Question Qualcomm's first Nuvia based SoC - Hamoa

[poke01]

Qualcomm's working on a 2024 PC chip codename "Hamoa" with up to 12 (8P+4E).

Said to the same cache layout as M1. large private L1$, per-cluster L2$ (cluster = 4 cores, 12MB for every cluster) and a lot of LLC.

source:

https://twitter.com/x/status/1589405172979339264

 

[DrMrLordX]

100% agree. GB6 ST is much more valuable and more accurate.

Am I the only one that hates GB5 and GB6?

Besides, we need to see more third-party benchmarks on this thing. It looks promising but still. You're talking about a new product in the WARM space which is still alien to many of the posters here (and to many PC enthusiasts in general).

 

[adroc_thurston]

Is it comfortably beating M3 and Snapdragon X Elite in GB6?

Yea.
M3 is literally a trainwreck meme.

Can it also do that at lower mobile TDPs?

Yea.
30W is the new 15W anyway.

 

[Hitman928]

Looked it up, according to Dylan Patel's estimates, M2 core size is ~2.76 mm2, Zen4 core area is ~2.56 mm2, and Zen4c is ~1.43 mm2. If we use his a15 estimate instead (2.58 mm2) then Zen4 and a15 are equal in core size on the same node. So one went larger area for a wider core and more IPC and the other went high frequency.

 

[Thibsie]

x86 optimizes for 2 things:
- servers
- die space

because the cover a market from top end server at 300W to mobile devices at 10w which use the same core.

Can't sell a chip in $300 craptop if it uses a ton of die space. Hence you need to clock it high and your efficiency starts to suffer. M2 and this nuvia chips are made for as single specific purpose, mobile client device so no compromise needed in core design and efficiency is a core aspect. So the go wide.

Both m2 and nuvia get there performance at the cost of die size. We know it for the m2 for sure, probably same for nuvia core. You can only do that if your target expensive devices which makes me think this SOC will not be a huge success due to pricing.

Mmm?
The nuvia core is a repurposed server core.

 

[Doug S]

Mmm?
The nuvia core is a repurposed server core.

Nuvia was designing a server core that was supposed to be released in like 2021. Its been how long since Qualcomm bought Nuvia? In all that time you don't think they had ample opportunity to change focus to something more mobile (at least laptop if not phone) focused? It seems to be pretty power efficient, if it had been targeting servers it would be burning more power unless they had been planning to cram 256 cores on a reticle sized die.

 

[coercitiv]

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

Nvidia has quietly begun designing central processing units (CPUs) that would run Microsoft’s (MSFT.O) Windows operating system and use technology from Arm Holdings(O9Ty.F), , two people familiar with the matter told Reuters.

Advanced Micro Devices (AMD.O) also plans to make chips for PCs with Arm technology, according to two people familiar with the matter.

It's just Reuters, but it does seem like they did their homework and verified the info using two independent source for each company.

 

[JoeRambo]

GB6 = Okay tool for estimating typical non-pro user performance scenarios.

There we go, i have news for You. 99% of Laptop and Desktop users are not PRO users and GB6 is in fact great benchmark for them.

GB5 has improved a lot since the days Linus Torvalds called Geekbench GB3 utter *** here:

Real World Technologies - Forums - Thread: Charlie re: Apple and ARM

www.realworldtech.com

GB4 was better:

Real World Technologies - Forums - Thread: Geekbench 4

www.realworldtech.com

And i am aware Crypto ended up only 5% in it due to past criticism. Yet GB6 is the next step in that evolution, removing those 5% completely and moving them into integral part of larger tasks.
You know what is better than such artificial multi gigabyte encode bs? Or just small component rendering HTML5?
Complete browser process as done in GB6. Same with some of the other subtasks that encompass multiple parts, without dedicating 5% of score to hashing or crypto.

