AMD phoenix 7040 AI engines
Are they real? Does any software use it? tensorflow versions for it?
It's All Fucked Shirt $22.14 |
AMD phoenix 7040 AI engines
Are they real? Does any software use it? tensorflow versions for it?
It's All Fucked Shirt $22.14 |
>Are they real?
Supposedly.
>Does any software use it?
Aside from rudimentary winblows copilot features? No
>tensorflow versions for it?
They don't exist.
Then why? Isn't it cool, to be able to run llamas from your ram, with higher speeds? 7040 ryzens are relatively popular.
Ayyymd can't even support ROCm on their lower-end GPUs.
Why would you waste silicon doing it, and not providing normal API?
https://www.amd.com/en/developer/resources/ryzen-ai-software-platform.html
https://ryzenai.docs.amd.com/en/latest/inst.html
I have no idea about AI shit, but those Zen 4 mobile CPUs make me salivate
It's just a bit of acceleration silicon meant for battery savings for laptops. the idea is that it provides simple things like video and audio processing for little power for corpo zoom meetings.
>It's just a bit of acceleration silicon meant for battery savings
Who the frick are you? Why would you even make this post.
The Phoenix Point APU die has a Xilinx AIE that has fricking absurdly high throughput for the power it consumes. It does literally nothing to save battery life for normal day to day use. It is there explicitly to make use of emerging workloads like stable diffusion, LLMs, etc.
https://www.amd.com/en/developer/resources/ryzen-ai-software-platform.html
first paragraph only speaks to efficiency. these are laptop chips, for crying out loud.
https://www.pcworld.com/article/1815008/why-amd-thinks-ryzen-xdna-ai-is-the-future.html
again it speaks to low power usage. it's basically just a little AI accelerator for basic AI task offload to save power, for now.
The ai silicon is gonna run like 2 watts max to be worthwhile offload from CPU/GPU doing it, much like hardware video acceleration is on the GPU these days.
back to the shadow realm, twerp.
Instead of being a shit eating child idiot you could look at the actual whitepaper for it and see what its capable of instead of making up asspulled statements about saving battery life
It is a Xilinx AIE
on a laptop chips you fricking moron, the whole chip is like 28w for the U series and the ai silicon is a small fricking sunset of the entire processor. learn to not be fricking moronic.
subset*
You are just throwing words together in desperation.
The Phoenix Point APU being used in laptops has no bearing on how the AIE functions. It doesn't magically take over tasks of the CPU and GPU to save power. It exists, explicitly, to be used in AI specific workloads like local LLMs. It does literally nothing to save battery life while your laptop is just idling, browsing the web, or playing videos.
Stop talking out of your ass. I know more about this than you do.
I never said anything to that, it's up to software devs to make software that utilizes the chip. that's pretty obvious, anon. it would save battery by offloading stuff to the ai chip, if the software can do it. currently there's like only one app on windows that can.
>I have literally no idea what this thing does
>software is magic!
>I'm going to keep talking out of my ass
Okay, little kid.
https://community.amd.com/t5/ai/amd-accelerates-ai-adoption-on-windows-11-with-new-developer/ba-p/607847
you aren't disproving my points at all, moron. the ai silicon is there to save battery by reducing CPU and GPU load for tasks it can handle.
You aren't making any points at all, nonwhite. You are parroting low IQ marketing bullet points that you don't even understand.
The AIE saves power vs the GPU *when you render an image in stable diffusion or generate a string with an LLM*
It does literally nothing, NOTHING, for battery life otherwise.
That's all that exists for the ryzen ai platform so far. Nothing has really been implemented yet. The silicon that is Ryzen ai is still just marketing hype until it becomes usable through software. There's not even linux tooling for it yet. You have to be on windows to even get the IPU driver to use it and develop with it. It's still a tiny 2w component of the overall 7040 series chips because that's all it can be justified to be. I really don't see where you are going with your argument anymore. You literally prove my point about the AIE saving power compared to the gpu and I don't know what else your anger-riddled brain wants to argue with me about. Saving battery is literally the only business justification of the silicon being on a mobile chip. Doubt AMD would even add ryzen AI to desktop chips since power(and battery, you fricking moron) is less of a concern on a desktop platform.
>until it becomes usable through software
Wow. I sure how Xilinx gets their software game on point! Not like they're the biggest FPGA/AI accelerator company in existence, who had software support available literally even before release! Haha!
The CPU doesn't have enough memory bandwidth for any significant AI bullshit.
>tensorflow
Nobody uses tensorflow