I think both Rembrandt and Escher/Hawk will get the boot after this year. MS will want every laptop to support AI PC sooner than later.The mainstream recently got Rembrandt at a good price, I think this+next year the mainstream will be 8540U
nope.The mainstream recently got Rembrandt at a good price, I think this+next year the mainstream will be 8540U (small phoenix)
that's 242.KRK 243
Zen 5 DT won't have an NPU and thus won't support AI PC without a dGPU. But it shouldn't matter.
So clearly Intel makes another judgement here w.r.t. whether it makes sense to incude NPU on desktop CPUs, since Arrow Lake DT will include it.Q - Do it Yourself Desktop Users who want AI PC experiences without being attached to an Laptop. How will they be able to get it?
They can buy an dGPU that meets the minimum system TOPs. Or can replace their system with Intel Arrow Lake which will have ~70 TOPs total, with 16 - 20 TOPs from the overclocked SoC NPU and ~50 TOPs from the brand new TSMC N4 Xe LPG with XMX units.
So it's actually 2 WGP instead of 3? Basically mirroring PHX2 for KRK 242 and PHX1 for KRK 444 (Sans the GPU).that's 242.
ARL DT meeting the requirements is more of a side-effect from reusing the same tiles as MTL and the need of adding back XMX units to the iGPU for ARL-H to meet AI PC requirements.So clearly Intel makes another judgement here w.r.t. whether it makes sense to incude NPU on desktop CPUs, since Arrow Lake DT will include it.
You can argue which path is the correct one, i.e. to include NPU on desktop CPUs or not. But to say that there is absolutely no discussion about what is correct path to take is simply not being honest. Especially with all the AI hype, and users buying a new PC wanting to be future-proof in this regard.
For those that do not need a dGPU on desktop, but want an AI-capable Win12 PC, they will have two options. Buy an Intel Arrow Lake CPU which already has an NPU included, or buy an AMD Zen5 without NPU included + dGPU (which they do not really need, except for the AI capabilities). For those, the former option will be much cheaper.
That said, I still think it is unclear whether or not AMD will have a solution to comply with the AI requirements on Zen5 DT. In their presentation slides they have stated that it’ll include ”Integrated AI and Machine Learning optimizations”. But no further details about what that means in practice. Could be that Zen5 DT will achieve the 40 TOPS via e.g. iGPU + some kind of generic AI optimized CPU instructions.
There s a test at computerbase, a mainstream GPU is miles above an integrated NPU, current low end dGPUs have several years advance, even next gen APUs will trail by a 4x ratio compared to a current low cost dGPU.It was a response needed since only mobile variants were mentioned regarding NPU. And DT PCs are not always paired with dGPU. You should know that by now. But you’re sounding like a broken record about it.
Well, Intel could have decided to use another config for ARL desktop, but they chose not to.ARL DT meeting the requirements is more of a side-effect from reusing the same tiles as MTL and the need of adding back XMX units to the iGPU for ARL-H to meet AI PC requirements.
You're correct in saying that Intel judged differently. But, as others said, it doesn't matter much because nor DT is the focus of AI PC push nor DT users will value AI PC compatibility.
This will have a bigger impact in DT OEMs. But that's a tiny slice of the market in comparison to mobile.
Of course a dGPU will be faster. But it'll also be overkill for the 40 TOPS requirement, and much more expensive compared to having it integrated in the CPU via NPU and/or iGPU.There s a test at computerbase, a mainstream GPU is miles above an integrated NPU, current low end dGPUs have several years advance, even next gen APUs will trail by a 4x ratio compared to a current low cost dGPU.
Will it though? The only AM4 desktop CPU's that had integrated graphics were repurposed mobile. The vast majority of OEM desktops use the G CPU's rather than normal desktop SKU's anyway. The OEM desktops that do use regular desktop SKU's are pretty much always using a discrete GPU. Perhaps this changed a bit with AM5 (and I didn't notice) but it seems like a pretty minor shift to go back to using more mobile derived SKU's on these types of machines. They're usually better suited to the SFF corporate desktops that constitute much of this market anyway.This will have a bigger impact in DT OEMs.
