Nuvia: don’t hold your breath

The big day approaches….soonish.


X gone give it you! Maybe.

Makes me sad, since the first full-blown chip i helped design was the X704. X was already taken, you Qualcomm ghouls.
 
So Qualcomm just announced their “Snapdragon Elite X” SoC which should be coming mid-24.

Some details:

4nm
12 Performance Cores running at 3.8, with a turbo up to 4.2 (for 2 cores?)
No Efficiency Cores at al!
4.6 TFlop GPU (at unspecified bit depth)
46 TOPS (INT4) NPU
8 x 16-bit LPDDR5x-8533 136GB/sec

Hidden in the presentation is the fact that not only have G Williams and other members of the chip team from Apple left for QC, but also the chart designer from the M1 presentations!
Look at these beauties:
1698176662515.png


1698176639909.png

1698176675996.png


Heavy focus on multi threaded performance (+50% vs an unspecified Arm competitor) and zero mention of single core performance. Not sure what that tells us, but it doesn’t seem great.

More details here


EDIT: wow, seems like I was wrong. Single core performance was mentioned. Very impressive Apple has some serious competition.
1698177358658.png

Yikes
1698177493213.png
 
Last edited:
So Qualcomm just announced their “Snapdragon Elite X” SoC which should be coming mid-24.

Some details:

4nm
12 Performance Cores running at 3.8, with a turbo up to 4.2 (for 2 cores?)
No Efficiency Cores at al!
4.6 TFlop GPU (at unspecified bit depth)
46 TOPS (INT4) NPU
8 x 16-bit LPDDR5x-8533 136GB/sec

Hidden in the presentation is the fact that not only have G Williams and other members of the chip team from Apple left for QC, but also the chart designer from the M1 presentations!
Look at these beauties:





Heavy focus on multi threaded performance (+50% vs an unspecified Arm competitor) and zero mention of single core performance. Not sure what that tells us, but it doesn’t seem great.

More details here


EDIT: wow, seems like I was wrong. Single core performance was mentioned. Very impressive Apple has some serious competition.
Suspicious of the phrase “peak”:
 
I hope Qualcomm can shake up the x86 laptop scene with this product! Curious what the actual performance and power consumption will be. They seem to choose the wording fairly carefully, e.g. they highlight that they can match M2 Max at 30% lower power but they avoid saying how much power they need to produce that 15% higher result. And there are a lot of little bits like that. Still, achieving the same IPC as Firestorm and maintaining the same frequency at lower power is very impressive nevertheless. They mentioned that there will be more technical data tomorrow, maybe there will be more detailed info there.

Less impressive are the weak GPU and narrow memory interface. But I suppose there was no way around if they want to keep it financially viable. Also, “50% higher multicore compared to M2” is a bit disappointing for a 12-core design running against a 4+4 core design that uses 2x less power… especially after the striking single core performance claims. But again, not surprising, as they likely have to reduce the frequency quite a bit when running a full multicore workload.
 
I hope Qualcomm can shake up the x86 laptop scene with this product! Curious what the actual performance and power consumption will be. They seem to choose the wording fairly carefully, e.g. they highlight that they can match M2 Max at 30% lower power but they avoid saying how much power they need to produce that 15% higher result. And there are a lot of little bits like that. Still, achieving the same IPC as Firestorm and maintaining the same frequency at lower power is very impressive nevertheless. They mentioned that there will be more technical data tomorrow, maybe there will be more detailed info there.

Less impressive are the weak GPU and narrow memory interface. But I suppose there was no way around if they want to keep it financially viable. Also, “50% higher multicore compared to M2” is a bit disappointing for a 12-core design running against a 4+4 core design that uses 2x less power… especially after the striking single core performance claims. But again, not surprising, as they likely have to reduce the frequency quite a bit when running a full multicore workload.
Good points. Do we think this event is partly motivation for Apple‘s upcoming one?
 
So Qualcomm just announced their “Snapdragon Elite X” SoC which should be coming mid-24.

