Nuvia: don’t hold your breath

[Jimmyjames]

Oh Asus.

Diablo IV doesn’t run on macOS. Is this a Crossover comparison??

ASUS Zenbook A16 (UX3607) | Copilot+ PCs | ASUS UK

ASUS Zenbook A16: Experience ultra-light 16" laptop with Snapdragon X2 Elite Extreme, 3K OLED display, and over 21hr battery life for multitasking on the go.

www.asus.com

 

[dada_dave]

Oh Asus.
View attachment 38447

Diablo IV doesn’t run on macOS. Is this a Crossover comparison??

ASUS Zenbook A16 (UX3607) | Copilot+ PCs | ASUS UK

ASUS Zenbook A16: Experience ultra-light 16" laptop with Snapdragon X2 Elite Extreme, 3K OLED display, and over 21hr battery life for multitasking on the go.

www.asus.com

View attachment 38449

On all of them (I see this in you first picture too but even for Diablo IV which looks okay in your picture) for me the bars are all screwed up - both going past the screen.

I *think* Diablo is also running on Prism for WoA? (not as much translation as for Xover) If memory serves when Geekerwan tested the first generation, the Apple chips were generally faster for gaming under emulation than the Qualcomm chips even though the Mac chips had more to emulate, so I guess this is an upgrade? Of course this is the M5.

 

[Yoused]

for me the bars are all screwed up

If you go to the link, the image shows the X2 Elite as 26% faster than the M5, based primarily on multi-core scores. Of course, the X2 accomplishes this with a mere 18 cores, compared to the M5's massive 10 core complement (it is the base M5, with genuine E-cores).

 

[dada_dave]

If you go to the link, the image shows the X2 Elite as 26% faster than the M5, based primarily on multi-core scores. Of course, the X2 accomplishes this with a mere 18 cores, compared to the M5's massive 10 core complement (it is the base M5, with genuine E-cores).

Aye same with the Diablo comparison too, the X2 Elite's GPU is I believe is 1.6x the size of the M5's (and I think faster clocks too). I guess for the consumer, it'll depend on price which I don't think is known yet (also this is a 16" device which the base M5 doesn't even go into a competing form factor - although it is a smaller, lighter 16" than Apple's).

But ... for me, based on all that only winning by 30% on a game that has to go through one (WoA) vs two (Mac) translation layers* isn't that impressive.

*and you could in theory count it as 4 translation layers for the Mac: x86 -> Arm, 4k -> 16k page tables, Windows -> macOS, DirectX -> Metal, I was combining the first two and last two together

Also when I go to the link, none of the bars render correctly and I don't get any of the ratios, they all look like @Jimmyjames' first picture.

 

[NotEntirelyConfused]

Asus' marketing is bullshit, but in the end it all comes down to money and performance.

Here, we care about the tech, but for the typical customer buying a laptop, core counts, clocks, etc. are largely irrelevant. Gaming, productivity, other types of performance - and how much you pay for it - is what matters.

This was true the last time QC tried to break into the market and they fell on their faces because they priced themselves out of their target market. They had dreams of competing with Apple for high-dollar purchases, which were largely insane.

This is, so far, pretty much a replay of what happened with their first attempt. They are using more silicon to compete (and win) on multicore against a smaller Apple chip. This could be a successful strategy, at least to some extent, if they actually priced their products accordingly. They didn't the first time, and if I had to bet, I would bet that they won't this time either.

I don't think they're going to find a lot of success. The first time around, they had a big window, with no serious competition on the x86 Windows side, for laptops with high performance, low heat/noise, and long battery life. Once Lunar Lake arrived, the game changed.

 

[MerryCherry]

Aye same with the Diablo comparison too, the X2 Elite's GPU is I believe is 1.6x the size of the M5's (and I think faster clocks too).

Size, as in die area?

We don't have dieshots for X2 Elite, but based on 8 Elite Gen 5 (which uses the same GPU architecture), we can estimate it to be about 30 mm². M5 is similar or larger than that, I believe.

This is, so far, pretty much a replay of what happened with their first attempt. They are using more silicon to compete (and win) on multicore against a smaller Apple chip.

The 12-core X2E has comparable multicore to the base M5 (17k GB6, ~1200 CB2024), and it's die size is also presumably similar to base M5.

 

[MerryCherry]

This is crazy. The dedication of Linux fans.

 

[leman]

Asus' marketing is bullshit, but in the end it all comes down to money and performance.

That is true. If this Asus model competes price-wise with the M5 machines, it’s absolutely fair to compare it to M5. So the price and positioning will be decisive.

I had some difficulty finding price information for UX3607. A Swiss retailer lists it for CHF 2400,- same price as a M5 Pro 14” in the comparable RAM/SSD configuration. Of course, the product is not yet practically available, so pricing might change.

