X2 Elite Extreme - SPECInt2017 Score
Posted by basedIITian@reddit | hardware | View on Reddit | 23 comments
David Huang has measured X2-94's P-core SPECInt 2017's score (in WSL2 environment) in the ASUS Zenbook A16
X2-94: 13.0
For comparison
M4 Pro: 13.7
9950X: 12.6
M3 Pro: 11.8
Max+ 395: 10.6
358H: 10.3
He also measured the power consumption via HW telemetry.
Total SoC consumption: 13-15 Watts
CPU Consumption: 10-13 Watts
(Source in comments)
-protonsandneutrons-@reddit
Great to see this added.
I really am amazed how slowly Intel & AMD release new uArches. Apple, Arm, and Qualcomm release a new P-core uArch every year, basically like clockwork. Every year. And Apple + Qualcomm ship hardware (!).
Intel desktop:
Intel laptop:
AMD desktop:
I'll do AMD mobile later, but it's even worse, IIRC.
I'm sure we can debate adding in the derivatives of Skylake, Alder Lake, Arrow Lake, Zen1, Ice Lake mobile derivatives, but those were nigh identical in uArch, if not actually the same uArch.
As far as I have seen, Apple, Arm, and Qualcomm have found to have actual IPC increases each generation.
basedIITian@reddit (OP)
The IPC gap b/w ARM and x86 keeps on widening every year and every generation.
AbhishMuk@reddit
IMO it's a direct result of competition. There's no other (big/western) x86 player to threaten Intel/AMD, so they can afford to "take it easy". Meanwhile Qualcomm is trying to one-up Arm (Nuvia), let alone its direct competitor (MediaTek). Phones are often sold on the basis of the specs sheets, and are a status symbol for many. And all this is to say nothing of all the other companies like RockChip that use ARM designs, or Samsung's former custom Exynos cores.
Arm wants to be competitive with Apple, Qualcomm wants to beat MediaTek... at least in India (and perhaps also in China and elsewhere), phone advertising is like 50x laptop ads. Makes sense that the players try to get more competitive.
Geddagod@reddit
Interesting power telemetry, \~30% better perf/watt in single core vs PTL. Just a depressing architectural gap.
basedIITian@reddit (OP)
He added the M5 score also now - 15.4 ~50% better than x86's latest and greatest.
Noble00_@reddit
And Huang mentioned, in a VM WSL2 env. Only thing next to it as far as Windows (non-desktop) goes is GB10 X925 core (cluster with 16mb l3$), 8% faster than it.
PTL (refreshed) core uArch are obvs behind, and IMO unlike headlines goes, the story isn't about the nodes, it really is about core designs. This performance gap with this on N3X/P mix, makes the compute tile on 18A underwhelming.
basedIITian@reddit (OP)
And it roughly matches Hardware Canuck's results @30 Watts
CB24 1T -
X2-94 160 388H 130 358H 125
Artoriuz@reddit
This puts it very close to desktop Zen 3, which looks considerably less impressive than what we see on GB6.
basedIITian@reddit (OP)
C1 Ultra is tested with Exynos 2600, it scores 11.9 - which is pretty impressive in itself. It sits between 8 Elite (11.1) and X2-94 (13.0)
Artoriuz@reddit
The point is that it loses to the X925 in the GB10, which is not supposed to be faster but ends up being faster due to having access to a better memory subsystem, more power and more cooling. The GB10 is not a phone SoC, so that makes perfect sense as it's not bound by the same constraints.
If Samsung released a laptop-class Exynos they'd almost certainly be able to push the C1 Ultra a little harder, and maybe catch Qualcomm.
basedIITian@reddit (OP)
The X925 in GB10 seems to score exactly the same as 8 Elite, no? That's what I'm looking at in the list.
R-ten-K@reddit
Huh? A mobile SKU outperforming the top desktop part is pretty impressive.
Geekbench mostly fits within cache, so it’s a good proxy for raw core performance. SPEC, on the other hand, exercises much more of the microarchitecture and exposes things like cache hierarchy and memory controller behavior. Looking at both gives a solid baseline for where things stand.
Right now, Apple still leads overall. And on the Windows side, ARM has taken the single core performance crown.
At this point, it’s pretty clear that ARM is where top end single core performance is heading (or has headed already really).
DerpSenpai@reddit
It's faster than Desktop Zen 5 and this is not the fastest SKU, there's another one with +6% perf in ST.
Geddagod@reddit
Zen 5 you mean. Desktop Zen 5.
This has it pegged at \~3% faster than the 9950x, gb6 has it at \~20% faster.
So you are right, this does look considerably less impressive than what we see on GB6, but there's also the point that Huang " tested in a virtual machine running a stripped-down OS" due to linux compatibility issues.
basedIITian@reddit (OP)
Also 4.7G vs 5G, that's another 6.4% just from the SKU difference.
AnyPomelo3352@reddit
I think Hyper-V doesn’t have much performance loss. From David’s previous tests, the ~4 GHz 8 Elite scores about 10.4. We also know that the SPEC integer IPC increase from 8 Eliteto 8 Elite 5 is around 8–9%. Based on that, we can estimate the 4.7G score to be slightly above 13, but not much higher.
The 8% integer IPC improvement was measured using Geekerwan, comparing a fixed 3.65 G8 Elite 5 with a 3.51 8 Elite.
AnyPomelo3352@reddit
I think 9950x means zen5, not zen3
Artoriuz@reddit
Yes, that was a typo. Brain got confused by the znver3 march used to compile the code, my bad.
Hour_Firefighter_707@reddit
Wait, so the x86 laptop paradigm is slower than an M3 Pro single core from 2023? And none of them is close? The only ones that might match it or surpass it slightly is th 9955HX/9955HX3D which are desktop CPUs stuck inside a laptop.
Embarrassing.
basedIITian@reddit (OP)
All current x86 laptop chips score worse in Specint2017 than 8 Elite, a mobile chip from 2024.
MysteriousLog6@reddit
hey do you know where I can find 8 Elite and 8 Elite Gen 5 Spec scores? Always been interested just couldn't find any
basedIITian@reddit (OP)
8EG5 hasn't been tested by David, but 8E is there in his list. Just go to the attached link.
IcyConsequence9107@reddit
No M5 Pro data?