Nuvia: don’t hold your breath

The pace of this team is fast. That and/or they started over working back to where they really had in mind.
It's been 3 years since the Nuvia acquisition. Qualcomm has since put out 3 cores.

Oryon (Gen 1)
Oryon-L (Gen 2)
Oryon-M (Gen 2)

It seems Gerard Williams and Co. are really starting to hit their stride.

The only thing that can stop them is that accursed lawsuit by ARM...
 
It's been 3 years since the Nuvia acquisition. Qualcomm has since put out 3 cores.

Oryon (Gen 1)
Oryon-L (Gen 2)
Oryon-M (Gen 2)

It seems Gerard Williams and Co. are really starting to hit their stride.

The only thing that can stop them is that accursed lawsuit by ARM...
also they suck
 
Sure they prob. Won’t get the original optimistic projection of 2000 at 2.5W or something soon, that’s history now.

However:

I think they did just deliver -57% power iso-performance on the X Elite with the 8 Elite and V2, primarily with architecture changes, and a massive increase in baseline ST they can yield from 2450 to 3200 — and doing so around the same power (or ~ 30% more perf iso-power).

Better just slightly than the X925. At worst they’re doing 10% better iso-power than an Arm core and with lower IP costs. Which is still exciting seeing as those X cores also Improved a great deal in the last 2-3 years still.

IMG_2919.jpeg


Intel is toss up on ST perf/W vs the X Elite, with a more expensive die. So an 8 Elite with Oryon V2 in a laptop (with more cores and diff bus width) would already blow that out of the water and in real use, same with the X925 and A7x stuff that Nvidia is going to use soon for 2025.

I agree the X Elite was disappointing in its core curve and GPU especially. But anyway, looking at the trajectory, things are looking up for Windows and Android both, software notwithstanding lol.
 
I’m not claiming what QC engineers have done is disappointing. I think they have done a very good job. I would just prefer if their CEO and marketers tried less to claim credit now, for things they will or may do in the future. That’s not exclusive to QC, but I find it fatiguing nonetheless.
 
I’m not claiming what QC engineers have done is disappointing. I think they have done a very good job. I would just prefer if their CEO and marketers tried less to claim credit now, for things they will or may do in the future. That’s not exclusive to QC, but I find it fatiguing nonetheless.
See I just ignore most of this stuff and look where the grass is actually green since usually there’s still a lot of improvement. Intel has done what QC has but on steroids for years, and they’re not the only one either obviously. Apple is the only one that really sandbags, AMD sometimes slightly though also does some disingenuous stuff just less so.


I did make salient the M2 Max shadiness here and elsewhere though and I agree the original claims about their performance and performance/W were way too enthusiastic.
 
Well, if that is the case, mobile users don't really care if the game they play show a few more FPS? Apple likely is again skating to where the the puck is going, rather than chasing the puck. I have a feeling that QC's mobile GPUs will have to play catch up soon.
You were damn right.
 
Sales figures for Qualcomm's Snapdragon X chips are here.


June-September.
720,000 devices.
0.8% of all PCs sold in this period.

Thoughts?
Only those that I’ve already had and shared. The main problem was that Qualcomm released too late with Lunar Lake and Strix Point right around the corner. Add in imperfect execution, especially from MS, and the low sales numbers are unsurprising.

In order for Qualcomm to break through, they have to offer a product way more compelling than AMD/Intel to get over the compatibility barrier (yes the CPU must be great but so too must the rest of the SOC) and MS has to do better if reviewers are saying that Apple/Linux has better compatibility with Windows-x86 than Windows on Arm. And neither company can leave developers hanging like they’ve been doing.
 
the CPU must be great but so too must the rest of the SOC

Qualcomm/Windows have the "advantage" that they are not wedded to the GPU on the SoC. If the QC GPU is not adequate to a need, the builder can just add in an AMD or nVidia card to boost performance as needed (as long as MS supports ARM drivers for cards). The OS could decide whether to use the GPU in the SoC for simple work or the card for the heavy work.
 
Qualcomm/Windows have the "advantage" that they are not wedded to the GPU on the SoC. If the QC GPU is not adequate to a need, the builder can just add in an AMD or nVidia card to boost performance as needed (as long as MS supports ARM drivers for cards). The OS could decide whether to use the GPU in the SoC for simple work or the card for the heavy work.
is that a thing? Arm Windows machines that support off-SoC GPUs?
 
More benchmark shenanigans. This time from OnePlus. It’s not their first time. Usual method.

1732660871482.png

1732660899950.png

1732660947892.png

Weird story. I wonder what is going on here? It will be interesting to see what happens with other phones as retail units come out.

View attachment 32388
View attachment 32389
 
is that a thing? Arm Windows machines that support off-SoC GPUs?

Not knowing much about WoA, I have no idea. But dGPU vendors supply drivers, right? And the drivers can be compiled for ARM, I would imagine. So one could imagine that Windows would be open to using a dGPU instead of the internal GPU. That windows would be smart enough to handle choosing one or the other for a given job seems unlikely at this time, but if there were enough demand, they might implement GFlex.
 
Qualcomm/Windows have the "advantage" that they are not wedded to the GPU on the SoC. If the QC GPU is not adequate to a need, the builder can just add in an AMD or nVidia card to boost performance as needed (as long as MS supports ARM drivers for cards). The OS could decide whether to use the GPU in the SoC for simple work or the card for the heavy work.

is that a thing? Arm Windows machines that support off-SoC GPUs?

