News [SA] SA teasers on WOA, Qualcomm, and new Arm entrants (NV/Samsung?)

[ikjadoon]

A note on Geekbench 6:

Geekbench 6.0 and Geekbench 6.1 scores are not comparable, so it'd be helpful if future leaks specify exactly which version, heh.

Thanks to these changes, Geekbench 6.1 single-core scores are up to 5% higher, and multi-core scores are up to 10% higher than Geekbench 6.0 scores. As a result of these methodological differences, which have a non-trivial impact on scores, we recommend users not compare Geekbench 6.1 scores against Geekbench 6.0 scores.

GB6.1 was released in June 2023, so it's possible recent leaks are a little off due to accidental GB6 vs GB6.1 comparisons.

 

[SpudLobby]

Assuming those Phoenix scores are an accurate leak:

CPU	"IPC" (Pts / GHz)	GB5 1T Pts	Frequency
Cortex-X3 (SD8G2)	438	1,398	3.19 GHz
Cortex-X2 (SD8G1)	409	1,309	3.20 GHz
Cortex-X1 (8CXG3)	362	1,087	3.0 GHz
Phoenix A	467	1,775	3.8 GHz
Phoenix B	470	1,600	3.4 GHz

So Phoenix would have the +7% IPC of the Cortex-X3, so there's a chance it matches the Cortex-X4 IPC. Of course, there's so much more to making a solid consumer SoC: efficiency, peak power draw, smaller Phoenix cores, [the entire rest of the Hamoa SoC].

But, at least Microsoft's decision makes more sense: if Microsoft's own laptops weren't basically the fastest Windows on Arm devices, that would've been shameful.

I imagine the "2024 NVIDIA SoC" will use the Cortex-X4. Arm estimates in Geekbench 5 IPC uplift is ~7%, so that tracks with the Cortex-X4 IPC = Phoenix IPC.

Whether Qualcomm thinks its decisions were worth $1.6 billion, a supplier lawsuit, and the loss of Microsoft Surface design wins: maybe they expect better results from Phoenix's successor (assuming the perf leaks are true).

Sources:

Lenovo ThinkPad X13s review: not quite a MacBook Air | Digital Trends

The Lenovo ThinkPad X13s has long battery life and optional always-connected internet in a thin and light chassis. Its performance is a step behind, however.

www.digitaltrends.com

Snapdragon 8 Gen 2 vs Snapdragon 8 Plus Gen 1: The new Adreno GPU is the best upgrade here

The Snapdragon 8 Gen 2 is the latest flagship SoC from Qualcomm, but how much better is it than the phenomenal 8 Plus Gen 1?

www.xda-developers.com

Well the original comparison was to their current Arm laptops, AKA the 8Cx Gen 3 and delivering in that vein.

I for one do agree losing to the X4 on IPC is a bit "come on" and indeed the X4 would have higher IPC in this case, but then again what matters as much as that within these variations is energy efficiency^1 if they're vaguely similar on performance, and I suspect Nuvia's second gen would have more in the tank on a revised IPC gain than the X5.


1: On this note, let's remember going wide and or having high perf/GHz in and of itself won't necessarily guarantee energy efficiency even on a good process node, in part due to other design decisions in the floorplan/operating frequency and more importantly the cache. They're related within a certain performance profile/class, but:

One can significantly improve energy efficiency via additional SRAM throughout the stack without affecting performance positively too much. I am pretty confident the 8Cx Gen 4's L1 will mirror Apple's moreso than Arm's and it's shared L2 clusters (12MB for every 4 cores) + decent L3/SLC rumors seem to be in line with this assumption.


Also FWIW the Nvidia SoC will be 2025 I think. I bet they'll use X5s, but either way it will be competitive for sure on performance.

 

[Doug S]

A note on Geekbench 6:

Geekbench 6.0 and Geekbench 6.1 scores are not comparable, so it'd be helpful if future leaks specify exactly which version, heh.



