Penello specifies why X1vsPS4 performance difference is overstated

[Zappykins]

petalpusher said:

Adinnieken said:

petalpusher said:

That's not how mathematic works.

 

(1 CU x 853 mhz x 64 ops x 2 cycles)  x 12  

is still the same than

12 CU x 853 Mhz x 64 ops x 2 cycles

= 1 310 208 fl ops

 

A 72% increase would be this gpu at 1376 Mhz (and probably burn quickly)

No.  A GPU is parrallel.  Therefore CU's can run independent of each other.  I can have one CU running a job each clock cycle, and another running one job over 853 clock cycles.  Because those each of those CUs will reduce the time to run an operation, the effect of those savings become accumulative.  For example, @ 800GHz I wouldn't be able to run 1000 operations.    This means it's capable of running 57 more operations. 

PS4 1 CU @ 800MHz @ 943 operations per clock cycle
XB1 1 CU @ 853MHz @ 1000 operations per clock cycle

So, at the end of one minute.

PS4 1CU @ 800MHz @ 943 operations per clock cycle = 754,400 operations per minute. 
XB1 1CU @ 853MHz @ 1000 operations per clock cycle = 853,000 operations per minute.

98,600 operations more.  Or 105 more clock cycles to do the same amount of work.  That's a 12% difference in performance right there.  Not to mention, that's one CU.

 

Please god you calculate things in minute for a gpu ? LOL

Do you even realize how much frames a gpu have to do in one minute ? that would be 1800 frames for a 30 fps games that's about 100 millions of operations for the gpu. It's not how you calculate ops anyway

Let's go back to basics and simple things...

1 CU @ 853 Mhz does 64 x 2 x 853 operations per second (floating point one). Thats 109 184 fl.op/s for each CU . Multiply it by 12 CUs and you get your 1.31 Tflops so this is the right calculation.

For fun if you want it per minute, it's 6 551 040 (yes +6 millions). So your math is wrong to begin with

And WE DON'T CARE if each CU is marginally faster by 6%, GPU are higly parrallel beast that's why we put more CU to get more powerfull cards.

Would you say an HD 7770 is faster or equal to a HD 7950 just because the 7770 CUs are clocked at 1000 Mhz and the 7950 are clocked at 800 Mhz on reference cards ?

NO, an HD 7950 blows away an HD 7770. And guess what's the main reason for that? It does have more CUs...it does also have more REAL bandwidth with GDDR5 (like the PS4 does) and twice the amount of ROPs (like the PS4 does)

 

See that chart ? the HD 7770 is all way down in performance and its CUs on their own are WAY faster than the HD 7950's CUs. 

 

It's not even a 53 Mhz difference it's 200 Mhz difference in 7770's favor, it's still get destroyed in real world performance (yes not gflops) 

Comparing a HD 7770 to a HD 7950 at just looking at Engine Clock speed doesn't really do it justice.  It's the only thing that is a tiny bit better.

While yes the HD 7770 does have a faster Engine Clock at 1000MHz, over the 7950's at 850 (boosted up to 925MHz), but it loses out on most every other spec.

The 7950 has nearly three times the CU Cores(28 vs 10), faster memory at 5 Gpbs vs 4.5 gpbs, and much higher bandwidth at 240BG/s vs only 72GB/s.

There is so much more that 7950 has better.  So it should have a higher performance.  But even with all those advantages, it just over doubles performance, even with nearly three times the CU Cores.  It should be much higher.