Why are you comparing an ultra-thin design to a CPU with PL2 at almost 90w? The notebookcheck tests show that the Zenbook runs up to 50% slower than the ProArt chassy.
It's definitely not a high performance chassis, despite being 16-inches. The default TDP is just 17 Watts; AMD asked reviewers to bump it up to 28W.
But this is what AMD sent out for review. Given the wide range of laptop TDPs out there, these review unit laptops can never cover the full spectrum. So it's more a reflection of what power level/form factor the chipmaker is choosing to prioritize in this generation.
What is the 90W laptop in this review? The other laptops are listed at 28W and 35W here. I did not see any indication of the power specifications of the ProArt on the other site, just some numbers provided. I strongly suspect that laptop is running at top TDP, 45-54W.
The HX 370 is in a different chassis pulling 80w sustained. Does that make the 35w vs 28w happening here more fair? I mean, if what the chips can draw elsewhere somehow matters here at all?
It matters if one wants to look at the maximum performance possible, regardless of power draw. But, in addition to what Ryan wrote, the attraction of the HX Series to me is the strong performance at lower power draws. I would have actually liked to see performance comparisons at 17 W, which IMHO is of special interest in such thin and light notebooks. The higher end (> 50 W) will be if interest for Strict Halo, which as far as I can tell is supposed to take on notebooks with smaller dGPUs.
From the benchmarks here, the 370 looks somewhat disappointing on the CPU front, with some losses to 8 cores Zen 4. A hybrid architecture is always a problem. I wonder if future scheduling changes will help or if the small 8MB L3 for the Zen 5c cores is a problem that can't be overcome.
The new GPU however looks like a good upgrade over the previous gen.
It's hybrid with different cache amounts, but it's also two CCXs instead of one after 3.5 generations of simple 8-cores. It's hard to say what's screwing it up.
The reviews I looked at didn't look too good for the GPU. Maybe it will do better with more power, but what it really needs is memory bandwidth. Hopefully AMD brings some Infinity Cache to its mainstream 128-bit APUs in the future.
I don't find it disappointing at all. It manages to perform as well or better (sometimes by a significant amount), while running on a 28W instead of 35W budget. I think it is a very important point when it comes to laptop CPUs.
It's pretty unfortunate that AT didn't also test the laptop with the 17W TDP. It would have been very interesting to see how big (or not) the impact on performances was. If the loss is limited, it may be an nice tradeoff for that kind of machine.
> There isn't a defined benchmark that tests AI performance from a level playing field
llama.cpp has a vulkan backend, and a couple of built in benchmarks. Its not the *fastest* backend (people on AMD/Intel just use rocm or sycl), but it's at least fair.
There are also vulkan ports of other models like esrgan.
Thank you for the review, nice work as always. Strix Point beating last-gen HS chip despite low cooling performance chassis is quite promising. It doesn't address the idle power, though, and in the mobile space it's very important as well.
By the way, do you plan to cover production laptops with Qualcomm X Elite/Plus? Yoga seems to be really interesting and only one with reasonable pricing of 32GB RAM option, but very few reviews were published so far.
I fully place the blame for that delay on Qualcomm and their launch partners. They have been quite "picky" about who gets what subnotebook or 2-in-1 and when. Which I believe was a mistake; they now face competition from AMD (like this one here) and soon also from Lunar Lake. At least AMD has apparently learned from their bad example, and sent out review units to many sites quickly.
A great piece of engineering on both sides for sure, very much where I was hoping AMD would land in terms of performance.
And I was very tempted until I saw the price: ASUS asks for more than €2000 on their direct sales shop and that is just way too much no matter what. Doing as good as the fruity cult doesn't mean you should charge their prices, especially the €400 markup for 8GB of RAM and 1TB of SSD.
I guess the ASUS exclusivity wasn't cheap, but I'll wait until prices have gone to saner levels, which I'm very sure they will.
And then I really wanted 64GB of RAM, which I believe is actually impossible with LPDDR5?
You might have to wait for the first LPCAMM2 notebooks for that. Micron and I believe Samsung are shipping memory modules that allow for 64 GB fast LP RAM. And yes, I agree on the pricing; it's, let's say, ambitious. And not helping with fast uptake. While Meteor Lake laptops are not as fast (but not that much slower either), they can be had, nicely equipped, for a lot less. It'll be interesting to see how Intel and launch partners will price Lunar Lake, and of course how Lunar Lake will perform.
