Discussion NV Re-Enter ARM PC market in 2025!

[Nothingness]

Some architectural dibits on the Nvidia CPUs:

GCC & LLVM Clang Merge Support For The NVIDIA Olympus Cores With The Vera CPU - Phoronix

www.phoronix.com

ARM v9.2-A cores and on top of that they support:

SVE2_BITPERM, RNG, LS64, MEMTAG, PROFILE, FAMINMAX, FP8DOT2, LUT, SVE2_AES, SVE2_SHA3, SVE2_SM4.

Finally decent SVE2 support in an ARM core it seems

Cortex-x925 already supports many of these extensions.

 

[soresu]

Finally decent SVE2 support in an ARM core it seems

Apparently it was in the Google Tensor SoCs, it's just that people don't pay nearly as much attention to them, because they were never top of the line.

 

[poke01]

Apparently it was in the Google Tensor SoCs, it's just that people don't pay nearly as much attention to them, because they were never top of the line.

Does android make use of SVE?

Can apps take advantage of this capability?

 

[adroc_thurston]

Apparently it was in the Google Tensor SoCs, it's just that people don't pay nearly as much attention to them, because they were never top of the line.

MTK shipped SVE2 too.
QC obviously disabled it since Nuvia IP will only ever have it like next year.

 

[poke01]

MTK shipped SVE2 too.
QC obviously disabled it since Nuvia IP will only ever have it like next year.

It looks like QC will have SVE2 this year

 

[soresu]

Does android make use of SVE?

Can apps take advantage of this capability?

It's hardly in anything at all so far.

dav1d has an open issue for it, but given the NEON path is currently pretty mature the SVE2 path has non existent priority against RVV and even AVX512.

On that note, I hope that AVX512 -> AVX10 code migration won't require significant rework.

 

[LightningDust]

It's hardly in anything at all so far.

dav1d has an open issue for it, but given the NEON path is currently pretty mature the SVE2 path has non existent priority against RVV and even AVX512.

On that note, I hope that AVX512 -> AVX10 code migration won't require significant rework.

AVX512 and AVX10-512 are essentially identical from the assembly programmer's point of view, IIRC, though there are encoding changes.

 

[Gideon]

GN discussion with Wendell from Level 1 about ARM and if x86 is screwed. Wendell thinks AMD can pivot fairly quickly but that Intel is boned. He says pricing and that it's "good enough" are strong points. Talks about how none of those Amazon Graviton or other ARM servers are ever coming back to x86; that biz is gone for good, just as with Apple.

This video also mentions Nvidias' WARM (Windows on ARM) drivers as a thing Wendell and Steve were not allowed to talk about .

All in all makes it pretty obvious that Nvidia is doing Windows ARM SoCs. Might also explain why their drivers suddenly become such dumpster fire (due to porting).

 

[Tigerick]

大举进攻x86！NVIDIA、联发科开发首款Arm PC处理器：台式机、笔记本都有

快科技5月7日消息，NVIDIA和联发科合作打造的首款PC处理器，很快就要来了。最新消息显示，NVIDIA和联发科预计将在Computex 2025台北电脑展推出共同开发采用Arm架构的PC处理器——N1

news.mydrivers.com

Hmm, NV might be launching two ARM SoC, N1 and N1x in Computex 2025.

The news indicate that one SoC for desktop and another for laptop which I find it unlikely. And the specs 10 x X925 and 10 x A725 are definitely refer to Project Digits, GB10.

I think reporter mixed up the specs. Anyhow, NV going to launch ARM SoC from top to bottom and Jensen is aiming big with ARM PC market. Definitely not custom SoC for Microsoft. Let's wait for Qualcomm AND AMD to response....

Nonetheless, let's welcome new player in the game...

 

[DZero]

大举进攻x86！NVIDIA、联发科开发首款Arm PC处理器：台式机、笔记本都有

快科技5月7日消息，NVIDIA和联发科合作打造的首款PC处理器，很快就要来了。最新消息显示，NVIDIA和联发科预计将在Computex 2025台北电脑展推出共同开发采用Arm架构的PC处理器——N1

news.mydrivers.com

Hmm, NV might be launching two ARM SoC, N1 and N1x in Computex 2025.

The news indicate that one SoC for desktop and another for laptop which I find it unlikely. And the specs 10 x X925 and 10 x A725 are definitely refer to Project Digits, GB10.

I think reporter mixed up the specs. Anyhow, NV going to launch ARM SoC from top to bottom and Jensen is aiming big with ARM PC market. Definitely not custom SoC for Microsoft. Let's wait for Qualcomm AND AMD to response....

Nonetheless, let's welcome new player in the game...

Maybe is plausible... both are work from nVIDIA and Mediatek.

Maybe... the desktop piece would go over 4.5Ghz+ with consumption of 60 or 90 watts. GPU performance of a GTX 3050 since the best iGPU (AMD one) is on that ballpark now
The laptop part could go lower (3.5 Ghz) with the consumption of 25 to 50 watts. GPU performance of a GTX 2050 since Intel iGPU is on that ballpark.

The issue will be compatibility

 

[Schmide]

AVX512 and AVX10-512 are essentially identical from the assembly programmer's point of view, IIRC, though there are encoding changes.

Explain this. This seems contradictory. Are the opcodes different? I highly doubt it.

