PS4 & XBone emulated?

[homerdog]

For the new consoles, the GPU side is like 8x-10x better than last gen, there is 8x the RAM, but I don't think there is anything close to an 8x increase in CPU power. 2-2.5x maybe?

Xbox One should have much more than a 2.5x advantage in CPU performance vs the Xbox 360. The badness of the Xbox 360 CPU is often understated. It is bad.

A single 1.75GHz Jaguar core is likely preferable to the entire 360 CPU.

 

[Corngood]

Xbox One should have much more than a 2.5x advantage in CPU performance vs the Xbox 360. The badness of the Xbox 360 CPU is often understated. It is bad.

A single 1.75GHz Jaguar core is likely preferable to the entire 360 CPU.

If you can load the 360 CPU up with hazard-free vector code it's extremely fast. I don't see how jaguar could get near it. It's only really bad with unoptimised spaghetti code. I'm not saying I prefer working with it, but it's pretty decent with the right workload (e.g. skinning, culling, physics.)

 

[pjbliverpool]

If you can load the 360 CPU up with hazard-free vector code it's extremely fast. I don't see how jaguar could get near it. It's only really bad with unoptimised spaghetti code. I'm not saying I prefer working with it, but it's pretty decent with the right workload (e.g. skinning, culling, physics.)

Don't 6 jaguar cores have similar theoretical vector throughput to the 3 Xenon cores anyway? Thus the actual throughput should be similar unless you're saying that the VMX units in Xenon are actually more efficient than the AVX units in the jaguars?

 

[Corngood]

Don't 6 jaguar cores have similar theoretical vector throughput to the 3 Xenon cores anyway? Thus the actual throughput should be similar unless you're saying that the VMX units in Xenon are actually more efficient than the AVX units in the jaguars?

Yeah, they are probably theoretically similar. Somewhere within 1-2x in jaguar's favour, I imagine, because of the huge clock difference.

I just think it's an exaggeration to say: "A single 1.75GHz Jaguar core is likely preferable to the entire 360 CPU."

 

[pjbliverpool]

Specifically the 3 VMX units in Xenon can output 76.8 GFLOPs on paper while 6 Jaguar cores at 1.75Ghz could push 84 GFLOPs. So yeah, 1 Jaguar CPU core at 14 GFLOPs wouldn't come close to Xenon in vector code.

 

[cal_guy]

Specifically the 3 VMX units in Xenon can output 76.8 GFLOPs on paper while 6 Jaguar cores at 1.75Ghz could push 84 GFLOPs. So yeah, 1 Jaguar CPU core at 14 GFLOPs wouldn't come close to Xenon in vector code.

There's is one VMX unit in each Xenon core with 2 execution ports. The first port deal contains a 128-bit vector FMA unit. The second pipe is for permute and load instructions. So each core is capable of 25.6 GFLOPS.

 

[Pixel]

They mention 1 CPU. Not sure if this is a per core basis, or per cluster (which only really applies to PS4/X1), or all cores. These figures include varying CPU overhead so take that into account.

http://gamingbolt.com/substance-eng...eneration-speed-to-14-mbs-12-mbs-respectively

 

[homerdog]

Specifically the 3 VMX units in Xenon can output 76.8 GFLOPs on paper while 6 Jaguar cores at 1.75Ghz could push 84 GFLOPs. So yeah, 1 Jaguar CPU core at 14 GFLOPs wouldn't come close to Xenon in vector code.

Wasn't really referring to peak Gflops when I made that statement. Cell has a terrifically high peak Gflops rate and I would definitely say that a single Jaguar core is preferable to Cell as a gaming CPU.

 

[DavidGraham]

No one is denying that a jaguar core is preferable to either the Xenon or the Cell, for on the basis of Programmability alone it holds the complete advantage. however Xenon and Cell are fast on their own with optimized code, they already compete with multi-core Jaguar on most metrics, that's the whole point.

 

[fellix]

Heavy FLOPs are already heading for the domain of the GPU compute. The choice of Jaguar is a compromise between the need of fast serial performance (integer and load/store tasks) and low-cost SoC-type of direct macro-block integration. HSA will iron out the critical CPU-GPU path (the system memory).

 

[homerdog]

No one is denying that a jaguar core is preferable to either the Xenon or the Cell, for on the basis of Programmability alone it holds the complete advantage. however Xenon and Cell are fast on their own with optimized code, they already compete with multi-core Jaguar on most metrics, that's the whole point.

I understand and agree.

 

[Corngood]

Wasn't really referring to peak Gflops when I made that statement. Cell has a terrifically high peak Gflops rate and I would definitely say that a single Jaguar core is preferable to Cell as a gaming CPU.

Are you saying you'd rather have a single Jaguar core at 1.8Ghz vs. the entire 3.2Ghz Cell from PS3? I can't agree with that. I know it's not all about flops, but there's like a 10x difference.

Maybe I'm misunderstanding you. I guess if your plan is to do the SPU work on a GPU, then you may have a point, but you're still losing out on a couple hundred GFLOPS overall.

 

[homerdog]

Yes, it depends on what GPU you're working with. With a 7850 riding shotgun we'd prefer the Jaguar, right?

 

[Corngood]

I'd still take the cell. It's fast as hell when you get stuff running well on the SPUs. Remember most modern games use a bunch of SPU time to do FSAA, vertex processing etc. to make up for the weak GPU.

I'd probably even take cell over 2 jaguar cores, 4 would tip it in jaguar's favour.

That's just my opinion, having optimised quite a few games for cell, but none for jaguar (yet.) I might have changed my mind in a year.

Why are we talking about this again? I feel off-topic guilt.