We do. Ps5 soc is a known quantity at this point. The gpu is essentially a 5700 + Ray tracing hardware, and on the cpu side they removed some of the vector instructions.
1 Like
Might be ways to port it over to use other formats that are only 2x instead of 4x delta. But that will take additional effort, over the DP4A version.
1 Like
True, if Ps5 supports rpm for int16 could be possible, perhaps even FP16 if that’s not available
“The AMD SoC, codenamed AMD Oberon, features AMD’s exclusive design IPs like Zen 2 and RDNA 2.”
Why is this well renowned site spreading half baked truths?
beats me, they are not anandtech who know about tech. I posted the link just to show the die shot
The absence of Infinity Cache has been proven by the earlier die shots (remember the wild “industry leaks” by some youtubers?). Also we have the Sony principal graphics engineer accidentally confirming that it’s more like a RDNA 1.5.
1 Like
very true, its the same as the false claimes infinity cache on PS5
So Infinity Fabric is not Infinity Cache?
What does any of this SoC shots means against the stuff that was debunked eariler this year?
Do they still stand?
1 Like
Well I would hope so.
That would be rather embarrassing if the PS5 BC version of Pyschonauts 2 was better than the Series X version which probably was the primary platform.
PS5 in theory can run it with FP16 calculation precision
Xbox series in theory can run it with Int 8 calculation precisions
So similar stuff of int8 vs FP16 advantage
Interviews with the devs from DF and some other sites. They were asked about running XeSS on platforms without XMX and without DP4A and had this to say:
Karthik: No, not at the moment. We will look into it, but I cannot commit to anything at this point. Even, if you were able to do it, there’s the big question of performance and whether its justified.
3 Likes
I believe he mentioned this to Richard’s question of running upscaling on CPU’s ML hardware.
Intel has trying this for some time with their Intel Optimised Python library but honestly unless you run a server grade Xeon processor it hardly makes any difference.
It was also mentioned in response to the other interview question (WccfTech). So pretty clear that their focus is on their XMX and DP4A versions, anything beyond that is unclear because of performance constraints.
Usman: Does XeSS work on hardware with no XMX or DP4a support with an FP16 or FP32 fallback?
Karthik: No, not at the moment. We will look into it, but I cannot commit to anything at this point. Even, if you were able to do it, there’s the big question of performance and whether its justified.
1 Like
That is not what RDNA2 means. RDNA2 just refers to their microarchitecture that has hw support for RT, VRS and the perf/watt gains over RDNA1 and GCN. PS5 is absolutely based on RDNA2, it just lacks hw VRS support.
Should be noted that some of the more important applications of ML like SR simply won’t run with FP16 in a way that makes the use of ML worthwhile.
The hardware for DirectML is also missing.