Really depends on the hardware. For example, Nvidia's Tensor cores accelerate ML by something like 700% compared to FP32.So is there difference from Dedicated Hardware ML acceleration and ML running on Hardware (CPU/GPU)
Technically doesn't a part of hardware freedup to run ML acceleration in itself become dedicated hardware for ML acceleration.
So is there difference from Dedicated Hardware ML acceleration and ML running on Hardware (CPU/GPU)
Technically doesn't a part of hardware freedup to run ML acceleration in itself become dedicated hardware for ML acceleration.
People said PS5 couldn't either with int4/8, only the Series consoles could. Turns out they were wrong, as usual.
But doesn't Mesh Shading run off the CUs too, where as the PS5 has a dedicated HW block (GE)
People say a lot of things, I'm interested in facts. I know for a fact that PS4 isn't int4/8 compatible
Precisely. Once PS5 software starts hitting all the available hardware feature of the system in unison things will truly shift to a new next gen paradigm.But doesn't Mesh Shading run off the CUs too, where as the PS5 has a dedicated HW block (GE)
The PS5 has less CUs but it appears to me it has other hardware freeing up said CUs e.g IO block with its own dedicated chip to free the CPU, Tempest Engine which handles all audio and perform further compute tasks if desired.
Not to mention the throughput of the architecture.
But doesn't Mesh Shading run off the CUs too, where as the PS5 has a dedicated HW block (GE)
The PS5 has less CUs but it appears to me it has other hardware freeing up said CUs e.g IO block with its own dedicated chip to free the CPU, Tempest Engine which handles all audio and perform further compute tasks if desired.
Not to mention the throughput of the architecture.
Insmoniac dev confirmed ML in the ps5 gpu.
Ass of Can Whooping
you are one of my favourite posters on this forum.
lol haha.
Right. I said that in another thread recently and someone got upset at me for it. It would be interesting if the Switch Pro has silicon dedicated for Tensor cores over shader cores.
GPU probably support INT4/INT8 just like RDNA2... after all the compute units are RDNA2.Insmoniac dev confirmed ML in the ps5 gpu.
GPU probably support INT4/INT8 just like RDNA2... after all the compute units are RDNA2.
That was what most believed but there are some FUD created to say the opposite
PS. It interesting to not like we were discussing in the Tony thread yesterday that all FUD is related to PS5 not having something.
Before RDNA 2 INT4/INT8 were emulated using INT16 units... you basically used all that silicon to do a single instruction in INT4 or INT8... now you use the silicon to do 4 INT4 or 2 INT8 instructions.
I think it will come down to silicon space vs performance. If tensor cores take 20% more space and give them 100% more performance via DLSS 2.0 or 3.0 then that would be a better usage of silicon space than adding 20% more shader cores for 20% more performance.
RDNA 1.1 is used in navi 12 (RX 5500xt)
I believe you meant Navi 10 at start of the third phase but I get it.
Yep sorry i mean in Navi 10 the implementation was broken
That would be wild since the info came from the update for Miles PS5 version and tweets before April 1.
'Of course, Monster Truck Championship is a cross-gen title, which means itās not built to leverage PS5 and Xbox Series Xās SSDs as much as next-gen exclusives in the future will be. More than anything, itāll be interesting to see how first party exclusives built specifically for either console will leverage their hardware.'
It was another
April fool in a update already released?
Some people are dense
An April Fool that updated a game a few days before April Fool's Day.
Cool.
Hardware based ML.
The original article is a complete utter joke/nonsense "Xbox series X has a much more impressive GPU" No need to read a single word after such a crude fallacy.
Yeah it's by the author and not the dev. Dev says about the loading speeds in their game, and talks about the future will be about fast SSDs used for more than just fast loading.
As far as I'm aware, Mesh and Primitive Shaders are still executed on the CU Shader Arrays, the GE along with the Command Processor are responsible for setting up graphics workloads and distributing them to the CU's. How those workloads are setup, scheduled and distributed to the CU is critical to performance (especially culling) and it's within the setup, scheduling and distribution the PS5 GE may differ from the standard RDNA 1/2 implementation and Series X/S.But doesn't Mesh Shading run off the CUs too, where as the PS5 has a dedicated HW block (GE)
The PS5 has less CUs but it appears to me it has other hardware freeing up said CUs e.g IO block with its own dedicated chip to free the CPU, Tempest Engine which handles all audio and perform further compute tasks if desired.
Not to mention the throughput of the architecture.
Yeah, i know about that. I talk about the author/article and the site. I don't like rewarding this kind of intentional misinformation/distortion by giving them clicks.
Think it was someone else, I only remember these guys doing the outriders demo.
Same as the demo, Series X looked worse and had a higher res, PS5 version was lower res and had better shadows and filtering.