At the same time the form factor is more similar to Air. It’s a strange one. I wonder who the target customer is. I feel that folks looking for an ultracompact laptop are better served with an Air, while those looking for performance would prefer Pro/Max.

 

[mr_roboto]

Size, as in die area?

We don't have dieshots for X2 Elite, but based on 8 Elite Gen 5 (which uses the same GPU architecture), we can estimate it to be about 30 mm². M5 is similar or larger than that, I believe.

There is no way 30mm^2 is anywhere near correct; that's far too small for chips similar to either base M5 or X2 Elite.

I don't know exactly where M5 is, but Apple's base M SoCs have been in the 120 to 150 mm^2 range.

As for X2 Elite, some searching brought me to a Hothardware article which estimates 220mm^2 based on photogrammetry.

 

[dada_dave]

There is no way 30mm^2 is anywhere near correct; that's far too small for chips similar to either base M5 or X2 Elite.

I don't know exactly where M5 is, but Apple's base M SoCs have been in the 120 to 150 mm^2 range.

As for X2 Elite, some searching brought me to a Hothardware article which estimates 220mm^2 based on photogrammetry.

I think he’s referring to just the GPU portion of the SOC.

I found it referenced here:

Size, as in die area?

Size as in reported ALU count (1.6x) and bus size (1.5x) + clock speed since I hadn't seen anything else at the time.

We don't have dieshots for X2 Elite, but based on 8 Elite Gen 5 (which uses the same GPU architecture), we can estimate it to be about 30 mm². M5 is similar or larger than that, I believe.

30mm^2 still seems rather small though, @MerryCherry do you have a link to original estimate?

Edit:

the original X Elite GPU according to this annotation is about 24-27mm^2 (depending on what you count):

Qualcomm Snapdragon X die shot reveals massive CPU cores with huge caches

GPU is quite modest though.

www.tomshardware.com

33% bigger would be in the 30s and then with a node shift, yeah ... I suppose it's possible.

 

[MerryCherry]

There is no way 30mm^2 is anywhere near correct; that's far too small for chips similar to either base M5 or X2 Elite.

I was referencing only the GPU area.

I don't know exactly where M5 is, but Apple's base M SoCs have been in the 120 to 150 mm^2 range.

According to Kurnalsalts, M4 is 168 mm².

M5 will be similar or larger.

 

[MerryCherry]

According to Kurnalsalts again,

A19 Pro GPU (6 core) is 21 mm².

M5 is 10 core. Assuming linear scaling, M5 GPU will be 67% larger at 35 mm².

(He hasn't published a dieshot for M5 yet, but there is one for M4).

 

[MerryCherry]

Based on his 8 Elite Gen 5 die shot, the full chip is 126 mm², and by pixel counting I estimate the Adreno 840 GPU to be 24 mm².

The X2 Elite's Adreno X2-90 GPU has 33% more CUs, so it can be estimated to be 32 mm².

 

[dada_dave]

I was referencing only the GPU area.

According to Kurnalsalts, M4 is 168 mm².

M5 will be likely a bit larger, since there is no node shrink from M4 to M5.

From the same article I linked to for the original X Elite, Piglin had 166mm^2 (what I used) for the M4, but close enough

I was thinking about how Qualcomm fit more ALUs into a smaller space than Apple and then I remembered that similarly, relative to Nvidia, Apple does like to go for chonky GPU cores where for the ALU (FP32) count, they have quite low compute density. This is in large part because Apple gives Int and FP16 their own separate pipelines (and maybe for the M5 2xFP16 pipelines -> not sure yet how they doubled FP16 throughput, has someone sussed that out?). Also in comparison specifically to the Snapdragon, from the Reddit post, still no matrix cores yet on the X2 Elite Adreno GPU? Don't know how much room those take up specifically in Apple's GPU but it's probably not nothing. Anyway, back to the ALU die density, Apple does this primarily to lower power, especially for FP16 calls, and (since the M3) increase throughput when you can call both FP16 and FP32 simultaneously. Right now, I honestly can't remember the details of Adreno, but I vaguely recall it doubled up its ALU duties. Also not sure how GPU cache sizes compare.

So admittedly per die size, the X2 Elite Extreme's GPU may indeed be around the M5 (and that is certainly important), but everything else about the GPU is closer to what I would expect from the binned M5 Pro or at least in between the two - e.g. power draw. And the game comparison is still between something going through a single translation layer vs multiple translation layers. Then again, I don't know how many native games WoA has at the moment, so maybe they're restricted in what they can actually compare to (even if Mac has more it's not exactly an overabundance either) and if they can't do native vs native, well I guess they don't have much choice but to compare translated vs translated.