How do they differ? One scales infinitely with cores, the other will plain out reveal trash multi core CPUs with bad uncores and so on. So that's GB6 value too, GB5's scaling also had value for 1% crowd.
GB5 was not that bad either, but it had completely and utterly unrealistic multi core scaling. It still has it's place like linpack, cb23 etc.

A 32 core TR would scale up to over twice what my 7950X performs at for my development workload, for example

So sir continue to run CB23 and enjoy Your perfect scaling estimation, I am developing on 13900KS with turned off E cores and HT and fixed clocks and for personal workstation it is awesome. It would be retarded to suffer some 70-80% of web, IDE, responsiveness/perf with some Threadripper, when i can SSH to 7950x and compile and build there?

In Nuvia laptop/desktop chip thread i was responding to unjust attack on GB6, and I evaluate GB6 score provided by QC with peace of mind, knowing they are not buffing up score with vector extensions, cache sizes that completely break some substest by fitting working set completely. And even GB6 MT scaling is revealing things about chip having not that great uncore, minimal L3 cache, 4C sized clusters. So a win for LAPTOP for mere mortals, PROs already ripping threads anyway.

 

[igor_kavinski]

NO EXECUTE! June 18, 2018

This guy is ex-Microsoft and ex-Intel. That's his impressions on one of the first ARM64 devices. He even worked on ARM64 emulation for Microsoft. Check his Aug 11, 2023 post: http://emulators.com/secrets.htm

Talks about M1 here: http://emulators.com/docs/nx43_ipc_race_is_afoot.htm

 

[JoeRambo]

This guy is ex-Microsoft and ex-Intel. That's his impressions on one of the first ARM64 devices. He even worked on ARM64 emulation for Microsoft. Check his Aug 11, 2023 post: http://emulators.com/secrets.htm

Truth to be told, they have done some ridiculously hard things like allowing to mix Arm64 and x64 libraries in single process*. Some seriuos efforts now that Intel's patents on SSE2 expired and x64 ABI emulation on ARM can no longer be touched by dirty lawyers.

*ARM64EC process, important detail for overzealous lawn owners

 

[Heartbreaker]

Chasing high clocks costs a huge amount of die space. Look at Zen4 vs. Zen4c.

No. The cores are identical. The die space savings come from cutting cache in half on Zen4c:

AMD Zen 4c Not an E-core, 35% Smaller than Zen 4, but with Identical IPC

AMD on Tuesday (June 13) launched the EPYC 9004 "Bergamo" 128-core/256-thread high density compute server processor, and with it, debuted the new "Zen 4c" CPU microarchitecture. A lot had been made out about Zen 4c in the run up to yesterday's launch, such as rumors that it is a Zen 4 "lite"...

www.techpowerup.com

The only thing that's changed is that the effective L3 cache per core has been reduced to 2 MB, from 4 MB on the 8-core "Zen 4" CCD. While the regular 8-core "Zen 4" CCD has eight "Zen 4" cores sharing a 32 MB L3 cache, the new 16-core "Zen 4c" CCD AMD introduced with "Bergamo" sees the chiplet pack two 8-core CCX (CPU core complexes), each with 16 MB of L3 cache shared among the 8 cores of the CCX. In this respect, the last-level cache and CPU core organization of the "Zen 4c" CCD has some similarities to the "Zen 2" CCD (which used two 4-core CCXs).

 

[moinmoin]

The cores are identical.

Yes.

The die space savings come from cutting cache in half on Zen4c

No, Zen 4c is not 35% smaller than Zen 4 only by cutting the L3 cache by half. That would otherwise mean that Zen 4 could be 70% smaller by not having an L3 cache at all. Zen 4 uses much more xtors to allow higher frequency.

 

[Heartbreaker]

Yes.

No, Zen 4c is not 35% smaller than Zen 4 only by cutting the L3 cache by half. That would otherwise mean that Zen 4 could be 70% smaller by not having an L3 cache at all. Zen 4 uses much more xtors to allow higher frequency.

