#1. Both Neon and SVE have GEMM instructions that deal with matrices, the CPU is not devoid of support for them without SME.KleidiAI relies on two technologies: SVE2 and SME.
One could argue that, for the purpose of AI, SME is more important than SVE, since it deals with matrices.
Possible, but not a given considering Hexagon DSP/NPU can do a lot besides what the CPU and GPU µArchs can manage with AI/ML code.But in the long term, I think it's reasonable yo assume that Qualcomm will add SME sometime in the future.
It should have Streaming SVE as part of the SME spec.M4 has SME but no SVE/SVE2.
Something ehthusiasts immediately called out.Also another silly reason why you'd want to add SME is because it boosts your Geekbench 6 score by a few hundred points
Sounds good. Much better than Passmark:We looked at over 82,000 Geekbench 6 for Windows results uploaded to the Geekbench Browser over the last 30 days. Intel CPUs powered 59.2% of results, AMD CPUs powered 32.4%, and Qualcomm Snapdragon X SoCs powered 6.4%. Snapdragon X is off to a good start.
In the past 30 days, we've seen 22,000 x86 Windows benchmark submissions vs. just 56 ARM Qualcomm Elite (that’s 0.3% of all Windows benchmark submissions). Given Qualcomm CEO's claim that ARM could capture 50% of the Windows PC market in 5 years, there's a long road ahead.
Bearing in mind that is new entries, and even then such tests represent a pretty limited subset of users overall (people more like the audience of this forum than not).
Sounds good. Much better than Passmark:
Cache quantity between unrelated CPU µArchs is much like TFLOPS between unrelated gaming GPU µArchs.Fun fact: X Elite's CPU has the same amount of L2 cache as the M3 Max CPU
#2.What do you guys think?
Means to an end.Fun fact: X Elite's CPU has the same amount of L2 cache as the M3 Max CPU
Far better.Well heck, +20% isn't bad. That should at least be Everest-equivalent iso-clock, no?
I called it: the GPU would be the most exciting aspect of the 8G4."Looking at the internal testing data of 8G4, the CPU efficiency improvement is in the single digit percentage, but the GPU improvement is huge, D9300 peak at half the power is possible. If you care about CPU performance, please skip this generation and wait for 8G5."
Is it enough, though?Far better.
20% would put it ahead of M4-P (if we discount SME).
Late 2026 seems too far out when considering this, more like early 2026:Is it enough, though?
Oryon V2 is not coming on the next generation of X processors, which should use roughly the same Phoenix variants that will be featured in 8G4. We are talking late 2026 at the earliest for Oryon V2, which means we could be getting well into M6 territory.
Being two generations behind Apple is better than the four or five they were with 8cx Gen 3, but I was expecting more. If anything, their delays in delivering Phoenix (and Hamoa as a whole) cost them a lot. Nuvia's team will be put to the test in the next two years. I hope QC doesn't cheap out and keeps them well-staffed.
The problem here is terminology. The poll said "Oryon V2 (Pegasus)", meaning the next uarch iteration, thus the parenthesis.Late 2026 seems too far out when considering this, more like early 2026:
V2 is the terminology for the CPU IIRC. The next generation of the SoC will be called X2.In the Dell leak the timeframe closely tracks the next generation of X SoCs, but this is V2 of the SoC itself, which should use roughly the same Phoenix variants to be debuted with 8G4
Year | SoC | CPU |
2024Q2 | X Elite | Oryon V1 |
2024Q4 | 8G4 | Oryon V1+ |
2025Q1 | X Elite Refresh | Oryon V1/Oryon V1+? |
2025Q4 | 8G5 | Oryon V2 |
2025Q4 | X2 Elite | Oryon V2 |
This is where the confusion lies. AFAIK there is no agreement on what Oryon V1/1+/2/etc means.Pegasus is the codename for Oryon V2, I believe. Oryon V1 is Phoenix. Oryon V1+ is Phoenix-L.
You are confusing µArch with ISA perhaps.The community started calling Oryon V2, but this should mean the next CPU, not uarch, right? After all, Oryon is the CPU.
Oryon is just the brand name for Qualcomm's new custom CPU core.Your Oryon V1+ is their Oryon V2, which might be someone else's Oryon V1 Gen 2.
Oryon is the name of the overall CPU architecture like Zen. Phoenix, Pegasus and so on are the codenames for specific generations, like Zen4(Persephone) and Zen5(Nirvana).Oryon is the name of the CPU block. Is it also the name of the individual CPU cores? There's some ambiguity.
Qualcomm did good by naming the GPU in X Elite as 'Adreno X1'. I wish they did that for the CPU as well.
So far, we have heard these names such as "Oryon V1", "Oryon V2", "Pegasus", "Phoenix-L" only from the rumour mill.
What I meant by CPU is the block itself, i.e., is Oryon "V1 to V2" like Cezanne to Rembrandt or like Phoenix to Strix?You are confusing µArch with ISA perhaps.
The µArch is the CPU core itself.
µArch means microarchitecture in a broader physical structure sense
This is my point. Qualcomm treats Oryon more as a brand name for its CPU block. So this Oryon V1 as Phoenix and Oryon V2 as Pegasus versioning is more of a convention the community/leakers are using.So Oryon V1/Phoenix will be different to Oryon V2/Pegasus from a microarchitectural sense, though not necessarily a hugely disruptive change.