By APUs I assume you mean the desktop G series chips. In which case, historically those have used the same die as the mobile parts. So the short answer is "most likely yes, they'll have 256-bit SIMDs".
Why, if you need full fledged AVX512 you can get the proper CPUs for that, people that do usually buy those anyway. The APU variant can still do AVX512 however, just not in one take cycle.
Good point. The thing is, apart from any power cost, it would be beneficial in a lot of places, considering that AVX-512 is being used more and more: dav1d, SVT-AV1, x265, and other places. Certainly, those who are doing heavy-duty work will buy the desktop CPUs; but if full-width AVX-512 can filter down to the lower parts, it will benefit many. And some folk, not interested in a discrete GPU, will buy an APU anyway.
I mean AMD still got you covered if you buy the desktop version, “not APU” it still has a small igpu in it, that’s plenty if you’re not gaming. And if you’re gaming the APUs aren’t much anyway. I think all in all the decision of AMD to save space and only give “proper” AVX512 to the usual desktop CPUs is fine, considering those who really need it will buy those or even higher parts anyway. What should be considered still, is the fact that it was saved in the APU/Laptop CPU because of economic reasons. You want full AVX512 there? Fine, chip would be bigger and thus CPUs pricier + bigger sizes reduces the amount of CPUs you can produce from single wafers = less CPUs overall = more pricing possibly = less availability. I think AMD did the right thing, Intel doesn’t even give you AVX512 in general. I mean you want AVX512? You still at least get the “light” version of it, it’s not like you’re not getting it at all, right m.
No, I agree. It's better than not having it, as on Intel, and I suppose it's just greed talking. What completely slipped my mind is that the ordinary "non-APUs" also come with graphics these days. So, my mistake there, and it changes the whole argument. In light of that, it makes more sense.
In one review I've seen, with the 28 Watt profile there was hardly any drop with the battery and I'd expect the "standard" and "whisper" profiles won't change any either. The 33 Watt "performance" mode most likely will default to something more 25 Watts without external power.
Twelve hours of sleep brought 4% battery drop in the same review.
The core-to-core latencies are disappointing. The Zen 4 apu has less than 21 ns worst case. Compare that to Zen 5+5c where you are lucky to get 21 ns at all. Most surprising to me is that threads within the same core have similar latency as between cores.
I couldn't find any Zen 4+4c core-to-core latencies. The issue might have already been present in previous generation where Zen 4c wasn't used for the top-of-the-line.
I was wondering the same. The division between the 2 core clusters is clearly obvious and this will make OS scheduling a problem. But even in the high-perf cluster the latencies are way worse than Zen4 ... also the fact that threads on the same core have the same latency as threads on different cores within the same cluster is unusual ... there seems to be some cache issue ...
wait wtf? rdna3.5 looks way faster than rdna3 (7940HS): in cp2077 it's 69% faster in returnal it's 58% faster in warhammer 3 it's 44% faster sure that's with 33% more CU's but then again, it's at lower power (28W vs 35W)
i didn't expect too much of strix halo, but with rdna3.5 this efficient it will undoubtedly make a real splash when it hits with ~2.5x this perf at ~70W
the new ryzen is running with lpddr5 7500MT/s which helps the graphics. if the 7940HS model is on 5600MT/s so-dimms then that's a huge difference right there ....
- better cores, arch is slightly upgraded - more cores 256 shaders more (1024 total), which now reached a respectable number that is 4 digits like a discrete gpu card - better vram and thus more bandwidth which also helps with performance like the other guy mentioned.
Wow this looks utterly disappointing. AMD's last chance to grab some mobile marketshare goes out the window. I won't be surprised if we see Lunar Lake taking over even the niche PC handheld marrket... ouch!
Ohhh, so I need local AI to better use Cloud AI. Next: AI in the keyboard to better use app AI to better use OS AI to better use Cloud AI. Plus an AI AI-aggregator ?
Seriously, what will these TOPS actually do for me ?
The questions are not dumb tho? What does this "AI" do? It "helps me with tasks"? How so? What does it do that I cannot already do myself? How does it change my life?
