Wednesday, May 22nd 2024
Qualcomm's Success with Windows AI PC Drawing NVIDIA Back to the Client SoC Business
NVIDIA is eying a comeback to the client processor business, reveals a Bloomberg interview with the CEOs of NVIDIA and Dell. For NVIDIA, all it takes is a simple driver update that exposes every GeForce GPU with tensor cores as an NPU to Windows 11, with translation layers to get popular client AI apps to work with TensorRT. But that would need you to have a discrete NVIDIA GPU. What about the vast market of Windows AI PCs powered by the likes of Qualcomm, Intel, and AMD, who each sell 15 W-class processors with integrated NPUs capable of 50 AI TOPS, which is all that Copilot+ needs? NVIDIA held an Arm license for decades now, and makes Arm-based CPUs to this day, with the NVIDIA Grace, however, that is a large server processor meant for its AI GPU servers.
NVIDIA already made client processors under the Tegra brand targeting smartphones, which it winded down last decade. It's since been making Drive PX processors for its automotive self-driving hardware division; and of course there's Grace. NVIDIA hinted that it might have a client CPU for the AI PC market in 2025. In the interview Bloomberg asked NVIDIA CEO Jensen Huang a pointed question on whether NVIDIA has a place in the AI PC market. Dell CEO Michael Dell, who was also in the interview, interjected "come back next year," to which Jensen affirmed "exactly." Dell would be in a front-and-center position to know if NVIDIA is working on a new PC processor for launch in 2025, and Jensen's nod almost confirms this
NVIDIA has both the talent and the IP to whip up a PC processor—its teams behind Grace and Drive can create the Arm CPU cores, NVIDIA is already the big daddy of consumer graphics and should have little problem with the iGPU, and the NPU shouldn't be hard to create, either. It wouldn't surprise us if the NPU on NVIDIA's chip isn't a physical component, but a virtual device that uses the AI acceleration capabilities of the iGPU with its tensor cores, as a hardware backend.
NVIDIA's journey to the AI PC has one little hurdle, and that is the exclusivity Qualcomm enjoys with Microsoft for the current crop of Windows-on-Arm notebooks, with its Snapdragon X series chips. NVIDIA would have to work with Microsoft to have the same market access as Qualcomm.
If all goes well, the NVIDIA PC processor powering AI PCs will launch in 2025.
Sources:
Bloomberg (YouTube), Videocardz
NVIDIA already made client processors under the Tegra brand targeting smartphones, which it winded down last decade. It's since been making Drive PX processors for its automotive self-driving hardware division; and of course there's Grace. NVIDIA hinted that it might have a client CPU for the AI PC market in 2025. In the interview Bloomberg asked NVIDIA CEO Jensen Huang a pointed question on whether NVIDIA has a place in the AI PC market. Dell CEO Michael Dell, who was also in the interview, interjected "come back next year," to which Jensen affirmed "exactly." Dell would be in a front-and-center position to know if NVIDIA is working on a new PC processor for launch in 2025, and Jensen's nod almost confirms this
NVIDIA's journey to the AI PC has one little hurdle, and that is the exclusivity Qualcomm enjoys with Microsoft for the current crop of Windows-on-Arm notebooks, with its Snapdragon X series chips. NVIDIA would have to work with Microsoft to have the same market access as Qualcomm.
If all goes well, the NVIDIA PC processor powering AI PCs will launch in 2025.
34 Comments on Qualcomm's Success with Windows AI PC Drawing NVIDIA Back to the Client SoC Business
Also, gg qualcomm for this huge accomplishment
But on a more serious note I bet this would require firmware, drivers and OS support that are rather extensive undertaking if we are talking something like making a Windows laptop out of this.
"Windows AI PCs powered by the likes of Qualcomm, Intel, and AMD, who each sell 15 W-class processors with integrated NPUs capable of 50 AI TOPS"
But Nvidia clearly outgrew catering to lowly penny pinching peasants - they practically don't offer low end GPUs, and with every generation they delay their lower end offerings more and more. And we can understand why - their server, AI mainframes are what's driving the stellar growth, not home users. Why would all of a sudden they want to deal with market that requires low margins and vast volume?