Some details:

4nm
12 Performance Cores running at 3.8, with a turbo up to 4.2 (for 2 cores?)
No Efficiency Cores at al!
4.6 TFlop GPU (at unspecified bit depth)
46 TOPS (INT4) NPU
8 x 16-bit LPDDR5x-8533 136GB/sec

Hidden in the presentation is the fact that not only have G Williams and other members of the chip team from Apple left for QC, but also the chart designer from the M1 presentations!
Look at these beauties:
View attachment 26887

View attachment 26886
View attachment 26888

Heavy focus on multi threaded performance (+50% vs an unspecified Arm competitor) and zero mention of single core performance. Not sure what that tells us, but it doesn’t seem great.

More details here


EDIT: wow, seems like I was wrong. Single core performance was mentioned. Very impressive Apple has some serious competition.
View attachment 26892
Yikes
View attachment 26894
Weird I couldn’t find that anywhere, and Anandtech they specifically said single core wasn’t mentioned. Where did you find that picture/slide?
 
I’ve seen that mentioned. Do you think they are referring to multi core performance with the 30% less power figure. These do seem very misleading if true.
No idea. Maybe it hits that figure only in short bursts. There are no efficiency cores, yes? If so, then the power number is even more perplexing - yeah we can hit this number in short bursts while chewing less power, but actually using the machine burns more power?
 
Oryon looks super promising for sure. If it meets all the claims presented well… they nailed it 👏
I’m willing to take the claims at face value for now. I don’t think they’d make such pointed claims about performance and power without having something genuinely great in the labs.
 
Good points. Do we think this event is partly motivation for Apple‘s upcoming one?

I doubt it. These things take a while and I doubt Apple would rush to challenge a product that doesn’t really threaten it. At any rate, I have little doubt that M3 Pro will significantly outperform the Elite X in every key metric, and it looks like it’s going to be half a year earlier to market as well.
 
Oryon looks super promising for sure. If it meets all the claims presented well… they nailed it 👏
I’m willing to take the claims at face value for now. I don’t think they’d make such pointed claims about performance and power without having something genuinely great in the labs.

Oh, there is no doubt about it, they did a great job. What’s interesting is that we now have three major ARM vendors converging on similar IPC and performance levels.
 
I doubt it. These things take a while and I doubt Apple would rush to challenge a product that doesn’t really threaten it. At any rate, I have little doubt that M3 Pro will significantly outperform the Elite X in every key metric, and it looks like it’s going to be half a year earlier to market as well.
Even single core? ~3250 seems pretty good.
 
I doubt it. These things take a while and I doubt Apple would rush to challenge a product that doesn’t really threaten it. At any rate, I have little doubt that M3 Pro will significantly outperform the Elite X in every key metric, and it looks like it’s going to be half a year earlier to market as well.
While I know N3 to N4P isn’t as big a difference one might expect, it’s still impressive if these numbers hold up.
 
Oh, there is no doubt about it, they did a great job. What’s interesting is that we now have three major ARM vendors converging on similar IPC and performance levels.
Yeah it’s pretty amazing 😀
The future looks bright for Arm in PCs
More strong Arms taking on the x86 dinosaurs makes me super happy
 
So Qualcomm just announced their “Snapdragon Elite X” SoC which should be coming mid-24.

Some details:

4nm
12 Performance Cores running at 3.8, with a turbo up to 4.2 (for 2 cores?)
No Efficiency Cores at al!
4.6 TFlop GPU (at unspecified bit depth)
46 TOPS (INT4) NPU
8 x 16-bit LPDDR5x-8533 136GB/sec

Hidden in the presentation is the fact that not only have G Williams and other members of the chip team from Apple left for QC, but also the chart designer from the M1 presentations!
Look at these beauties:
View attachment 26887

View attachment 26886
View attachment 26888

Heavy focus on multi threaded performance (+50% vs an unspecified Arm competitor) and zero mention of single core performance. Not sure what that tells us, but it doesn’t seem great.

More details here


EDIT: wow, seems like I was wrong. Single core performance was mentioned. Very impressive Apple has some serious competition.
View attachment 26892
Yikes
View attachment 26894

So 30% less power and 70% less power would mean, that i9-13980HK consumes single core only 2.3x more than M2 Max single core. But isn’t it much more?
 
I find the slides more confusing the more I look at them. I’m unclear if they are matching single core or multi core at 30% less power.

Edited for clarity hopefully.
 
Back
Top