoles and oodles of BW, there may well be circumstances where the consoles can edge out the i7 920 however.

This is just based on the comparisons available. There may be tasks where the advantages of Intel's architecture opens up much wider gaps, so I wouldn't assume these comparisons hold across the board for all workloads.

 

I have a notebook with i7 920XM and a little KangarooPC portable with Atom X5 Z8500 (Cherrytrail Quad 2.24GHz Turbo). The 920XM is ridiculously faster than that Atom and isn't Cherrytrail faster per clock than Jaguar? 920XM is so much more pleasant to use that I feel it's silly to put them in the same comparison.

 

Single threaded comparisons won't change, multi-threaded performance may be up to double. Meaning there are probably instances where - for the same clock speed - the console arrangements may edge out the 920. With *actual clocks* taken into account the 920 will beat consoles even at heavily multithreaded stuff, most likely even the X1X (though some workloads may be quite close).

Single threaded, the results from Anandtech show an advantage of as little as 20~30% for an i7 920 at the same clock speed. But single threaded doesn't benefit from hyperthreading on the Intel cores, which potentially gets you another ~25%. But you can make the case that if you're going to factor in hyperthreading, you should really compare one Intel core to two Jaguar cores...

That is simply going by these (limited) results. As I mentioned earlier, there may be situations where the Intel architecture can stretch its legs further.

 

Performance wise, probably in a similar ballpark. Even at the same process node, I'd wager any "big" Intel core (Bloomfield, Sandy Bridge, Ivy Bridge) to be as big a Jaguar cluster (4 cores), which is exactly why Bulldozer modules made no sense. Jaguar was simply the best solution AMD had for the given die area, power draw and TDP.

Effectively these big wide cores are giving you double the IPC, running twice the clock speed, hence why older dual core Core i3's can generally match and even exceed the base PS4 and Xbone. But they draw more power in total as a result. Newer Core i3s are even better, but the consensus these days is to have quad-cores to guarantee enough CPU side performance for a consistent 60 FPS with the proper GPU, as most multiplatform games are built around 30 FPS on console. And of course you have the clock bumps on the PS4 Pro and XB1X.

 

Cinebench tends to be very forgiving with memory for example; often games are not,
I wonder how the PS4 CPU would look in this regard, high bandwidth but with a high performance GPU taking priority, likely very high latency? also lack of l3 cache (which on PC typically means a significant performance penalty for gaming performance, going by the AMD CPUs that had version with l3 and no l3 cache), also isn't the PS4 limited to use less than 8 cores for gaming? and isn't the dual module nature going to incur some penalty to scaling compared to the 5150 with 4c?

another test, geekbench with the 920 and the 5150 (I looked for some non OC, average looking result from the first page)
https://browser.geekbench.com/v4/cpu/compare/9057210?baseline=8922732

have to keep in mind that the i7 920 boosts to 2.93GHz for single thread (but how constantly it sustains that on first gen i7s?), while the Athlon is locked at 1.6GHz and obviously the memory performance is very different from the consoles.

for a PC I think 1.6GHz 4c/8t 920 still makes more sense than 1.6 Jaguar with 8c, I think the l3 cache is a big help, and the single thread performance advantage should still be pretty significant, Jaguar is slower than Core 2 Duo at the same clock on many tests.