Competition is better, we had good ride from cortex A72 to A78 of good cores, this killed all custom cores projects, when Arm messed up A710 whole market were affected, if we had custom cores maybe things would be better now. Am rooting for Oryon too and Dream chip, let three horse race begin in android space.
FlameTail
Diamond Member
- Dec 15, 2021
- 4,384
- 2,756
- 106
FlameTail
Diamond Member
- Dec 15, 2021
- 4,384
- 2,756
- 106
Coz Oryon has a legendary backstory. Startup Nuvia, $1.4 billion acquisition, multiple lawsuits (Apple and ARM) etc.....
I don't want them to fail.
DrMrLordX
Lifer
- Apr 27, 2000
- 22,525
- 12,390
- 136
Yeah I know. It would have made tinkering with ARM SoCs and/or DiY mini PCs a much-more attractive option. Sadly that market just hasn't matured over the years.
I’ve seen this joke or claim countless times now about the X4 or X5 and Nuvia if the performance is similar.
It’s wrong. Power is another concern independently of performance and while Arm still do far better than Intel or AMD in performance < 10W (on total platform power with DRAM which counts) they don’t quite match Apple even recently — they’re in the same cluster with Apple on a graph but not the same. See the 8 Gen 3 vs A14 from Geekerwan on ST.
Qualcomm’s Nuvia almost certainly is closer to Apple on perf/W if not ahead (similar to the first Nuvia graphs) judging by the humongous L2 and likely humongous L1, and this area penalty is luckily buttressed with lower licensing cost. It’s true R&D/labor is another variable, but Qualcomm for now have the scale to amortize that cost between phones/tablets, laptops, automotive, and XR/gaming. (They said they will charge more for 8 Gen 4 but I bet that will be just as true of the Dimensity 9400 on N3E lol).
Now I don’t think the X5 will be able to blow up power any further than the X4, and that would defeat some of the point here. They’ll also have a node gain with the Dimensity 9400. So it’s entirely possible they really do add +20-30% to their peak performance via a new architecture (possibly more SRAM I feel has to happen to be efficient still?) and N3E clock boost around the same wattage.
That range would still put them in line with the 8 Gen 4 ST performance rumors, and at that point would you rather be MediaTek paying the IP licensing costs and area or Qualcomm with the area, ISA licensing (which is small), engineering labor — and quite possibly still more efficient cores?
I would take Qualcomm. I don’t know it will be hold true in the long run, they do have to retain talent and keep scale to make it worthwhile but it’s tough to see this case as a “waste” in the near term.
Yeah I agree it would be cool but the reason we haven’t seen an SBC is that these are still flagship cores — we’re more likely to see a Mac Mini-style box with the Snapdragon X Elite than an SBC on N4/5 or N3 with great X cores, it just doesn’t make sense economically to do that.
I also want to point out: it’s not necessarily just Qualcomm that is annoyed or threatened by Arm keeping pace and being more competitive — it’s also Intel, AMD and to a lesser degree Apple.
Even today, I think like 4-8 X4 cores and 4 A720’s in a laptop chip on N4P would be good. We know they’d do idle power much better than Intel and AMD with Arm’s fabric, and X4’s are still very efficient from a platform perspective, likewise A720’s all the moreso especially on area.
Right now the 8 Gen 3’s X4 @ 3.3GHz is surprisingly in line with a lot of of 7840U results or about 10% down the best, and the 8 Gen 3’s X4 does so with ~ 5W +- a bit. That 7840U is using 18-20W+ at peak ST. And as for the curves, I doubt a 7840U clocked slightly down (say to 4.4GHz) to match the 8 Gen 3’s 1690 GB5 or 2329 GB6 is going to do so with 5W +- 1W on total power.
… and Zen 4C or Crestmont not going to be as efficient as cut down X4’s or A720’s for background tasks at the same performance lol.
