mckmas8808
Mckmaster uses MasterCard to buy Slave drives
Against "the Xbox SX either has only 7.5 GB of interleaved memory operating at 560 GB/s for game utilisation", argument.
From https://www.eurogamer.net/articles/digitalfoundry-2020-inside-xbox-series-x-full-specs
When the game is at full screen, OS front-end render will be suspended.
James Prendergast failed to understand CPU is not pretending to be a half-ass'ed GPU like in PS3's CELL SPUs i.e. CPU and GPU memory bandwidth intensity are different e.g. computation intensity difference for XSX's CPU : GPU ratio is 1 to 13, hence Prendergast's averaging argument is flawed.
Prendergast's argument is flawed when the CPU is the bottleneck i.e. a similar argument to why XBO GPU's 1.32 TFLOPS with +200 GB/s memory bandwidth was nearly useless against PS4 GPU with 1.84 TFLOPS and 176 GB/s memory bandwidth.
Targeting 60 Hz game loop, game programmer slices machine's potential in a 60 Hz game loop
PS5:
448 GB/s turns into 7.467 GB per 16ms frame budget potential
XSX:
336 GB/s turns into 5.60 GB per 16ms frame budget potential. 3.5 GB memory storage
560 GB/s turns into 9.33 GB per 16ms frame budget potential. 10 GB memory storage
2.86: 1 memory storage ratio between 10 GB vs 3.5 GB
As long XSX doesn't give equal time between 5.6 GB and 9.33 GB memory pools, XSX has the advantage.
Scenario 1
15% of 16 ms for 5.6GB memory pool= 0.84 GB
85% of 16 ms for 9.33 GB memory pool = 7.93 GB
Frame total bandwidth budget: 8.77 GB
XSX has 17.5% per frame BW advantage over PS5
Scenario 2
10% of 16 ms for 5.6GB memory pool= 0.56 GB
90% of 16 ms for 9.33 GB memory pool = 8.397 GB
Frame total bandwidth budget: 8.957 GB
XSX has 20% per frame BW advantage over PS5
Scenario 3
5% of 16 ms for 5.6 GB memory pool= 0.28 GB
95% of 16 ms for 9.33 GB memory pool = 8.8635 GB
Frame total bandwidth budget: 9.1435 GB
XSX has 22.5% per frame advantage over PS5
Scenario 4
2% of 16 ms for 5.6 GB memory pool= 0.112 GB
98% of 16 ms for 9.33 GB memory pool = 9.1434 GB
Frame total bandwidth budget: 9.2554 GB
XSX has 24% per frame advantage over PS5
Assigning CPU and GPU memory bandwidth for frame
Scenario A
Let CPU consumes 0.85 GB for 16 ms frame similar to 16 ms frame slice from 51 GB/s 128 DDR4-3200 PC config
PS5: 6.617 GB available to GPU per frame
XSX: 7.92 GB available to GPU per frame with 8.77 GB (from Scenario 1)
XSX GPU has 19.7% memory bandwidth advantage per frame over PS5 GPU
Scenario B
Let CPU consumes 0.85 GB for 16 ms frame similar to 16 ms frame slice from 51 GB/s 128 DDR4-3200 PC config
PS5: 6.617 GB available to GPU per frame
XSX: 8.107 GB available to GPU per frame with 8.957 GB (from Scenario 2)
XSX GPU has 22.5% memory bandwidth advantage per frame over PS5 GPU
Scenario C
Let CPU consumes 0.85 GB for 16 ms frame similar to 16 ms frame slice from 51 GB/s 128 DDR4-3200 PC config
PS5: 6.617 GB available to GPU per frame
XSX: 8.296 GB available to GPU per frame with 9.1435 GB (from Scenario 3)
XSX GPU has 25.3% memory bandwidth advantage per frame over PS5 GPU
Tile compute methods on both CPU's and GPU's multi-MB caches with TMU/ROPS can conserve external memory IO access.
--------------------
The reason for 970's slow 0.5GB
0.5GB DRAM bottlenecked without its own L2 cache and dedicated I/O link into the crossbar.
So the memory bandwidth difference is about the same as the TF difference? Makes sense to me. We all said it's a 20 or something percent difference for the most part.