As some have pointed out, TSMC is the real competitor, not really Apple or AMD. So I was surprised how quickly Pat managed to open up Foundry with industry standard tools and ecosystem. I have been saying this since 2012 even before the original Intel custom foundry were announced. ( And constantly being attacked and abused by Intel Fans online) But to actually see Intel announcing it is a completely different thing. Especially knowing their strong inertia. In hindsight, it probably took that beating from 10nm before they could really go all in with Custom Foundry. ( "All in" as in with industry tools, they are still an IDM, not Pure Play )
And how the tide has turned. All of a sudden Leading Edge Foundry becomes a national security concern. And Intel should rip all the benefits from it.
It was Andy Grove first started moving away from DRAM in the late 1980s, and may be his disciple will be the one who start to move away from x86 as core business.
Meanwhile, AMD gets a headstart on ARM and leaves Intel in the dust on that front. I don't see Intel winning one way or another. I think it will take many years for them to potentially come back as no. 1 in the industry.
Not sure if this is a new comer or new account. You may want to explain yourself. As you have nothing in your previous comment to "rehash".
If the past 10 years are records that I am way more plugged into what's going on than 99.99% of the internet.
And if you dont think TSMC is the competitor, it tells others more about you than about me.
But that's mostly because of apple reserving TSMC entire 5nm line, so AMD is still on 7nm (although they look extremely promising even on an older node)
Never seen one in the wild, though.
Personally I haven’t used CUDA, but if anybody could comment on its ease of use compared to Vulkan that would be great.
Source? I can't find anything for '"cublas" scheduler'…
Bespoke accelerator FPGA solutions are the future and will probably scale a lot longer than CPU/GPU combinations.
(I use "fixed function" loosely here. It's not a fixed function T&L pipleline, it's meant in the "the circuit is fixed but can execute arbitrary code" sense.)
This isn't my space as I operate on low-latency networking side.
> A more traditional "many 'fixed function' processor" a la GPU will beat it for any algorithm which at least semi-successfully maps to the provided operations. And most math is matrix multiplication anyway, if you squint your eyes and hold your tongue at the right angle ;)
I don't think it is as simple as this. Certainly not for low-latency scenarios or for HPC efficiency where the gains are done my moving as much computation as possible onto accelerator cards and off the CPU.
The cost of moving things back and forth from the CPU can be expensive (data especially). We are already seeing dedicated solutions from Google, Graphcore that beats the pants off the traditional GPU/CPU approach.
In this case being video streaming in from the NIC, encoding/decoding occurring directly on the NIC.
Same to a degree with the bespoke smartssd where if you have some complex encryption, encoding, decoding or data driven application than the processing happens directly on the storage device via the FPGA.
this is a big deal especially since ai tech advanced.
The other thing worth noting on the consumer side is that Zen 4 should also bring Navi 2 and DDR5 to their APUs. AMD's current APUs can already handle most eSports/low-end games at 1080p [1] but this next iGPU bump has the potential to displace the low-end AIB/dGPU market completely. When looking outside of the AIB numbers, AMD (and Intel) continue to increase unit shipments/gain market share at Nvidia's expense. [2]
As for GPU compute, there's no question that Nvidia continues dominate and it feels like if anything, AMD is falling further behind. While I have some hopes for CDNA/2 and ROCm continuing to mature, their ongoing lack of support for even basic stuff like desktop OpenCL or any RDNA cards means that AMD is moving backwards as far as ecosystem support goes. I really can't think of a better way to stunt adoption than to have no ML/compute support for the majority of their products. ¯\_(ツ)_/¯
On the process side, assuming Intel suffers no more delays on 7nm (a big assumption), it will still be half a step and half a year behind TSMC's N3 process in 2023 (beyond that, I suppose it'll be an open question of who gets their GAAFET working). AMD's CPU architecture team has a pretty solid track record at this point though, and I don't think Zen 4 or Zen 5 will be at any particular disadvantage vs Granite Rapids.
Obviously, AMD has lots of competition, but as long as they keep executing, I think they have a pretty bright future (also, we'll have to see how Xilinx plays out).
[1] https://static.techspot.com/images2/news/bigimage/2021/04/20...
[2] https://www.jonpeddie.com/press-releases/gpu-shipments-soar-...
TBH, the only reason I pay 800$ for a card today is to play games with ray tracing on at an acceptable resolution. This is only possible with nvidia RTX + DLSS.
I wouldn't pay 800$ for an AMD card that gets me similar rasterization performance but none of the "modern" (3 year old already...) features I care about.
Right now, _any_ GFX card is better than having no GFX. Both NVIDIA and AMD are sold out.
I can't know how real the AMD mindshare is. We probably could only know if we had a market with enough supply of both.
AMD only have two more years to enjoy their advantage from Intel's multiple missteps. After that it is going to be tough. And that is excluding the threat from ARM.