Right now if Nvidia threw something like 8 X4’s (maybe where only two of them are full size for ST purposes) and 4 A720’s down on an Arm DSU with their own GPU IP, they could have a pretty competitive chip for thin and lights while getting as much peak throughput as the others, if not more under real power constraints. With the X5 and say N3E that product grows even more competitive.
Even today, I think like 4-8 X4 cores and 4 A720’s in a laptop chip on N4P would be good. We know they’d do idle power much better than Intel and AMD with Arm’s fabric, and X4’s are still very efficient from a platform perspective, likewise A720’s all the moreso especially on area.
Right now the 8 Gen 3’s X4 @ 3.3GHz is surprisingly in line with a lot of of 7840U results or about 10% down the best, and the 8 Gen 3’s X4 does so with ~ 5W +- a bit. That 7840U is using 18-20W+ at peak ST. And as for the curves, I doubt a 7840U clocked slightly down (say to 4.4GHz) to match the 8 Gen 3’s 1690 GB5 or 2329 GB6 is going to do so with 5W +- 1W on total power.
… and Zen 4C or Crestmont not going to be as efficient as cut down X4’s or A720’s for background tasks at the same performance lol.
Right now if Nvidia threw something like 8 X4’s (maybe where only two of them are full size for ST purposes) and 4 A720’s down on an Arm DSU with their own GPU IP, they could have a pretty competitive chip for thin and lights while getting as much peak throughput as the others, if not more under real power constraints. With the X5 and say N3E that product grows even more competitive.
I will bet against them backtracking on front end width specifically decode. We’ll see in a few months and you can check back with me.
The cost of those decoders and fetching was different at those times, MOP caches- which they usually expanded when they backtracked like with the X1-> X2 for example, don’t improve their efficiency or area anymore now they switched to Arm64-only decoders (dropping 32-bit compatibility). The 64-bit only ones are also just 1/4 the size which is pretty wild. In this case it’s because Arm’s 32-bit ISA was genuinely different unlike X86 so that’s why it added area and apparently some power cost.
Agree re roadmaps pushed back though. I wonder how much of this is going to come down to just needing a bigger L1 and L2 though.
They already do cheaper X4 cores with a gutted SIMD (128x2 not 128x4) and they’ll allow you to do that & 512kb of L2 for up to 3(?) others as long as the prime X4 has 1MB of L2 and the full SIMD units.
And then when you count slight area costs of hitting higher frequencies (cell or track choices or domino logic stuff, idk), you have three separate variables from phydes to SIMD to smaller L2 that allows you to save on marginal X cores after the first.
MediaTek has talked about this with respect to physical design — using different transistors or whatever — of their biggest X4 core that hits 3.25GHz while the other 3 hit 2.8GHz in the D9300, besides the L2/SIMD difference.
FlameTail
Diamond Member
- Dec 15, 2021
- 4,384
- 2,756
- 106
I wonder if ARM is handing Qualcomm the ARMv9 ISA, what with the lawsuit in between them and all.
The first gen Oryon core in the X Elite is definitely not ARMv9, but ARMv8 as confirmed by Geekbench listings.
No ARMv9 means no SVE2, SME or SME2.
The first gen Oryon core in the X Elite is definitely not ARMv9, but ARMv8 as confirmed by Geekbench listings.
No ARMv9 means no SVE2, SME or SME2.
moinmoin
Diamond Member
- Jun 1, 2017
- 5,203
- 8,365
- 136
"Annoyed" is not the word I'd use. It's basic competition every competitor ought to (actually must) consider for its business' well-being.
OG Kryo showed that they can fail and still come up with a win - albeit with a single generation misstep for SD810 as they navigated big.Little SoC design issues for the first time.
Nuvia was acquisition bait from day 1 - they were years from tape out when Qualcomm bought them, whatever their optimistic projections at the time.
QC was basically just buying a concept with a design team to bring it to fruition, so it's never been Nuvia failing or not, but rather the firm that acquired them, which considering their market modem dominance is basically impossible.