Khanan: AI, in its traditional sense, is a legitimate field. Artificial neural networks and the transformer architecture are advances on the road to strong AI. The idea, as science-fiction has always speculated, is to match the human mind; this has its own philosophical and ethical questions. Now, what's wrong with today's "AI" is that greedy corporations, seeing dollar signs, latched onto this technology, abusing an important field by trying to make money out of it, forcing it onto people. (And nobody is fond of being forced.) The truth is, what they're offering makes little difference to most people's tasks, so they're trying to invent tasks we can do with it, and understandably, people are fed up hearing and seeing it everywhere. It is like the TV infomercial selling you a bizzare product that will, they allege, make your life better. So, the comments of StormyParis and TheinsanegamerN are quite valid questions to ask.
TL;DR. Today's AI is, in my opinion, opportunistic monetisation of important science. And when the leather-jacket-wearing man says it's democracy, he's thinking about his bank balance.
That is currently my main question/complaint about all these "AI/NPU" PCs.
I'd like to know how much die area is used for the NPU in this APU or, for that matter, PHOENIX/Hawk and Meteor Lake? Are there any estimates from die shots?
It’s practically maxed out, these are high end mobile parts. However there will be a APU later with an extra big GPU called Strix Halo, the only difference will be that it will have a mid range desktop level GPU with over 32 CUs - comparable to console APUs.
You can find annotated die shots for Strix like here: https://www.techpowerup.com/325035/amd-strix-point... The die area spent for the NPU is quite significant, someone estimated around ~14mm². As for Phoenix, I've never seen a die shot of it (only Phoenix2 which doesn't feature the NPU). According to AMD, the NPU in Phoenix though was made up of 5x4 tiles, whereas the one in Strix is 8x4 tiles (with twice the MAC throughput per tile, and other improvments), as can be seen in the slides from AMD: https://www.anandtech.com/show/21469/amd-details-r... - so maybe it was like half the size? But that's just a crude guess.
It’s not, as per MS it needs to be at least 40 TOPS fast and AMD made sure to be faster than the competitors, it’s a marketing win. And copilot doesn’t run on CPU or GPU as that costs too much energy on the go, and in general it won’t use those, even if you’re on desktop and have no battery constraints.
> . The first to market in the consumer space with a dedicated AI engine in the notebook space was AMD with their Zen 4-based Phoenix Point or the Ryzen 7040 series
Ryzen 7040 released March 2023
Macbook M1 with dedicated ML chip released in 2020, that one already has dedicated ML chip with support for a slew of models, including text and image (stable diff); ....
And why no comparison with the Nvidia-powered ARM Windows PCs? Was it an AMD request?
Also, which model of MacBook Pro is included in the benchmarks? Apple's naming convention is currently very confusing and there are 14" MacBook Pros with M3, 2 flavors of M3 Pro and 2 flavors of M3 Max. Also, I do not see the point of comparing, weight-wise, this Asus with a 16" MacBook Pro with M3 Max, which is in a different league, performance and price-wise.
The argument to use LLVM for SPEC based on cross-platform ease is not a good one. Compilers don't perform equally competitively across different platforms. A compiler that works well on, say, apple silicon need not be optimal nor typical elsewhere. Fortunately, LLVM is fairly broadly liked, so hopefully this isn't a huge impact, but we don't know. Could be this alone pretty invalidates the spec2017 results
Additionally, particularly for *new* platforms the exact version of the compiler and whatever tuning it has had can make a difference; it's easy to test with a compiler that's older than the platform and thus tunes poorly. As it happens the LLVM version listed - 10.0.0 is simply ancient; over 4 years old - the current version is 18. A bit of web searching suggests that at least until fairly recently no znver5 tuning was merged into even dev LLVM; GCC happened to have tuning merged in march (not sure if this is typical between those two compilers).
Finally, the text implies you disabled avx512. I'm not sure why; even on zen4 with its half-rate execution path this can have considerable performance advantages; it pretty plausible it will on zen5 too (even on models with half-rate execution rates such as mobile chips).
All in all, I'm not sure what the point of the spec result is in this form. On the one hand, the vast majority of software run is compiled targetting old processors, so that's realistic. On the other hand, you already have benchmarks for that, and kind of the whole allure of a compiled-from-scratch benchmark like SPEC is that you can see what the platform is capable of natively, not when running legacy software. It's not too different from how it's interesting to see ARM chips running both emulated x86 code, but also see native results even if that's not 100% representative.