GB6.1 was released in June 2023, so it's possible recent leaks are a little off due to accidental GB6 vs GB6.1 comparisons.

While that's true the fact that leaks are showing GB5 numbers instead of GB6 is just stupid and unnecessarily confusing. Why is anyone still using GB5 for anything?

GB6.1 changing results by that much means it should have been called GB7 to avoid confusion, IMHO.

 

[SpudLobby]

While that's true the fact that leaks are showing GB5 numbers instead of GB6 is just stupid and unnecessarily confusing. Why is anyone still using GB5 for anything?

GB6.1 changing results by that much means it should have been called GB7 to avoid confusion, IMHO.

Probably because it's just the standard everyone knows off the top of their head index-wise, e.g. a 1500 or 2000 in GB5 registers in a way 2200 in GB6 doesn't yet. At any rate yeah, you still have to check basically.

 

[ikjadoon]

Well the original comparison was to their current Arm laptops, AKA the 8Cx Gen 3 and delivering in that vein.

I for one do agree losing to the X4 on IPC is a bit "come on" and indeed the X4 would have higher IPC in this case, but then again what matters as much as that within these variations is energy efficiency^1 if they're vaguely similar on performance, and I suspect Nuvia's second gen would have more in the tank on a revised IPC gain than the X5.


1: On this note, let's remember going wide and or having high perf/GHz in and of itself won't necessarily guarantee energy efficiency even on a good process node, in part due to other design decisions in the floorplan/operating frequency and more importantly the cache. They're related within a certain performance profile/class, but:

One can significantly improve energy efficiency via additional SRAM throughout the stack without affecting performance positively too much. I am pretty confident the 8Cx Gen 4's L1 will mirror Apple's moreso than Arm's and it's shared L2 clusters (12MB for every 4 cores) + decent L3/SLC rumors seem to be in line with this assumption.


Also FWIW the Nvidia SoC will be 2025 I think. I bet they'll use X5s, but either way it will be competitive for sure on performance.

Yeah, that's fair. I was thinking what NVIDIA might put out, but right, the 8CX Gen3 is QC's internal comparison. 2025: did not know that. Wow, so it's is a ways out: I assumed 8CX Gen3 this year in a Surface Arm refresh, but NVIDIA in 2024. That may not track as I assumed; Windows Central doesn't mention any Surface Pro X update rumors.

That's a good point re: X5, though rumors claimed Cortex-X5 was a ground-up rewrite uArch (unlike X1 → X4), but then that rumor's codename ("Logan") didn't pan out: instead we got "Blackhawk".

//

A good caveat: I don't know where some of these SoCs are wasting their power, putting out the perf that they do. As long it's closer to Apple's consumption vs AMD/Intel, at least it'll be an interesting launch.

Agreed: even Arm doubled L2 to 2MB in X4 (with no increase in latency, they claim, by moving the SRAM closer to the core), so Qualcomm using higher caches is a positive sign for efficient uplift.

While that's true the fact that leaks are showing GB5 numbers instead of GB6 is just stupid and unnecessarily confusing. Why is anyone still using GB5 for anything?

GB6.1 changing results by that much means it should have been called GB7 to avoid confusion, IMHO.

I think Spud's has the right vibe: feels like vernacular is still with GB5 anchor points (e.g., "1700 pts is a modern CPU, 1000 is Skylake", etc.).

GB5 also had three version changes that were not comparable, fwiw. I guess they want to keep it updated enough between the years 😂

These changes mean Geekbench 5.1 scores will be higher than Geekbench 5.0 scores. As a result, we recommend users not compare Geekbench 5.0 and Geekbench 5.1 results.

GB5.0 = one standard
GB5.1, 5.2 = another standard
GB5.3, 5.4, 5.5 = another standard (current)

I'm still warming up GB6 numbers, so I guess a good thing I'm not wedded yet to GB6.0 numbers yet, ha.

