Yes, SVE disabled on Qualcomm SoCs:What!?
They did WAHT?!!Do you mean the one about ARMv8 vs ARMv9? Isn't the main selling point of ARMv9, SVE2? If yes, losing it is no big deal in the current situation given that Qualcomm has disabled it in the current generation of chips
This is very interesting. We talk about vector units for x86 cores all the time, but rarerly so for ARM cores.X2 did have 256 bit units I think, but then X3 went back to multiple 128 bit units instead for some reason - possibly the same reason AMD are sticking with 256 bit instead of upgrading to 512 bit units in Zen4.
ST performance absolutely matters, and it is what determines how 'smooth' and 'snappy' a phone is.Doesn't concern me at all.
It's an interesting thing to measure with Geekbench, but in reality smartphones really don't see much benefit from better CPUs anymore, even with Android's less than efficient system architecture.
Ooo. What about Mediatek/Exynos/Tensor?Yes, SVE disabled on Qualcomm SoCs:
Problem is, I'm not even sure at this point what the average user would test it with.Ooo. What about Mediatek/Exynos/Tensor?
As far as I am aware even the latest still correspond to an advanced v8-A ISA.What about Apple's P cores?
They did WAHT?!!
Sheesh, that's almost as bad as Samsung gimping the A57 to a 32 bit only chip in its first Exynos implementation 😅
That being said tho, SVE2 designs in mobile so far really take advantage of the higher vector width capacity.
X2 did have 256 bit units I think, but then X3 went back to multiple 128 bit units instead for some reason - possibly the same reason AMD are sticking with 256 bit instead of upgrading to 512 bit units in Zen4.
As far as I remember that's correct: Fugaku has SVE with 512-bit registers and Neoverse V1 has SVE with 256-bit registers. Both of these chips also are SVE only, not SVE2.This is not the case. No Cortex-X or Cortex-A core has ever had SVE beyond 128b, probably to avoid issues when migrating threads between smaller and larger cores. The only devices with greater-than-128b SVE have been server silicon.
So does SVE2 not solve that issue?This is not the case. No Cortex-X or Cortex-A core has ever had SVE beyond 128b, probably to avoid issues when migrating threads between smaller and larger cores. T
Ah therein was my confusion, I was equating V1 to X2 😅This is not the case. No Cortex-X or Cortex-A core has ever had SVE beyond 128b, probably to avoid issues when migrating threads between smaller and larger cores. The only devices with greater-than-128b SVE have been server silicon.
Ah therein was my confusion, I was equating V1 to X2 😅
I still don't get the whole migrating limits malarky tho.
Why not just finish the code currently running with larger vectors and then start running the rest with smaller vectors if the situation demands it?
So does SVE2 not solve that issue?
Very good catch that the omission of Apple seems to be an oddly specific one.That's about as wide of a net as you could cast without catching Apple, I suppose.
Very good catch that the omission of Apple seems to be an oddly specific one.
Not lifestyle companies?Apple, for reference, is usually lopped into the big bucket industry of Information Technology.
ODM's are playing a numbers game with this statistic hoping that the average consumer gets blinded by it into oooooooh reactions without actually researching whether it actually benefits them.And also better AI performance (45 TOPS vs 10 TOPS for MTL per the latest leak)
Is RTX actually beyond early days now? I don't think an average consumer is intending to look at the price class of GPUs necessary to make today's RTX run smoothly.On the consumer side AI/ML feels much like the early days of RTX right now.
The average consumer won’t even know which of 1 GOPS or 1 TOPS is larger.ODM's are playing a numbers game with this statistic hoping that the average consumer gets blinded by it into oooooooh reactions without actually researching whether it actually benefits them.
Source: https://www.xda-developers.com/windows-hudson-valley-update/Getting ready for the Snapdragon X Elite
One thing that's interesting is that Hudson Valley will be based on the Germanium platform release, and that platform update is set to hit RTM in April. However, Hudson Valley itself won't RTM until August, with a general release in September or October. The platform changes have more to do with the underlying tech, and it looks like Microsoft wants to have it ready earlier so that Arm devices powered by the Snapdragon X Elite can ship with it preinstalled.
Indeed, it's said that the Germanium platform has important changes for the Snapdragon X Elite and these PCs can't be shipped with the current version of Windows 11, but manufacturers want to ship them in June 2024. As such, these Arm-based PCs will ship with the Germanium platform release, but they won't have all the Hudson Valley features out of the box. They'll have to wait for the update coming a few months later, but it will simply be a cumulative update. For everyone else, Hudson Valley will release alongside the Germanium platform release as one big feature update, like Windows 11 was to Windows 10.
DCS: 8G4 is codenamed SUN, adopts 2+6 structure, and the production process is TSMC 3nm.
Furthermore, he made an interesting statement, saying, "The design performance of Qualcomm's Oryon architecture has significantly improved, posing a threat to Apple."