Still I guess an interesting result, but those caveats are large and worth mentioning IMHO. It's very likely a more reasonably tuned run could be significantly different for all of these chips, particularly newer ones that deviate most from the platform described by -march=x86-64 -mtune=core-avx2 -mfma -mavx -mavx2 - is that haswell? It's ancient anyhow.
What’s important isn’t the numbers, but the scaling over previous CPU’s. The fact is the iGPU is an absolute monster over the competition regardless of TDP.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
72 Comments
Back to Article
Dante Verizon - Sunday, July 28, 2024 - link
Why are you comparing an ultra-thin design to a CPU with PL2 at almost 90w? The notebookcheck tests show that the Zenbook runs up to 50% slower than the ProArt chassy.Ryan Smith - Sunday, July 28, 2024 - link
Sorry, which notebook are you referring to? We have multiple Zenbooks here.Terry_Craig - Sunday, July 28, 2024 - link
He's probably talking about the Zenbook in the review: https://www.notebookcheck.net/Asus-Zenbook-S-16-la...Strix performs much worse on the Zenbook than inside the ProArt, probably due to more aggressive power and temperature management.
Ryan Smith - Sunday, July 28, 2024 - link
It's definitely not a high performance chassis, despite being 16-inches. The default TDP is just 17 Watts; AMD asked reviewers to bump it up to 28W.But this is what AMD sent out for review. Given the wide range of laptop TDPs out there, these review unit laptops can never cover the full spectrum. So it's more a reflection of what power level/form factor the chipmaker is choosing to prioritize in this generation.
The Hardcard - Sunday, July 28, 2024 - link
What is the 90W laptop in this review? The other laptops are listed at 28W and 35W here. I did not see any indication of the power specifications of the ProArt on the other site, just some numbers provided. I strongly suspect that laptop is running at top TDP, 45-54W.So, like, a different comparison.
Terry_Craig - Sunday, July 28, 2024 - link
https://www.notebookcheck.net/AMD-Ryzen-9-7940HS-P...Depending on the model, the 7940HS goes up to 100w.
The Hardcard - Monday, July 29, 2024 - link
But, is the 7940HS pulling 100w in this review instead of the reported 35w? Otherwise,, what is the point of the complaint?https://www.ultrabookreview.com/69005-asus-proart-...
The HX 370 is in a different chassis pulling 80w sustained. Does that make the 35w vs 28w happening here more fair? I mean, if what the chips can draw elsewhere somehow matters here at all?
eastcoast_pete - Monday, July 29, 2024 - link
It matters if one wants to look at the maximum performance possible, regardless of power draw. But, in addition to what Ryan wrote, the attraction of the HX Series to me is the strong performance at lower power draws. I would have actually liked to see performance comparisons at 17 W, which IMHO is of special interest in such thin and light notebooks. The higher end (> 50 W) will be if interest for Strict Halo, which as far as I can tell is supposed to take on notebooks with smaller dGPUs.ET - Sunday, July 28, 2024 - link
From the benchmarks here, the 370 looks somewhat disappointing on the CPU front, with some losses to 8 cores Zen 4. A hybrid architecture is always a problem. I wonder if future scheduling changes will help or if the small 8MB L3 for the Zen 5c cores is a problem that can't be overcome.The new GPU however looks like a good upgrade over the previous gen.
nandnandnand - Sunday, July 28, 2024 - link
It's hybrid with different cache amounts, but it's also two CCXs instead of one after 3.5 generations of simple 8-cores. It's hard to say what's screwing it up.Phoronix's review was more positive for the CPU. The main benefit is power efficiency:
https://www.phoronix.com/review/amd-ryzen-ai-9-hx-...
The reviews I looked at didn't look too good for the GPU. Maybe it will do better with more power, but what it really needs is memory bandwidth. Hopefully AMD brings some Infinity Cache to its mainstream 128-bit APUs in the future.
kkilobyte - Sunday, July 28, 2024 - link
I don't find it disappointing at all. It manages to perform as well or better (sometimes by a significant amount), while running on a 28W instead of 35W budget. I think it is a very important point when it comes to laptop CPUs.It's pretty unfortunate that AT didn't also test the laptop with the 17W TDP. It would have been very interesting to see how big (or not) the impact on performances was. If the loss is limited, it may be an nice tradeoff for that kind of machine.
brucethemoose - Sunday, July 28, 2024 - link
> There isn't a defined benchmark that tests AI performance from a level playing fieldllama.cpp has a vulkan backend, and a couple of built in benchmarks. Its not the *fastest* backend (people on AMD/Intel just use rocm or sycl), but it's at least fair.