Not knowing much about WoA, I do not know. But dGPU supply drivers, right? And the drivers can be compiled for ARM, I would imagine. So one could imagine that Windows would be open to using a dGPU instead of the internal GPU. That windows would be smart enough to handle choosing one or the other seems unlikely at this time, but if there were enough demand, they might implement GFlex.

In theory, yes ... in practice though? Not so far. Perhaps next year or two years from now after Nvidia and maybe AMD release their own rumored ARM-SOCs and public ARM drivers for AMD/Nvidia GPUs might exist. Otherwise, there is zero incentive for Nvidia, AMD, or Intel to help Qualcomm release a laptop or desktop with a capable dGPU and gain market share*. Plus, beyond the GPU apparently a lot of other parts of the SOC were found wanting by the reviewers, the NPU was fine, but if I remember correctly things like (de-)compression were woeful compared to Intel/AMD.

According to NotebookCheck, in addition to low sales figures now, Qualcomm has drastically cut its sales figures for its PC chips:

Qualcomm originally had ambitious plans for the Snapdragon X Elite, according to which Snapdragon X was intended to account for 40 to 60% of the PC market by 2027. Qualcomm recently revised these plans and is now targeting a market share of 30 to 50% by 2029. Even more drastic is the limitation that this market share is intended to be achieved with AI notebooks not based on x86; Qualcomm apparently no longer even counts Intel and AMD in the notebook market.
(I haven't been able to verify that Qualcomm has cut their sales forecast by so much that they only expect to have 50% of the ARM PC laptop market by 2029 which is what NBC is suggesting ... unless they expect ARM to get 80% of the PC market by then? ;) EDIT: that's possible or something similar since apparently Qualcomm expect to have $4 billion in PC chip revenue by then, which is roughly what AMD brings in now in consumer revenue so maybe not quite as pessimistic as NBC is suggesting)

*especially not when Qualcomm apparently still has an exclusivity deal with MS to supply SOCs for Windows on ARM, which I thought that expired this year but apparently that's next year.
 
Last edited:
In theory, yes ... in practice though? Not so far. Perhaps next year or two years from now after Nvidia and maybe AMD release their own rumored ARM-SOCs and public ARM drivers for AMD/Nvidia GPUs might exist. Otherwise, there is zero incentive for Nvidia, AMD, or Intel to help Qualcomm release a laptop or desktop with a capable dGPU and gain market share*. Plus, beyond the GPU apparently a lot of other parts of the SOC were found wanting by the reviewers, the NPU was fine, but if I remember correctly things like (de-)compression were woeful compared to Intel/AMD.
How many (and what revision) PCIe lanes do the Qualcom SoCs have? That may be a problem for discrete GPUs.
 
Slide from Qualcomm Investor Day 2024
View attachment 32886
Qualcomm says 30-50% of "AI Notebooks" will be ARM based essentially.
So NBC interpreted that slide as saying Qualcomm would have 30-50% of non-x86 AI notebooks, but in actual fact Qualcomm were saying that their serviceable addressable market will be 30-50% - either of the 90% or the full laptop market, unclear from the slides alone. Of revenue they expect to get roughly 1/9th of the total consumer silicon revenue for 2029, so just over 10% of the consumer market by revenue. That's still a far cry from what they were predicting originally (although I'd have to double check if what NBC said about that was accurate).

How many (and what revision) PCIe lanes do the Qualcom SoCs have? That may be a problem for discrete GPUs.

I believe they had enough bandwidth to run dGPUs, but @Artemis or @The Flame would know better than I. I do remember they had a goal of releasing Snapdragons with standard dGPUs, but obviously that hasn't happened. So I would assume they would engineer the SOC with that mind even if they so far haven't managed it.
 
Last edited:
I believe they had enough bandwidth to run dGPUs, but @Artemis or @The Flame would know better than I. I do remember they had a goal of releasing Snapdragons with standard dGPUs, but obviously that hasn't happened. So I would assume they would engineer the SOC with that mind even if they so far haven't managed it.
X Elite does have the PCIe lanes to support a dGPU. That's what Qualcomm claimed in October 2023, IIRC.

According to rumours, and corroborated by die shots, X Elite has;

2 PCIe G3 lanes for WiFi/BT
2 PCIe G3 lanes for 5G Modem
4 PCIe G4 lanes for SSD
8 PCIe G4 lanes for dGPU
 
Sales figures for Qualcomm's Snapdragon X chips are here.


June-September.
720,000 devices.
0.8% of all PCs sold in this period.

Thoughts?
The article author has issued a correction via an edit that the 720,000 number was for Q3 only.

Snapdeagon X laptops were released on June 18th, at the tail end of Q2.

Let's assume they sold 80,000 laptops between June 18th and July 1st.

Then Q2 + Q3 = 800,000 units

Now this 800k figure is almost entirely composed of premium consumer laptops using the Hamoa die (170 mm² N4P), which is what Snapdragon X Elite and Snapdragon X Plus 8-core are based on.

In September, Qualcomm announced the Snapdragon X Plus 8-core, which is based on the smaller and cheaper Purwa die (~125 mm² N4P?).

Their OEM partners have since;

1. Released budget laptops based on Snapdragon X Plus 8-core.
2. Released more business laptops

Due to these 2 factors, I think sales of Snapdragon laptops in 2024Q4 will be higher than in Q3.

I think there is a possibility that Qualcomm will be able to sell 2 million laptops in 2024 as a whole. They have sold 800k in Q2+Q3. They need to sell 1.2 million in Q4 to reach that milestone.
 
Back
Top