That's what Geekerwan said about Adreno 700 series, let's wait until real system arrives then we will know..I doubt that. Almost certainly just driver issue since 8 Gen 2 has ray tracing support.
Are ALUs the same across different GPU architectures?
You are comparing old products to unreleased ones.The scores for the 23 W TDP configuration are much more impressive than the higher TDP version. Intel and AMD cannot catch it in CPU performance unless they up the core count of the big cores with their next mobile CPUs.
What's even more impressive is the iGPU performance. To basically match AMD, who have the most mature iGPU drivers, with a first gen product is really impressive. Granted it is just one game.
This wasn't the case previously.Whereas Qualcomm is basically using the same GPU with same core count in the laptop chip as the phone chip.
Their DX12 driver is likely trash tier so I doubt that they are getting maximum out of it. Still, surprisingly good performance in Control all things considered (x86 emulation, drivers and probably zero optimisations from game side).Well well well.
I was expecting the X Elite's GPU to be much more stronger than Apple M2. Kinda disappointed.
And the M2 Max GPU is on another league
Zen 5 mobile won't be scoring 90% of AMD's fastest mobile CPU (the 16 core 7945HX) in Geekbench 6 MT at 20-25 W.Wait for Zen 5 and Arrow Lake.
Nah, they just had to fit something around the Nuvia cores.So I am guessing Qualcomm is going to be leaning in on dGPUs instead of taking the Apple approach of having humongous iGPUs.
I do not like this.
dGPUs with what drivers?So I am guessing Qualcomm is going to be leaning in on dGPUs instead of taking the Apple approach of having humongous iGPUs.
I do not like this.
GB6 is a horrible benchmark for discussing MT performance. It scales exceptionally poorly with core count.Zen 5 mobile won't be scoring 90% of AMD's fastest mobile CPU (the 16 core 7945HX) in Geekbench 6 MT at 20-25 W.
And Zen 5/ARL mobile are even further out than this Snapdragon in terms of launch window.
And of course the 1536 ALU Adreno in SD8G3 is a very formidable GPU (even better than Apple A17).
Not 3rd party but some users seem to love it: https://www.lenovo.com/us/en/p/lapt...kpad--x13s-(13-inch-snapdragon)/21bxcto1wwus1Find me one good 8cx Gen 3 laptop review aside from Notebookcheck.
It scales perfectly fine for the kind of tasks it is designed to measure which is much more relevant for the vast majority of laptop users.GB6 is a horrible benchmark for discussing MT performance. It scales exceptionally poorly with core count.
On more performance-oriented tests, the Snapdragon just didn't compete. It took the ThinkPad X13s 89 minutes to run a big spreadsheet in Excel (a native ARM application), compared with 41 minutes for the current (Intel-based) X1 Carbon Gen 10. On the X13s, converting a large video using Handbrake took 3 hours and 38 minutes with the native version of the app; and 4 hours and 5 minutes in an emulated 64-bit version, compared with 1 hour and 55 minutes on the X1 Carbon. And a big Matlab model just simply would not run. Bottom line: this isn't the right machine for you if you run heavy-duty applications.
So if you're aware of that, why would you make a comparison you know isn't a valid one? You know perfectly well that GB6 doesn't scale well all the way up to 16 cores, and that a few faster cores are better suited to the benchmark. Nuvia themselves are claiming ~25% faster than PHX in MT, and only 5.5% faster than the 13800H. Why not use those numbers instead?It scales perfectly fine for the kind of tasks it is designed to measure which is much more relevant for the vast majority of laptop users.
Also, I didn't want to comment on availability.
Thanks for the Cinebench 24 comparison. That's impressive indeed.IMO the Cinebench 24 is a lot more impressive as it's a much more taxing MT test.
I just ran it on ma 8+2 M1 Pro work laptop and the MT score was about 819 points.
While the M2 Pro has 2 more small cores, it still won't reach anywhere near the Qualcomm scores:
- 997 for the smaller reference design
- 1227 for the bigger one.
Even M2 Ultra still scores around 1600 after all (2x CPUs on one package).
The single threaded results are good too.
- The 4.3 Ghz design is the same as 13900KS
- The 4.0 Ghz smaller design is on par with the 7950X
And while it's true, that these should rather be compared against the upcoming M3 Pro, Meteor Lake and Zen 5 - it's still a very solid result from the first implementation.
All of the competitors need to do a very good job with next-gen chips to really outperform it.
I would really like to see a power-unconstrained Desktop version as this also supports external GPUs. Too bad it's a pipedream for now.
EDIT:
Ok when closing everything I managed a 819 point run on the M1 Pro.
Corps buy Apple MacBooks.Corporate users are not sensitive about prices, and they are likely to buy in volumes, thus I am expecting Qualcomm will surpass AMD to become No.2 key player in commercial notebook market by end of 2025
Jon Masters said:Gerard Williams only knows how to be the best. If you take him on, you had better be prepared to lose. Not surprised that Oryon is dunking on the competition. That’s why we did NUVIA
Another random Gerard story: when he got married, his wife’s dress looked absolutely stunning. He sewed that dress himself, by hand, easily beating high end designers out there
That guy can’t be another other than the very best at everything he tries to do
And remember: they’re just getting started. Be prepared for insanely ridiculous gen on gen perf uplift
...
One more random Gerard story: he worked on many of the earlier 32-bit Arm holdings designs. In at least one case, he was still perfecting the design *two years* after it shipped. For no reason other than that he can’t stop himself from trying to make things better
Give that guy enough hours in the day, and the result is dangerous
I’ve witnessed him hacking on perf models at 2am on several occasions