There are also vulkan ports of other models like esrgan.
Gavin Bonshor - Sunday, July 28, 2024 - link
Hey Bruce, thanks for the heads up on that. I'll check it outMakaanPL - Sunday, July 28, 2024 - link
Thank you for the review, nice work as always. Strix Point beating last-gen HS chip despite low cooling performance chassis is quite promising. It doesn't address the idle power, though, and in the mobile space it's very important as well.By the way, do you plan to cover production laptops with Qualcomm X Elite/Plus? Yoga seems to be really interesting and only one with reasonable pricing of 32GB RAM option, but very few reviews were published so far.
Ryan Smith - Sunday, July 28, 2024 - link
"By the way, do you plan to cover production laptops with Qualcomm X Elite/Plus?"Yes. It's taken entirely too long to get a sample in, but we should finally have one at the start of August.
eastcoast_pete - Monday, July 29, 2024 - link
I fully place the blame for that delay on Qualcomm and their launch partners. They have been quite "picky" about who gets what subnotebook or 2-in-1 and when. Which I believe was a mistake; they now face competition from AMD (like this one here) and soon also from Lunar Lake. At least AMD has apparently learned from their bad example, and sent out review units to many sites quickly.Khanan - Sunday, July 28, 2024 - link
Good article.Terry_Craig - Sunday, July 28, 2024 - link
What is most impressive is that even in this ultra-thin model the framerate is super stable.abufrejoval - Sunday, July 28, 2024 - link
A great piece of engineering on both sides for sure, very much where I was hoping AMD would land in terms of performance.And I was very tempted until I saw the price: ASUS asks for more than €2000 on their direct sales shop and that is just way too much no matter what. Doing as good as the fruity cult doesn't mean you should charge their prices, especially the €400 markup for 8GB of RAM and 1TB of SSD.
I guess the ASUS exclusivity wasn't cheap, but I'll wait until prices have gone to saner levels, which I'm very sure they will.
And then I really wanted 64GB of RAM, which I believe is actually impossible with LPDDR5?
eastcoast_pete - Monday, July 29, 2024 - link
You might have to wait for the first LPCAMM2 notebooks for that. Micron and I believe Samsung are shipping memory modules that allow for 64 GB fast LP RAM.And yes, I agree on the pricing; it's, let's say, ambitious. And not helping with fast uptake. While Meteor Lake laptops are not as fast (but not that much slower either), they can be had, nicely equipped, for a lot less.
It'll be interesting to see how Intel and launch partners will price Lunar Lake, and of course how Lunar Lake will perform.
GeoffreyA - Sunday, July 28, 2024 - link
I hope the mobile cores' having 256-bit SIMDs doesn't mean the APUs are going to also?Ryan Smith - Sunday, July 28, 2024 - link
By APUs I assume you mean the desktop G series chips. In which case, historically those have used the same die as the mobile parts. So the short answer is "most likely yes, they'll have 256-bit SIMDs".GeoffreyA - Monday, July 29, 2024 - link
Yes, that's what I thought. A bit of a disappointment, that.Khanan - Monday, July 29, 2024 - link
Why, if you need full fledged AVX512 you can get the proper CPUs for that, people that do usually buy those anyway. The APU variant can still do AVX512 however, just not in one take cycle.GeoffreyA - Monday, July 29, 2024 - link
Good point. The thing is, apart from any power cost, it would be beneficial in a lot of places, considering that AVX-512 is being used more and more: dav1d, SVT-AV1, x265, and other places. Certainly, those who are doing heavy-duty work will buy the desktop CPUs; but if full-width AVX-512 can filter down to the lower parts, it will benefit many. And some folk, not interested in a discrete GPU, will buy an APU anyway.Khanan - Monday, July 29, 2024 - link
I mean AMD still got you covered if you buy the desktop version, “not APU” it still has a small igpu in it, that’s plenty if you’re not gaming. And if you’re gaming the APUs aren’t much anyway. I think all in all the decision of AMD to save space and only give “proper” AVX512 to the usual desktop CPUs is fine, considering those who really need it will buy those or even higher parts anyway. What should be considered still, is the fact that it was saved in the APU/Laptop CPU because of economic reasons. You want full AVX512 there? Fine, chip would be bigger and thus CPUs pricier + bigger sizes reduces the amount of CPUs you can produce from single wafers = less CPUs overall = more pricing possibly = less availability. I think AMD did the right thing, Intel doesn’t even give you AVX512 in general. I mean you want AVX512? You still at least get the “light” version of it, it’s not like you’re not getting it at all, right m.GeoffreyA - Monday, July 29, 2024 - link
No, I agree. It's better than not having it, as on Intel, and I suppose it's just greed talking. What completely slipped my mind is that the ordinary "non-APUs" also come with graphics these days. So, my mistake there, and it changes the whole argument. In light of that, it makes more sense.Khanan - Monday, July 29, 2024 - link
Absolutely no problem, that’s why we are having these talks!id4andrei - Sunday, July 28, 2024 - link
What is the delta between on battery performance and plugged in?How is the battery drain under sleep?
