• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Question x86 and ARM architectures comparison thread.

Page 33 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
511 said:
Skymont on decent bit behind node refinement and horrible uncore we don't have an Apples to Apple comparison pun not intended insert core on windows vs core on MacOS.
Click to expand...
this was a Linux vs macOS SPEC17 comparison. macOS is 4% faster in SPEC.
1773986880063.png

But the main point is the M core uses only 2.5 watts to reach 10.5 points in SPEC Int. I don't think theres any core from others that comes close to that perf/w yet.
 
poke01 said:
this was a Linux vs macOS SPEC17 comparison. macOS is 4% faster in SPEC.
View attachment 140355

But the main point is the M core uses only 2.5 watts to reach 10.5 points in SPEC Int. I don't think theres any core from others that comes close to that perf/w yet.
Click to expand...
well the issue is he is using -O3 with clang vs David Huang using GCC 12 and and -O2 flag matters and we won't know how the power measurements were taken Apple Power metrics shouldn't be taken at face value.
 
Last edited:
poke01 said:
he also tested gaming and Strix Halo won there. That’s embarrassing for Apple considering RDNA3.5 is one generation behind.
Click to expand...
Halo is using a slightly tweaked 2022 architecture with about half the memory bandwidth of the 32 GPU core M5 Max. And it's on 4nm instead of 3nm. But I guess ~40W can solve lots of problems.
 
poke01 said:
he also tested gaming and Strix Halo won there. That’s embarrassing for Apple considering RDNA3.5 is one generation behind.
Click to expand...
Doesn't Apple Silicon being TBDR means benchmarking the same titles that has a native Mac port being better? Also Apple Silicon relies on on chip bandwidth for GPU bandwidth rather than DRAM isn't it?
 
MerryCherry said:
What? No.

In Apple Silicon the DRAM is on-package, not on-chip. It is technically no different from DRAM soldered to the motherboard like most laptops, except that you get some power and space savings.
Click to expand...
Different graphics rendering technique, TBDR relies on advanced on chip cache bandwidth and efficiency for rendering workloads/tiles and passes before being read/written to DRAM is what I'm saying, unlike IMR that AMD uses for Strix Halo
 
S'renne said:
Different graphics rendering technique, TBDR relies on advanced on chip cache bandwidth and efficiency for rendering workloads/tiles and passes before being read/written to DRAM is what I'm saying, unlike IMR that AMD uses for Strix Halo
Click to expand...
IMR's havent been IMR's for like 10+ years , they have binned things for ages.
 
poke01 said:
he also tested gaming and Strix Halo won there. That’s embarrassing for Apple considering RDNA3.5 is one generation behind.
Click to expand...
In Blender, the basic M5 outperforms the Strix Halo. There are still plenty of ways to optimise performance for gaming, but unfortunately the developers aren’t making full use of them.
 
mvprod123 said:
In Blender, the basic M5 outperforms the Strix Halo. There are still plenty of ways to optimise performance for gaming, but unfortunately the developers aren’t making full use of them.
Click to expand...
www.michaelstinkerings.org

Apple M5 GPU Roofline Analysis

In this deep dive, we will examine M5's GPU performance across various workloads. TLDR: It can be very powerful if the programmer knows how to use it correctly.
www.michaelstinkerings.org www.michaelstinkerings.org

Idk how accurate this guy is, but it suggests that games needs a good port or the games will only use half of the GPU's power
 
mvprod123 said:
In Blender, the basic M5 outperforms the Strix Halo. There are still plenty of ways to optimise performance for gaming, but unfortunately the developers aren’t making full use of them.
Click to expand...
i know it does in blender but the 3nm M5 shader core is weaker than a RDNA 3.5 on 4nm in pure raster but this only matters for old games. I bet if RT was enabled here, M5 Max would be much faster.

All AAA games released this year use RT and the PS6 AAA games will use even more demanding RT.
 
poke01 said:
i know it does in blender but the 3nm M5 shader core is weaker than a RDNA 3.5 on 4nm in pure raster but this only matters for old games. I bet if RT was enabled here, M5 Max would be much faster.

All AAA games released this year use RT and the PS6 AAA games will use even more demanding RT.
Click to expand...
The 8060s performs on par with the M5 Pro in Steel Nomad. I think the issue is not with the M5’s shader cores (they’re not weak) but with game optimisation. The reason Blender works so well is not only down to Apple’s superior RT implementation, but also to excellent optimisation and a well-implemented Metal backend. Apple has assigned a dedicated team of engineers to work on optimising Blender for Mac. If Apple had formed a separate engineering team to optimise game ports for Mac, the results would have been better.

1774263020168.png
www.3dmark.com

Result

www.3dmark.com www.3dmark.com
 
mvprod123 said:
The 8060s performs on par with the M5 Pro in Steel Nomad. I think the issue is not with the M5’s shader cores (they’re not weak) but with game optimisation. The reason Blender works so well is not only down to Apple’s superior RT implementation, but also to excellent optimisation and a well-implemented Metal backend. Apple has assigned a dedicated team of engineers to work on optimising Blender for Mac. If Apple had formed a separate engineering team to optimise game ports for Mac, the results would have been better.

View attachment 140534
www.3dmark.com

Result

www.3dmark.com www.3dmark.com
Click to expand...
If only Apple acquired a game studio from back then lol
 
S'renne said:
www.michaelstinkerings.org

Apple M5 GPU Roofline Analysis

In this deep dive, we will examine M5's GPU performance across various workloads. TLDR: It can be very powerful if the programmer knows how to use it correctly.
www.michaelstinkerings.org www.michaelstinkerings.org

Idk how accurate this guy is, but it suggests that games needs a good port or the games will only use half of the GPU's power
Click to expand...
That is an impressive and detailed analysis. I am puzzled, given how much he assesses LLM performance implications, that he doesn’t even mention the neural accelerators, let alone test them.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top