 

[SpudLobby]

Yeah, that's fair. I was thinking what NVIDIA might put out, but right, the 8CX Gen3 is QC's internal comparison. 2025: did not know that. Wow, so it's is a ways out: I assumed 8CX Gen3 this year in a Surface Arm refresh, but NVIDIA in 2024. That may not track as I assumed; Windows Central doesn't mention any Surface Pro X update rumors.

That's a good point re: X5, though rumors claimed Cortex-X5 was a ground-up rewrite uArch (unlike X1 → X4), but then that rumor's codename ("Logan") didn't pan out: instead we got "Blackhawk".

//

A good caveat: I don't know where some of these SoCs are wasting their power, putting out the perf that they do. As long it's closer to Apple's consumption vs AMD/Intel, at least it'll be an interesting launch.

Agreed: even Arm doubled L2 to 2MB in X4 (with no increase in latency, they claim, by moving the SRAM closer to the core), so Qualcomm using higher caches is a positive sign for efficient uplift.



I think Spud's has the right vibe: feels like vernacular is still with GB5 anchor points (e.g., "1700 pts is a modern CPU, 1000 is Skylake", etc.).

GB5 also had three version changes that were not comparable, fwiw. I guess they want to keep it updated enough between the years 😂



GB5.0 = one standard
GB5.1, 5.2 = another standard
GB5.3, 5.4, 5.5 = another standard (current)

I'm still warming up GB6 numbers, so I guess a good thing I'm not wedded yet to GB6.0 numbers yet, ha.

Yeah, exactly RE: 1000 is Skylake, or the range of Gracemont or an A7x core in the 3GHz range, or lower clocked Zen 2.

I don’t know if this is 6 or 6.1 but

“Geekbench 6 scores are calibrated against a baseline score of 2,500 (which is the score of a Dell Precision 3460 with a Core i7-12700 processor). Higher scores are better, with double the score indicating double the performance.”

This is annoying honestly. Prefer Skylake @ 1000.

 

[Doug S]

Yeah, exactly RE: 1000 is Skylake, or the range of Gracemont or an A7x core in the 3GHz range, or lower clocked Zen 2.

I don’t know if this is 6 or 6.1 but

“Geekbench 6 scores are calibrated against a baseline score of 2,500 (which is the score of a Dell Precision 3460 with a Core i7-12700 processor). Higher scores are better, with double the score indicating double the performance.”

This is annoying honestly. Prefer Skylake @ 1000.

Not sure why they couldn't have scaled the numbers to match. If Skylake gets 1000 in GB5 then when developing GB6 they could see whatever it scores and multiply it by whatever is required to make it score 1000 there too. It would scale differently since it has different tests and maybe SIMD stuff helps more or memory bandwidth helps less or whatever, but at least something that scored 1500 in GB5 would probably be somewhere in that ballpark for GB6.

Obviously the i7-12700 wasn't 2.5x faster in ST than Skylake so I can't see any good reason they rescaled like that. Is GB7 going to pick a then current Intel CPU and call it 6000? Heck if you're going to do that make it 60,000 then at least there will be no confusion about which version of GB that score came from!

 

[SpudLobby]

Not sure why they couldn't have scaled the numbers to match. If Skylake gets 1000 in GB5 then when developing GB6 they could see whatever it scores and multiply it by whatever is required to make it score 1000 there too. It would scale differently since it has different tests and maybe SIMD stuff helps more or memory bandwidth helps less or whatever, but at least something that scored 1500 in GB5 would probably be somewhere in that ballpark for GB6.

Obviously the i7-12700 wasn't 2.5x faster in ST than Skylake so I can't see any good reason they rescaled like that. Is GB7 going to pick a then current Intel CPU and call it 6000? Heck if you're going to do that make it 60,000 then at least there will be no confusion about which version of GB that score came from!

Indeed. The 2500 choice was really silly and brings us back to the GB4 days of unnecessarily inflated absolute ST indexes (and it's not like the marginal precision is worth anything here with this index, it's not an order of magnitude, just a factor of 2.5). Agree RE: the tests would change - I forgot whatever crap they added, SIMD may well be something, but yeah it would still make far more sense.