 

[marees]

Nvidia and Mediatek's AI CPU may not see mass rollout until late 2026 — Asus, Dell, and Lenovo reportedly developing N1X desktops and laptops​

News
By Stephen Warwick published 3 hours ago
The chip is tipped for an unveiling at Computex 2025

the report notes, early N1X benchmarks hint at performance that lags behind some Arm-based PC chips, noting "the results have raised industry concerns."

Perhaps more worryingly, Digitimes reiterates reports that there are "unresolved integration issues with endpoint devices." These manufacturing headaches have previously been reported elsewhere, and could explain the hefty lead time of 2H26 on these chips.


https://www.tomshardware.com/pc-com...eportedly-developing-n1x-desktops-and-laptops



Nvidia and MediaTek's hotly anticipated AI PC chips may not see meaningful shipment volumes to the mass market until the second half of 2026, according to a new report. As reported by Digitimes, Nvidia and MediaTek are expected to jointly unveil their new 'N1' Arm chips for Windows PCs at Computex.

According to Digitimes, the joint chip will "likely debut under the Nvidia brand," with both N1X and N1 models planned, echoing previous reports. Digitimes says that both companies are well into production ramp-up, however, states "insiders believe meaningful shipment volumes won't emerge until the second half of 2026."

https://twitter.com/x/status/1914321933853655321

 

[poke01]

What a fail. Nvidia can’t do consumer hardware right especially CPUs/SoCs.
The X925 will be outdated crap by then especially with Nova lake and M6 out

 

[adroc_thurston]

What a fail. Nvidia can’t do consumer hardware right especially CPUs/SoCs.

That's MTK too.

The X925 will be outdated crap by then especially with Nova lake and M6 out

well yeah kinda.
at least it's not a Tegra so it won't have mountains of catastrophic errata.

 

[gdansk]

What a fail. Nvidia can’t do consumer hardware right especially CPUs/SoCs.
The X925 will be outdated crap by then especially with Nova lake and M6 out

If it has working graphics drivers in Windows (might be a big if for a Blackwell GPU) then they'll be way ahead of Qualcomm. X925 is good enough if the GPU can sell the platform.

 

[Tigerick]

Hoho, once again the medias (Digitimes & Charlie) are trying to interpret why NV will unveil N1 and N1x at Computex without shipping the products. "unresolved integration issues with endpoint devices." WTH?, GB10 (with LinuxOS) should be shipping this month; there shouldn't have any hardware issue. With Blackwell essential the same as x86 GPU IP, I don't see NV going to have issues with driver developments.

If there is showstopper bug, Jensen will simply NOT GOING TO unveil N1 and N1x. Jensen jumps the gun because he wants to be the first to unveil / showcase ARM SoC before others (AMD, Mediatek or Samsung). And the real reason is simply all CPU makers are waiting for WoA12 which is supposedly be reveal by the end of the year. As explained here; Microsoft going to refresh Windows 12 by the end of the year to support all new ARM vendors. "unresolved integration issues with endpoint devices." simply meant the system is ready to go but stuck waiting for RTM version of WoA12. Of course, this is my assessment; time will tell whether I am correct or not...

 

[poke01]

If it has working graphics drivers in Windows (might be a big if for a Blackwell GPU) then they'll be way ahead of Qualcomm. X925 is good enough if the GPU can sell the platform.

CPU has be good. NV doesn’t need convince people about their GPUs, it’s ARM that’s the problem.

How good is the emulation for games and for professional apps?
If it’s not Nova Lake level then it’s a waste.

You can get NV GPUs on x86 laptops so the CPU has to shine the most.

 

[adroc_thurston]

You can get NV GPUs on x86 laptops so the CPU has to shine the most.

It'll be an alright CUDA stick on Linux.
WoA? more like DoA.

 

[DZero]

It'll be an alright CUDA stick on Linux.
WoA? more like DoA.

Just like Raptor Lake.

 

[Gideon]

Another hint that Nvidia is working on ARM driver stack

https://twitter.com/x/status/1922582064139837704

 

[marees]

Is POPCNT relevant to discussion of ARM processors 🤔

https://www.tomshardware.com/pc-com...-change-opens-up-bizarre-system-build-options

 

[coercitiv]

Another hint that Nvidia is working on ARM driver stack

https://twitter.com/x/status/1922582064139837704

Not sure if this was discussed here, but the biggest hint we got was from a discussion between Steve from Gamers Nexus and Wendell from Level, they jokingly said they're not supposed to talk about Nvidia's ARM driver stack, hinting they were under NDA. I'll post a timetamped video if I can find it quickly.

Got lucky:

 

[MS_AT]

Another hint that Nvidia is working on ARM driver stack

To be fair, it proves nothing. The reason popcnt got there, could be simply that somebody was too hopeful when specifying the baseline to compile for, thinking that people have moved past 15 years old CPUs. I would find that unlikely they would use popcnt from inline assembly. It's not to say, they are not working on ARM drivers, it's just this is not a good argument for the case, imo.

 

[adroc_thurston]

Not sure if this was discussed here, but the biggest hint we got was from a discussion between Steve from Gamers Nexus and Wendell from Level, they jokingly said they're not supposed to talk about Nvidia's ARM driver stack

They already have the aa64 driver stack how would Grace Hopper work otherwise

 

[LightningZ71]

Doesn't ARM have the "CNT" instruction that does essentially the same thing? I don't see ditching this instruction as related to ARM driver development.