If they were to do something with risc V where they can keep more of the pie... that might be something different.
Some background on the current Cortex-X4:
www.hwcooling.net/en/arm-unveils-record-breaking-cortex-x4-core-with-eight-alus/
Geekbench 6 puts its 1T perf at 5950X / 7840U or i9-11900 / i3-12100.
SPECint2017 puts the 1T perf at 6900HS / Core Ultra 155H.
And that's all at low 3.2 GHz clocks. NVIDIA will have a lot of strong options even in the X4, tbh. But it's more likely NVIDIA will use the 2025 core, the Cortex-X5, which Arm claims has much higher IPC.
At the moment, any SoC manufacturer can get an ultra-high-performance and high efficiency uArches and need not develop their own by using Arm's Cortex X3, X4, and upcoming X5. It's honestly never been a better time to make a high-end SoC.
Apple has its custom uArches.
Arm has its custom uArches
Intel has its custom uArches.
AMD has its custom uArches.
Qualcomm has its custom uArch.
I personally can't wait to see Arm filter to DIY desktops in the next decade.
I don't see why a scaled down version of it couldn't compete with the snapdragon.
And yeah, I don't know that NVIDIA needs to go full-custom. They could pull the architecture off the shelf and probably get more out of it by using advanced nodes like Apple does. It sure seems like they could easily answer Snapdragon if they wanted to, and now there's a window of opportunity for such devices. It makes me wonder if MS hasn't already asked NVIDIA, and NVIDIA wasn't interested. Or maybe MS didn't want to deal with NVIDIA, I dunno.
They just have to build a Windows compatible one, it's trivial.
Meanwhile in the real world we have ~
www.phoronix.com/review/nvidia-gh200-amd-threadripper
openbenchmarking.org/result/2402191-NE-GH200THRE98&sgm=1&ppd_U3lzdGVtNzYgVGhlbGlvIE1ham9yIHI1IC0gVGhyZWFkcmlwcGVyIDc5ODBY=15076&ppd_SFAgWjYgRzUgQSAtIFRocmVhZHJpcHBlciBQUk8gNzk5NVdY=30041&ppd_R1BUc2hvcC5haSAtIE5WSURJQSBHSDIwMA=42500&ppt=D&sor
It's easy to forget how bandwidth starved regular zen4 chips are, I think I saw that analysis on Chips & Cheese. With more memory channels &/or higher speed memory they easily pull way past Grace Hopper & Emerald (Sapphire?) Rapids as well. This is why Strix point & Halo would be interesting to watch & whether AMD can at least feed zen5 better on desktops/mobile platforms!
The Tegra X1 was used in a couple other devices(Pixel C) but it end up not seeing wide adoption. It had a cluster of A53 in it but I believe there was a silicon bug that ended up with those not being used(and later removed).
Competitors SoCs from the same period had a cluster of quad core A72 and another quad core cluster of A53, with about the same clocks.
The big advantage that Nvidia had at the time was GPU but that ends up not being as important.
Remember how Apple, EVGA, MS, Sony and others tasted a nice fat knife in their backs courtesy of Ngreedia.
Also, Denver was pretty good back when Nvidia was trying to cook their own. Pretty sure they have the know-how.
Tegra did not fail spectacularly, it kind of slid out of our view. They pivoted from consumer stuff to automotive and industrial. Most likely due to profit margins.
We already have a whole landscape of affordable and practical CPUs in a wide variety of ways. If Nvidia does off the shelf cores, how will they differentiate? Probably with a segment of Nvidia sauce. So they can present some USP on their AI capability. Its going to be a massive bunch of bullshit. Much like RT there's going to be a black box 'NPU performance' to compare between constantly moving models and locally updated tools? Its going to fail spectacularly.