Link?

 

[adroc_thurston]

The cores are identical.

No.

The die space savings come from cutting cache in half on Zen4c:

No.

 

[Hitman928]

Link?

You can find a breakdown here:

Zen 4c: AMD’s Response to Hyperscale ARM & Intel Atom

Bergamo Volumes, ASP, Performance, Hyperscale Order Shift, Die Shot, Floorplan, Physical Design, and Future Use of Dense Core Variants

www.semianalysis.com

Comparison table for core area breakdown. Does not include L3 area.

 

[moinmoin]

Beaten by @Hitman928

Die shot of the cores from that semianalysis article:

No.

Physically no, functionally yes.

 

[HurleyBird]

Yea.
M3 is literally a trainwreck meme.

So, here's an issue:

This sounds exactly like your "I have insider info" voice, and it just makes your entire persona seem unlikely.

I'm willing to entertain the idea, at least until disproven, that you might have confidential information regarding AMD server tech, AMD in general, or server tech in general; that you may be an insider at AMD, supermicro, etc.

I refuse to believe that you have some position in the industry that places you at the non-existent intersection of AMD server parts and Apple consumer SOCs. To the extent that any such position exists, it's at an abstract level that has been compartmentalized from the information you claim to possess. And I refuse to entertain the idea that you have sources feeding you information across various domains for the purpose of communicating that information to a random tech forum. Both ideas are insane.

Obviously it's possible to speculate that M3 will come out poorly based on delays and the performance of A17. But if you are speculating, you ought to do it in a different voice; one that isn't speaking from the position of the omnipotent insider. Otherwise, you deserve the reaction you get.

 

[poke01]

M3 is not a "trainwreck". More specifically M3 Max (CPU) will be the undisputed king for laptops based on IPC and perf/w.
IF it launches next month.

Please don't judge M3's performace based on A17 Pro. The iPhone has crap cooling( more like NO cooling) and design is only meant for iPhones and mini tablets.

 

[gdansk]

One may have extrapolated M3 into a trainwreck based on A17.
Whether or not that is correct remains to be seen but would not surprise me.

And on topic: SD X Elite (lol) GPU looks fast enough - but I wonder about their D3D/Vulkan/OGL implementations.

 

[poke01]

From Techinsights.
A17 P core: 2.18mm^2

EDIT: I wonder what size SD Elite core is?

 

[Denly]

For a desktop pc enthusiast I doubt it’ll ever be relevant in our market.

As much as I’d like to buy an Nvidia ARM desktop CPU I don’t see it ever happening. I also doubt Qualcomm will ever become a player in our market.

dv
Currently using a 8cx gen laptop from Samsung, long battery life, 5G and fanless is a breath of fresh air. I don't care about apps and games as I can remote in if I need something more intense done. With Windows moving into cloud/virtual MS need something like Macbook Air with reasonable performance.

 

[adroc_thurston]

Please don't judge M3's performace based on A17 Pro. The iPhone has crap cooling( more like NO cooling) and design is only meant for iPhones and mini tablets.

Ughhh cooling isn't the issue.
Coll big-core is just bad.

 

[poke01]

Edit: Deleted

 

[jeanlain]

One may have extrapolated M3 into a trainwreck based on A17.

Why? A17 is close to the best performers in single thread tasks while being a mobile chip.

 

[jeanlain]

Coll big-core is just bad.

Whatever that sentence means, I don't think it will age well.

 

[FlameTail]

M3 is not a "trainwreck". More specifically M3 Max (CPU) will be the undisputed king for laptops based on IPC and perf/w.
IF it launches next month.

Please don't judge M3's performace based on A17 Pro. The iPhone has crap cooling( more like NO cooling) and design is only meant for iPhones and mini tablets.

Also apparently a software update fixed the overheating issues with A17 and pushed the Single core CPU performance beyond 3000 points in GB6