Sadly I suspect it's psychological and their mass market that bothers to buy the mobile app or $99 "Pro" edition are e.g. YouTubers or run-of-the-mill tinkerers that find this exciting in part due to ignorance.

 

[soresu]

That's a good point re: X5, though rumors claimed Cortex-X5 was a ground-up rewrite uArch (unlike X1 → X4), but then that rumor's codename ("Logan") didn't pan out: instead we got "Blackhawk".

I swear that ARM are making it hard to find clean PDFs of those presentation slides on purpose 😱😒

Seriously their PR department is truly terrible for such an influential tech company.

 

[Doug S]

Not sure why they couldn't have scaled the numbers to match. If Skylake gets 1000 in GB5 then when developing GB6 they could see whatever it scores and multiply it by whatever is required to make it score 1000 there too. It would scale differently since it has different tests and maybe SIMD stuff helps more or memory bandwidth helps less or whatever, but at least something that scored 1500 in GB5 would probably be somewhere in that ballpark for GB6.

Obviously the i7-12700 wasn't 2.5x faster in ST than Skylake so I can't see any good reason they rescaled like that. Is GB7 going to pick a then current Intel CPU and call it 6000? Heck if you're going to do that make it 60,000 then at least there will be no confusion about which version of GB that score came from!

Come to think of it, that's basically what SPEC did at one point - after SPEC92 and SPEC95 they rescaled to very low numbers for SPEC2000 to avoid confusion with previous results.

 

[A///]

I swear that ARM are making it hard to find clean PDFs of those presentation slides on purpose 😱😒

Seriously their PR department is truly terrible for such an influential tech company.

ever since the economic downturn a decade and more ago the ir pages of companies have been a jumbled mess. almost as if it's done on purpose.

 

[SpudLobby]

I swear that ARM are making it hard to find clean PDFs of those presentation slides on purpose 😱😒

Seriously their PR department is truly terrible for such an influential tech company.

I assume it's on the presser page but yeah the site has a lot of material in part I think because it's for both developers + hardware design clients (developers of a diff kind) + general info, idk.

 

[ikjadoon]

NVIDIA does have a fair bit of Arm + GPU → platform → end-user device experience with their Jetsons, one good review here.

How did I forget the Nintendo Switch's SoC, the NVIDIA Tegra X1 / X1+?

120+ million units, 4x Cortex-A57, 256-core Maxwell GPU.

NVIDIA has an abundance of experience in shipping consumer Arm CPUs + its GPUs.

 

[NTMBK]

How did I forget the Nintendo Switch's SoC, the NVIDIA Tegra X1 / X1+?

120+ million units, 4x Cortex-A57, 256-core Maxwell GPU.

NVIDIA has an abundance of experience in shipping consumer Arm CPUs + its GPUs.

Nintendo got a great deal on the X1 because it was a flop elsewhere. If I remember correctly the A53 cores were bugged, or something, so only the big cores worked? Nintendo didn't need the little cores and wanted to underclock the thing for a handheld anyway so it worked out great for them.

 

[Heartbreaker]

Nintendo got a great deal on the X1 because it was a flop elsewhere. If I remember correctly the A53 cores were bugged, or something, so only the big cores worked? Nintendo didn't need the little cores and wanted to underclock the thing for a handheld anyway so it worked out great for them.

I've never read about anything wrong with the A53 cores in X1. NVidia uses a different core migration strategy than ARM, but I doubt that's the Issue.

It's NVidia, so I expect the issue was price. SoC are largely a low margin commodity business, and NVidia likely wanted a fat premium for having their proprietary GPU. Something most general users probably didn't care about at all.

Though Nintendo likely did care about the GPU... So were the ideal customer.

Though if NVidia pushes their luck NVidia could get a decent GPU from somewhere else. Samsung as a deal with AMD. Qualcomm has old Radeon tech they are doing well with.