abufrejoval - Monday, July 29, 2024 - link
In one review I've seen, with the 28 Watt profile there was hardly any drop with the battery and I'd expect the "standard" and "whisper" profiles won't change any either. The 33 Watt "performance" mode most likely will default to something more 25 Watts without external power.Twelve hours of sleep brought 4% battery drop in the same review.
Rudde - Sunday, July 28, 2024 - link
The core-to-core latencies are disappointing. The Zen 4 apu has less than 21 ns worst case. Compare that to Zen 5+5c where you are lucky to get 21 ns at all. Most surprising to me is that threads within the same core have similar latency as between cores.I couldn't find any Zen 4+4c core-to-core latencies. The issue might have already been present in previous generation where Zen 4c wasn't used for the top-of-the-line.
haplo602 - Monday, July 29, 2024 - link
I was wondering the same. The division between the 2 core clusters is clearly obvious and this will make OS scheduling a problem. But even in the high-perf cluster the latencies are way worse than Zen4 ... also the fact that threads on the same core have the same latency as threads on different cores within the same cluster is unusual ... there seems to be some cache issue ...bernstein - Sunday, July 28, 2024 - link
wait wtf? rdna3.5 looks way faster than rdna3 (7940HS):in cp2077 it's 69% faster
in returnal it's 58% faster
in warhammer 3 it's 44% faster
sure that's with 33% more CU's but then again, it's at lower power (28W vs 35W)
i didn't expect too much of strix halo, but with rdna3.5 this efficient it will undoubtedly make a real splash when it hits with ~2.5x this perf at ~70W
haplo602 - Monday, July 29, 2024 - link
the new ryzen is running with lpddr5 7500MT/s which helps the graphics. if the 7940HS model is on 5600MT/s so-dimms then that's a huge difference right there ....Khanan - Monday, July 29, 2024 - link
- better cores, arch is slightly upgraded- more cores 256 shaders more (1024 total), which now reached a respectable number that is 4 digits like a discrete gpu card
- better vram and thus more bandwidth which also helps with performance like the other guy mentioned.
dwade123 - Sunday, July 28, 2024 - link
Wow this looks utterly disappointing. AMD's last chance to grab some mobile marketshare goes out the window. I won't be surprised if we see Lunar Lake taking over even the niche PC handheld marrket... ouch!abufrejoval - Monday, July 29, 2024 - link
Do you at least get paid for comments like this?Khanan - Monday, July 29, 2024 - link
No i think he lacks the knowledge to understand the review.TheinsanegamerN - Wednesday, July 31, 2024 - link
This is your brain on Intel Rap- *blue screen*Duncan Macdonald - Monday, July 29, 2024 - link
Could you run at least one of the tests at the 4 different power levels that this laptop supports to show the impact of power on performance.haplo602 - Monday, July 29, 2024 - link
I second this request. Would be nice to see at least a brief analysis how the CPU behaves on different power levels.abufrejoval - Monday, July 29, 2024 - link
Could you break out the iGPU power consumption on the gaming benchmarks?Do you see a Wattage ceiling for the iGPU and do you see it getting priority over the CPU when both are maximized (e.g. combo of Furmark and Prime95)?
