XB1 vs. PS4: Which is most powerful? *dead thread

[AzBat]

Oh boy. Interesting...

• 18 CU's vs. 12 CU's =/= 50% more performance. Multi-core processors have inherent inefficiency with more CU's, so it's simply incorrect to say 50% more GPU.
• Adding to that, each of our CU's is running 6% faster. It's not simply a 6% clock speed increase overall.
• We have more memory bandwidth. 176gb/sec is peak on paper for GDDR5. Our peak on paper is 272gb/sec. (68gb/sec DDR3 + 204gb/sec on ESRAM). ESRAM can do read/write cycles simultaneously so I see this number mis-quoted.
• We have at least 10% more CPU. Not only a faster processor, but a better audio chip also offloading CPU cycles.
• We understand GPGPU and its importance very well. Microsoft invented Direct Compute, and have been using GPGPU in a shipping product since 2010 - it's called Kinect.
• Speaking of GPGPU - we have 3X the coherent bandwidth for GPGPU at 30gb/sec which significantly improves our ability for the CPU to efficiently read data generated by the GPU.

http://www.neogaf.com/forum/showpost.php?p=80951633&postcount=195

I bolded that part above. I have a feeling some of you will disagree though.

Tommy McClain

 

[McHuj]

Oh boy. Interesting...


http://www.neogaf.com/forum/showpost.php?p=80951633&postcount=195

I bolded that part above. I have a feeling some of you will disagree though.

Tommy McClain

I don't know. Simultaneous read/write is the only way close to 200+ GB/sec is possible in the eSRAM.

The transistor count of the chip doesn't seem to add up to me. 5B, imo, is too much for the CPU, GPU, eSRAM, even with the additional processors. While Tensillica cores offer a lot bang, there not expensive in terms of transistor count. IMO, the only way you get up to 5B transistors is with a lot of redundancy and/or a dual ported (or some hybrid eSRAM).

 

[3dilettante]

I wish we could get the words straight from the technical fellow at Microsoft he consulted with.
The bullet point summary is an endpoint for which there are many paths one can take to get to.

The audio processor's superiority is an assumption at this point, although it looks very, very likely.

I don't know about the coherent bandwidth number, however. The numbers we've seen leaked for Sony are unidirectional. Perhaps Durango really is 30 in both directions, although up to this point I've interpreted the Vgleaks data to be 30 GB/s in total. If not, the advantage is more like 50% instead of 300%.
To add another quibble, it's not clear from Vgleaks that the GPU's link to the northbridge has as much bandwidth. This is another case where we need more precision to know exactly what each side is looking at when they get their numbers, which PR often gets wrong even if they try.

 

[XpiderMX]

each of our CU's is running 6% faster. It's not simply a 6% clock speed increase overall

Can 12 CUs "do the same" than 18 CUs if each CU from the 12 is 6% faster than the 18CUs?

 

[Xenus]

Yeah it seems like that entire sentace is math fail. Unless he's trying to say only the CU's clock is increased not the rest of the GPU which is actually worse. The problem I think lies in people feeding him information he doesn't understand. So the context is entirely lost.

 

[blakjedi]

Can 12 CUs "do the same" than 18 CUs if each CU from the 12 is 6% faster than the 18CUs?

The xb1 gpu would have to be clocked at 1.2 ghz to do equiv operations so no.

 

[MrFox]

each of our CU's is running 6% faster. It's not simply a 6% clock speed increase overall

So... that would be 6% x 12CU, it's 72% increase !!!! WOW !!! Maths are overrated.

 

[eastmen]

The xb1 gpu would have to be clocked at 1.2 ghz to do equiv operations so no.

well only the cu's would have to be clocked at 1.2ghz do to equiv operations and that's if they are the same cu's

 

[temesgen]

The math in the quote seems fine, or at least most of it is accurate--if not always very precise. (edit: For clarity, it's not the numbers, but the context of the data that we have to make assumptions about.)
The quibbly part I have is the 3x coherent bandwidth claim, mainly because I don't know which sources he's drawing from. The leaks we've seen give Durango an edge, but it's possibly 25-50% instead of 300%.
Perhaps that's something that wasn't leaked or not described fully, but it's a bigger gap than anything else mentioned.

My issue with the math is he seems to minimizing the 50% difference in CUs and then 3 bullet points later saying we have 10% more CPU...

In the middle of this cognitive dissonance he says multi core processors have inherent inefficiency so he seems to be wanting his cake and eating it too with the math.

Don't really see how you can say a 50% difference is not a big deal then turn right around and emphasize a 10% difference. Again I don't really have a problem with comments but they don't seem to be very coherent. I think he would be better served by focusing on the vision and less on specs.

Further if MS is going to keep stating the bandwidth advantage they should give some indication how they derive their numbers. I think its reasonable to assume their peak is what they say it is but how likely are developers to hit this on real life scenarios? The move engines probably help in this regard but what is a good average? Surely its >68gb/sec but what is a good number? We don't have any idea and since MS has been happy to provide specs that are equal or show themselves in a favorable light, its fair to be somewhat intrigued by all of this.

 

[3dilettante]

My issue with the math is he seems to minimizing the 50% difference in CUs and then 3 bullet points later saying we have 10% more CPU...

The debate shifts depending on where the bottlenecks go. My first uncertain glances at the architecture were at the choice of CPU and the non-native octo-core setup.
The GCN setup looks to be taking a design exactly where it is able to go, whereas the CPU modules show traits that show they may not have had this scenario in mind when they were conceived.
Mileage may vary.

Again I don't really have a problem with comments but they don't seem to be very coherent. I think he would be better served by focusing on the vision and less on specs.

I believe he is responding to a rather incoherent and spec-obsessed narrative.

Further if MS is going to keep stating the bandwidth advantage they should give some indication how they derive their numbers. I think its reasonable to assume their peak is what they say it is but how likely are developers to hit this on real life scenarios? The move engines probably help in this regard but what is a good average? Surely its >68gb/sec but what is a good number? We don't have any idea and since MS has been happy to provide specs that are equal or show themselves in a favorable light, its fair to be somewhat intrigued by all of this.

I don't think the move engines have much to say here.
One thing I noted back before it was "discovered" that the eSRAM's bandwidth had a higher peak than 102 GB/s was that even then I thought it might have some nice facets of consistent performance versus a DRAM bus whose load-balancing abilities for the bandwidths in question is a question mark. The upclock and bonus 30% achieved bandwidth are icing on the cake to me.

 

[3dilettante]

I'm going to note that while a Microsoft spokesperson technically started the versus discussion, I think the discussion has gone to the bad place.

I do think at some point there is worthwhile discussion to be had, if we can get enough information to make well-founded comparisons.
However, until then, I worry that we've managed to make a versus comparison, argued about sources, and ascribed relevance to GAF's feelings in the technical thread--a thread I've grown fond of.

 

[Ekim]

Can 12 CUs "do the same" than 18 CUs if each CU from the 12 is 6% faster than the 18CUs?

12 CUs can even do more without being 6% faster when having a low-latency embedded memory block and no optimization to prevent lots of cache misses on the 18CU machine.