A similar thing is true of Samsung and Exynos (renamed XNS, I'm too lazy 😆) - though in their case while they can still continue on without Mongoose it's the (lacking) competitiveness of their own fab nodes that is tripping up XNS vs SD at the moment.
Even beating TSMC to Nanosheet/GAA/MBCFET hasn't seemingly improved their fortunes any here.
They know that TSMC is the better way to go, but the market perception of their high end fab nodes would very likely tank if they sent their own SoC's to be produced somewhere else.
IMHO they should just spin off the Exynos outfit under another name.
They're just shooting themselves in the foot over and over if this carries on, especially as AMD's RDNA IP is much more closely tied to TSMC's fab nodes currently, which sets Exynos at a double disadvantage.
Apple's current Axx and Mx SoC ARM cores also do not correspond to ARMv9-A ISA.
This is not so very detrimental as you think, at least not for now.
As long as you have 4x 128 bit NEON SIMD units you can still handle vector processing on CPU pretty well vs 2x 256 bit, or 1x 512 bit SVE2 SIMD units.
Despite there being 3 generations of v9-A hardware in the market the AV1 video decoder dav1d still hasn't had a single SVE2 contribution to it - the fact that Apple has yet to make the plunge on v9-A probably isn't helping matters with SVE2 adoption either.
SarahKerrigan
Senior member
- Oct 12, 2014
- 735
- 2,035
- 136
Considering the absolute flop that consumer SVE has been so far, I'm not sure Qualcomm cares all that much.
That being said, I expect ARM and Qualcomm will ultimately reach a settlement, probably along the lines of "Qualcomm pays ARM (either lump-sum or higher royalties) to go back to status quo on their arch license and release Oryon unencumbered."
Yes Poseidon/V3 is almost certainly based on the same core µArch as X5.
If ARM follow this V1=X1 -> V2=X3 cadence then Aphrodite/V4 will be based on X7.
N series seems to be every 3rd big Axx/xxx core, unless they change that with the slowing improvement rate now seen in this segment from A78 onward.
I wouldn't say flop, there are likely SVE2 instructions in Android code or anything else major, but right now anything that doesn't have money going into it regularly doesn't yet feel the need to adopt SVE2 over polishing their NEON code paths.
The situation isn't a million miles from early AVX512 adoption with Apple's continued v8-A use acting as a sort of analogue to the fragmentation issues that had (and still has 😂) plus the lack of AMD support.
I don't really see much benefit to Apple going ARMv9. They may not care about SVE2 if their homegrown AMX unit is their chosen way forward. That gives them flexibility to implement what they want without waiting for ARM to put it into their ISA and ignore what they don't want rather than being forced to implement it because it is part of the ISA.
Apple was able to front run ARM with ARMv8 so at this point not going ARMv9 has to be a deliberate decision. The one hope I could see for those who want to see Apple go ARMv9 is the possibility their architectural license was only for ARMv8. If so the extension into the 2040s they agreed to with ARM last year would clear the way for ARMv9 stuff from Apple. While I'm throwing that out as a possibility I'm pretty skeptical. If that was necessary for ARMv9 adoption and Apple wanted to adopt ARMv9 they would have negotiated that extension several years earlier.
Reading into it too much. No company likes competition but it’s a boon for us. Of course a great X5 on par with Apple/QC on perf & in the same league on perf/W is a PITA — frustrating, annoying, even, for everyone but MediaTek, Samsung and Nvidia.
I assume that extension included an Arm V9 provision and was only reported in generic Arm ISA terms, and they might’ve had Arm V9 too already (just not till 2040) and so the report left some details left out.
I think you’re right though it’s not a priority though and I don’t see why they’d go with SVE2 and SME over their current AMX stuff.
I mean Qualcomm’s 8 Gen 2 don’t have SVE2 even included/enabled fwiw, and the X Elite is Arm V8. So it’s pretty dire lol
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 24K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 19K
-
-
-
Facebook
Twitter