StormyParis - Monday, July 29, 2024 - link
So much written about NPU and TOPS and I still have no idea what this will do for me.Khanan - Monday, July 29, 2024 - link
If you’re not into chat gpt- nothing.StormyParis - Monday, July 29, 2024 - link
I'm using it - a bit - on my 11 yo old PC, so I don't a need new fancy NPU with lotso of TOPS for that. Is there anything else I'm actually missing ?Khanan - Monday, July 29, 2024 - link
it's a bit different than chatgpt as it's assisted computing on the fly.StormyParis - Tuesday, July 30, 2024 - link
What does this even mean ? Clippy II ?Khanan - Tuesday, July 30, 2024 - link
Clippy2 but he’s kinda smart (and not an faq with graphics animation) and can actually help you with various things like chat gpt.StormyParis - Tuesday, July 30, 2024 - link
Ohhh, so I need local AI to better use Cloud AI. Next: AI in the keyboard to better use app AI to better use OS AI to better use Cloud AI. Plus an AI AI-aggregator ?Seriously, what will these TOPS actually do for me ?
GeoffreyA - Tuesday, July 30, 2024 - link
TOPS will make your computer go fast, like stickers on the car :)StormyParis - Tuesday, July 30, 2024 - link
Right now they mostly make noise, like a muffler on a car ^^Khanan - Tuesday, July 30, 2024 - link
From asking dumb questions to pretending to be an expert.TheinsanegamerN - Wednesday, July 31, 2024 - link
The questions are not dumb tho? What does this "AI" do? It "helps me with tasks"? How so? What does it do that I cannot already do myself? How does it change my life?Khanan - Wednesday, July 31, 2024 - link
Do you have a brain? Then use it. If not... well.GeoffreyA - Thursday, August 1, 2024 - link
Khanan: AI, in its traditional sense, is a legitimate field. Artificial neural networks and the transformer architecture are advances on the road to strong AI. The idea, as science-fiction has always speculated, is to match the human mind; this has its own philosophical and ethical questions. Now, what's wrong with today's "AI" is that greedy corporations, seeing dollar signs, latched onto this technology, abusing an important field by trying to make money out of it, forcing it onto people. (And nobody is fond of being forced.) The truth is, what they're offering makes little difference to most people's tasks, so they're trying to invent tasks we can do with it, and understandably, people are fed up hearing and seeing it everywhere. It is like the TV infomercial selling you a bizzare product that will, they allege, make your life better. So, the comments of StormyParis and TheinsanegamerN are quite valid questions to ask.GeoffreyA - Thursday, August 1, 2024 - link
TL;DR. Today's AI is, in my opinion, opportunistic monetisation of important science. And when the leather-jacket-wearing man says it's democracy, he's thinking about his bank balance.Khanan - Friday, August 2, 2024 - link
This isn’t the issue. I’m just not taking my time and effort to reply to trolls, maybe you missed him behaving in a very odd way.eastcoast_pete - Monday, July 29, 2024 - link
That is currently my main question/complaint about all these "AI/NPU" PCs.I'd like to know how much die area is used for the NPU in this APU or, for that matter, PHOENIX/Hawk and Meteor Lake? Are there any estimates from die shots?
Khanan - Monday, July 29, 2024 - link
you should know one thing: the cpu or gpu wouldn't be any bigger without the npu, as those are maxed out.eastcoast_pete - Monday, July 29, 2024 - link
How do you know that to be the case?Khanan - Tuesday, July 30, 2024 - link
It’s practically maxed out, these are high end mobile parts. However there will be a APU later with an extra big GPU called Strix Halo, the only difference will be that it will have a mid range desktop level GPU with over 32 CUs - comparable to console APUs.mczak - Tuesday, July 30, 2024 - link
You can find annotated die shots for Strix like here:https://www.techpowerup.com/325035/amd-strix-point...
The die area spent for the NPU is quite significant, someone estimated around ~14mm².
