- Nov 12, 2016
And here I took this graphics card with 14 Gbps (448 GB/s) and was able to up its bandwidth to 512 GB/s easily. I'm sure Sony could have done even better...As much as I like PS5 hardware, and fascinated by the talent of Mark Cerny, I'm still wondering why they opted for 448 GB/s instead of 512 GB/s. I believe they were not sure till last minute, but this is just my thought. That was the only missing thing to make PS5 'perfect'. What would be the extra cost to give the beast 512 GB/s of memory bandwidth ? Just wondering if it was feasible or not.
That said, the actual performance gains from this 14% increase in BW was very little based on my tests. You need way more than 512 GB/s to see any noticeable perf gains...Take 3070 Ti, for instance, it comes with GDDR6X which gives it 608 GB/s, a massive 36% increase over 3070 but on avg. the Ti is only about 7% faster than 3070 according to the TPU database and the compute power between Ti and non-TI is basically identical.
So my guess is that PS5 wouldn't have benefitted much from 512GB/s due to diminishing returns and probably needed way more than 600+ GB/s for any meaningful perf gains or something like an infinity cache-like setup for much higher effective BW which must not have been viable due to limited power and silicon real estate.