But they did.Well, Intel could have decided to use another config for ARL desktop, but they chose not to
Was just thinking about this... Zen 5 can't becoming out at a worse time with the whole industry apesheet over AI.
A 4070 is 6-8x faster in AI tests than current mobile chips with NPUs, so even a 7600/7600XT or 4060 will be largely ahead of a next gen APU s NPU + CPU + GPU compute throughput.Of course a dGPU will be faster. But it'll also be overkill for the 40 TOPS requirement, and much more expensive compared to having it integrated in the CPU via NPU and/or iGPU.
Its been like almost 2 years there is still no Lenovo Legion 7 Pro(AMD) here at all, only Intel lolSame here. OEMs have a fetish into bringing Intel only options. While the share of AMD laptops has drastically increased, they still stick to up to Zen 3 + Vega SKUs. I'm hoping next year they'll bring Kraken based designs due to Intel having no viable alternative in the space
Watch them bring MTL/ARL-U based designs and try to claim they're KRK competitive
Then why are you doing DIY PC anyway?Of course a dGPU will be faster. But it'll also be overkill for the 40 TOPS requirement, and much more expensive compared to having it integrated in the CPU via NPU and/or iGPU.
In AMD's case most systems that will be running without a dGPU will likely be mobile APU based anyway.Of course a dGPU will be faster. But it'll also be overkill for the 40 TOPS requirement, and much more expensive compared to having it integrated in the CPU via NPU and/or iGPU.
In AMD's case most systems that will be running without a dGPU will likely be mobile APU based anyway.
Mobile APUs are generally better suited to office PC style usecases than stuff like Raphael. They idle lower, those kinds of office style PC have no real usecase for the handful of cores you get on mobile anyway and of course now they'll have NPUs. That's just the more obvious solution.
I feel like you're heavily overestimating how much of the PC market is tower PCs running high end 12+ core CPUs without dGPUs. It's a very, very, very small subset of systems.
It should but there's nothing else besides the kernel on Linux that explicitly uses them I thinkIs the NPU not a part of the GPU or is it a discrete/separate part of the die ?
Also has there been any news on the Linux Kernel utilizing the upcoming NPU's ? (I guess more of a question for the phoronix forums)
I believe the problem is still that Intel basically chews the integration work for OEMs in laptops while AMD is just sending some docs and goes "GL m8".Same here. OEMs have a fetish into bringing Intel only options. While the share of AMD laptops has drastically increased, they still stick to up to Zen 3 + Vega SKUs. I'm hoping next year they'll bring Kraken based designs due to Intel having no viable alternative in the space
Watch them bring MTL/ARL-U based designs and try to claim they're KRK competitive
They're a seperate block.Is the NPU not a part of the GPU or is it a discrete/separate part of the die ?
That is just so incredibly dumb for fixed hours work. They do it here as well. Awful loud machines though with annoying fans. On permanent dock. 3x as expensive. But when I request a nicer monitor or a nicer mouse, nope. IT people are dumb here.
Did you not even read my post which you responded to? As stated in it, I'm not questioning that a dGPU can be faster than the NPUs currently used in mobile CPUs.A 4070 is 6-8x faster in AI tests than current mobile chips with NPUs, so even a 7600/7600XT or 4060 will be largely ahead of a next gen APU s NPU + CPU + GPU compute throughput.
We re talking dGPUs that will cost around 200-250$ when the next APUs will be released, only Strix Halo will be better in AI than those low cost discrete solutions.
Why are you assuming everyone needs a powerful dGPU, just because they have a powerful CPU? They handle completely different types of workloads.I feel like you're heavily overestimating how much of the PC market is tower PCs running high end 12+ core CPUs without dGPUs. It's a very, very, very small subset of systems.