As for Phoenix, I've never seen a die shot of it (only Phoenix2 which doesn't feature the NPU). According to AMD, the NPU in Phoenix though was made up of 5x4 tiles, whereas the one in Strix is 8x4 tiles (with twice the MAC throughput per tile, and other improvments), as can be seen in the slides from AMD: https://www.anandtech.com/show/21469/amd-details-r... - so maybe it was like half the size? But that's just a crude guess.
zamroni - Tuesday, July 30, 2024 - link
the npu is overkill.i'd rather have bigger igpu or more cpu caches.
gpu can do anything that npu can anyway
Khanan - Tuesday, July 30, 2024 - link
It’s not, as per MS it needs to be at least 40 TOPS fast and AMD made sure to be faster than the competitors, it’s a marketing win. And copilot doesn’t run on CPU or GPU as that costs too much energy on the go, and in general it won’t use those, even if you’re on desktop and have no battery constraints.jmke - Tuesday, July 30, 2024 - link
> . The first to market in the consumer space with a dedicated AI engine in the notebook space was AMD with their Zen 4-based Phoenix Point or the Ryzen 7040 seriesRyzen 7040 released March 2023
Macbook M1 with dedicated ML chip released in 2020, that one already has dedicated ML chip with support for a slew of models, including text and image (stable diff); ....
jeromec - Tuesday, July 30, 2024 - link
Did I miss the battery test?And why no comparison with the Nvidia-powered ARM Windows PCs? Was it an AMD request?
Also, which model of MacBook Pro is included in the benchmarks? Apple's naming convention is currently very confusing and there are 14" MacBook Pros with M3, 2 flavors of M3 Pro and 2 flavors of M3 Max.
Also, I do not see the point of comparing, weight-wise, this Asus with a 16" MacBook Pro with M3 Max, which is in a different league, performance and price-wise.
lmcd - Tuesday, July 30, 2024 - link
Nvidia-powered ARM Windows PCs?I'm so sorry to inform you that the last Nvidia ARM Windows PC was the Surface RT 2, which launched with a Tegra 4. We have not seen anything since.
Hopefully that will change, but seriously: what are you talking about?
nandnandnand - Tuesday, July 30, 2024 - link
There's rumors of a MediaTek/Nvidia chip coming within a couple years, but jeromec meant the Qualcomm Snapdragon X Elite.patel21 - Tuesday, July 30, 2024 - link
Minor Correction:>> Core Ultra 7 155H, which is a 6P+8E/22T chip
It should be:
Core Ultra 7 155H, which is a 6P+8E/20T chip
emn13 - Wednesday, July 31, 2024 - link
The argument to use LLVM for SPEC based on cross-platform ease is not a good one. Compilers don't perform equally competitively across different platforms. A compiler that works well on, say, apple silicon need not be optimal nor typical elsewhere. Fortunately, LLVM is fairly broadly liked, so hopefully this isn't a huge impact, but we don't know. Could be this alone pretty invalidates the spec2017 resultsAdditionally, particularly for *new* platforms the exact version of the compiler and whatever tuning it has had can make a difference; it's easy to test with a compiler that's older than the platform and thus tunes poorly. As it happens the LLVM version listed - 10.0.0 is simply ancient; over 4 years old - the current version is 18. A bit of web searching suggests that at least until fairly recently no znver5 tuning was merged into even dev LLVM; GCC happened to have tuning merged in march (not sure if this is typical between those two compilers).
Finally, the text implies you disabled avx512. I'm not sure why; even on zen4 with its half-rate execution path this can have considerable performance advantages; it pretty plausible it will on zen5 too (even on models with half-rate execution rates such as mobile chips).
All in all, I'm not sure what the point of the spec result is in this form. On the one hand, the vast majority of software run is compiled targetting old processors, so that's realistic. On the other hand, you already have benchmarks for that, and kind of the whole allure of a compiled-from-scratch benchmark like SPEC is that you can see what the platform is capable of natively, not when running legacy software. It's not too different from how it's interesting to see ARM chips running both emulated x86 code, but also see native results even if that's not 100% representative.
Still I guess an interesting result, but those caveats are large and worth mentioning IMHO. It's very likely a more reasonably tuned run could be significantly different for all of these chips, particularly newer ones that deviate most from the platform described by -march=x86-64 -mtune=core-avx2 -mfma -mavx -mavx2 - is that haswell? It's ancient anyhow.
The_Assimilator - Thursday, August 1, 2024 - link
Please can you update the "Cinebench 2024 (multi-thread)" chart on the last page to include the core and thread counts of the tested CPUs.Samus - Thursday, August 1, 2024 - link
What’s important isn’t the numbers, but the scaling over previous CPU’s. The fact is the iGPU is an absolute monster over the competition regardless of TDP.