 

[soresu]

Qualcomm has old Radeon tech they are doing well with

GC's Adreno GPU series started with AMD's Imageon µArch design and patents, but at this point I would not expect it to resemble what AMD sold to them in anything but the abstract terms that patent lawyers so love to make big cases about.

Same thing with ARM Mali which was originally acquired from a Norwegian company I think back in its very early days before undergoing umpteen iterations and at least 2 radical deisgn overhauls.

Edit: The Norwegian company was Falanx Microsystems, since renamed to ARM Norway, and as far as I can tell still based in Trondheim, Norway.

 

[Heartbreaker]

GC's Adreno GPU series started with AMD's Imageon µArch design and patents, but at this point I would not expect it to resemble what AMD sold to them in anything but the abstract terms that patent lawyers so love to make big cases about.

Same thing with ARM Mali which was originally acquired from a Norwegian company I think back in its very early days before undergoing umpteen iterations and at least 2 radical deisgn overhauls.

Edit: The Norwegian company was Falanx Microsystems, since renamed to ARM Norway, and as far as I can tell still based in Trondheim, Norway.

I didn't mean it was still the same as when they bought it. Just that was the origin, and it was doing well today.

I don't think Nintendo is locked into NVidia GPUs.

 

[soresu]

I don't think Nintendo is locked into NVidia GPUs.

I think potentially not, and certainly with the amount of movement we have seen from Nintendo specifically on the Vulkan API extension development front we can infer that they are leaning towards breaking from dependence on the nVidia provided NVN API that many Switch games employ currently.

If nVidia becomes cute about Nintendo migrating away from NVN with a shim to Vulkan (a la DXVK/Zink) it won't be pretty at all.....

 

[Saylick]

Exclusive: Nvidia to make Arm-based PC chips in major new challenge to Intel

Nvidia dominates the market for AI computing chips. Now it is coming after Intel’s longtime stronghold.

www.reuters.com

Looks like it might be Nvidia after all?

 

[Heartbreaker]

Exclusive: Nvidia to make Arm-based PC chips in major new challenge to Intel

Nvidia dominates the market for AI computing chips. Now it is coming after Intel’s longtime stronghold.

www.reuters.com

Looks like it might be Nvidia after all?

All the players should be working on Windows SoCs, including Intel and AMD:

NVIDIA and AMD reportedly working on ARM-based processors for PCs - VideoCardz.com

NVIDIA’s push into the consumer CPU market NVIDIA, as reported by Reuters, is discreetly in the process of designing new chips tailored for personal computers, with the aim of challenging the well-established CPU market. Sources familiar with the matter have informed Reuters that NVIDIA is...

videocardz.com

So far nothing about Intel's efforts.

 

[poke01]

I think potentially not, and certainly with the amount of movement we have seen from Nintendo specifically on the Vulkan API extension development front we can infer that they are leaning towards breaking from dependence on the nVidia provided NVN API that many Switch games employ currently.

If nVidia becomes cute about Nintendo migrating away from NVN with a shim to Vulkan (a la DXVK/Zink) it won't be pretty at all.....

That's far away. "Switch 2" uses NVN2 but I do think Vulkan will still be used regardless of NVIDIA GPU in the future.

 

[moinmoin]

Be it on ARM or not, I honestly wish Windows would just go away.

 

[soresu]

Be it on ARM or not, I honestly wish Windows would just go away.

Fo sho, if Wine/Proton ever manages to match it's compatibility with general Windows apps the same way it has for games then I could see myself waving a dry eyed adieu to my time with it.

 

[Heartbreaker]

Be it on ARM or not, I honestly wish Windows would just go away.

I just wish Microsoft, would stop making it worse.

 

[moinmoin]

I just wish Microsoft, would stop making it worse.

Microsoft for quite some time now has no interest in it anymore aside turning it into a platform offensively selling the user their service of the day.

Come to think of it same is actually happening now with all Apple devices as well... And Android never was anything else to start with. Why can't we just get a clean slim basic OS anymore?