 

[MerryCherry]

So... GPU area;

A19 Pro = 21 mm²
8 Elite Gen 5 = 24 mm²

GPU能效曲线

www.socpk.com

In terms of traditional graphics performance/efficiency the two are almost identical.

But A19 Pro GPU is 12.5% smaller, while also having dedicated RT cores for BVH traversal and Tensor cores (neither of which the Adreno has).

So in terms of area efficiency, Apple's GPU seems to be superior.

There was a time when Qualcomm was the mobile GPU king, but Apple's aggressive iteration in the recent years seemed to be have paid off.

 

[dada_dave]

So... GPU area;

A19 Pro = 21 mm²
8 Elite Gen 5 = 24 mm²
View attachment 38455

GPU能效曲线

www.socpk.com

In terms of traditional graphics performance/efficiency the two are almost identical.

But A19 Pro GPU is 10% smaller, while also having dedicated RT cores for BVH traversal and Tensor cores (neither of which the Adreno has).

So in terms of area efficiency, Apple's GPU seems to be superior.

I thought Adreno had gotten RT cores?

 

[casperes1996]

So admittedly per die size, the X2 Elite Extreme's GPU may indeed be around the M5 (and that is certainly important), but everything else about the GPU is closer to what I would expect from the binned M5 Pro or at least in between the two - e.g. power draw. And the game comparison is still between something going through a single translation layer vs multiple translation layers. Then again, I don't know how many native games WoA has at the moment, so maybe they're restricted in what they can actually compare to (even if Mac has more it's not exactly an overabundance either) and if they can't do native vs native, well I guess they don't have much choice but to compare translated vs translated.

From a chip analysis perspective it's a case of "oh well, it's the only thing we can get so we'll take it" but from a consumer perspective, it's also a case of "who cares how well it can run native games, if I want to play games that aren't available natively". For that same reason, a comparison of a native x86 game on Windows versus that same game through CrossOver is also a valid comparison, from an end-user perspective. It is the performance they can expect to get on the two computing platforms running the same workload. The same as in, same result to the user, even if the workload involves more steps on one platform. As with all benchmarks it is important to understand the context and what it actually shows, but it is still valuable to say "if you want to play XYZ, this is how it performs. It is not like for like, it is not a pure hardware comparison, but it is what you can expect from the platforms with the available software stacks for running said game"

 

[MerryCherry]

I thought Adreno had gotten RT cores?

I suppose they do have "RT cores", but they can't do BVH traversal in hardware, which shows in benchmarks such as Solar Bay Extreme:

Also notice the comically huge 3x improvement from Dimensity 9400 to 9500. That's what happened when ARM added RT cores with BVH traversal in hardware to their latest Mali GPU.

Edit:

According to Kurnal, the Mali G1 Ultra GPU in D9500 is also 24 mm², equal in size to 8 Elite Gen 5.

 

[dada_dave]

From a chip analysis perspective it's a case of "oh well, it's the only thing we can get so we'll take it" but from a consumer perspective, it's also a case of "who cares how well it can run native games, if I want to play games that aren't available natively". For that same reason, a comparison of a native x86 game on Windows versus that same game through CrossOver is also a valid comparison, from an end-user perspective. It is the performance they can expect to get on the two computing platforms running the same workload. The same as in, same result to the user, even if the workload involves more steps on one platform. As with all benchmarks it is important to understand the context and what it actually shows, but it is still valuable to say "if you want to play XYZ, this is how it performs. It is not like for like, it is not a pure hardware comparison, but it is what you can expect from the platforms with the available software stacks for running said game"

Oh absolutely, and by the same token Apple can advertise in comparison to the Elite how well the M5-series can run say CP 2077 or Assassins Creed or Baldur's Gate or ... you know any of the other games that are actually native to Mac and do not have a native Windows on Arm version, even if they are obviously native to Windows on x86. Again, maybe not a long list, but I don't know how long the reverse list is either?

I suppose they do have "RT cores", but they can't do BVH traversal in hardware, which shows in benchmarks such as solar bay Extreme.
View attachment 38457
Also notice the comically huge 3x improvement from Dimensity 9400 to 9500. That's what happened when ARM added RT cores with BVH traversal in hardware to their latest Mali GPU.

Got it!

 

[casperes1996]

Oh absolutely, and by the same token Apple can advertise in comparison to the Elite how well the M5-series can run say CP 2077 or Assassins Creed or Baldur's Gate or ... you know any of the other games that are actually native to Mac and do not have a native Windows on Arm version, even if they are obviously native to Windows on x86. Again, maybe not a long list, but I don't know how long the reverse list is either?

I know of no WoA native games. Potentially Minecraft? Certainly a tiny list.

Though at reasonable resolution and quality settings you'll be more GPU than CPU bound anyway and I don't think CPU translation should impact that very much. HLSL will still compile through the GPU driver to native GPU code in the end I assume