Ampere Altra vs. Amazon Graviton2 Linux Performance Benchmarks

[pWe00Iri3e7Z9lHOX2Qx]

it is a simple idea what makes the M1 fast...

it is the concept of zero copy... means "no buffers" and fast direct connect to very fast and paralell ram. M1 has 8 channel ram for its 12 cores 8 channel

[chomp]

the 8 high performance cores have 12 MB L2 cache
the 4 low performance have 4 MB L2 cache

[chomp]

problem with this design is not the 12core version M1 but the M2 with 32 cores...

M1 is an 8 core part with a 4/4 split, not a 12 core one with an 8/4 split.

[qarium]

M1 is an 8 core part with a 4/4 split, not a 12 core one with an 8/4 split.

ok. no problem with that. then M2 will have 8-32 fast cores.

[coder]

So after comparing some of the results, the main takeaway is that the performance per MHz is identical - in all the 64core test results I checked the Ampere Altra is about 32% faster, at 32% higher clockrate.

For the benchmarks that are compute-bound, sure. But many are clearly memory-bound. According to the geomean, the 64-core Altra is only 17,6% faster, which is about half what you'd expect for linear scaling.

One thing to watch out for is benchmarks like Coremark, which is carefully to avoid memory bottlenecks. So, they really tell you nothing about scaling (other than thermal or power throttling, which Altra avoids due to its efficiency and the lack of an effective turbo).

[coder]

believe it or not but they even plan to put the SSD on the cpu die do. to save the copy between ssd and ram.

This can't be true, for at least 2 reasons. First is practical -- just look at all of the space occupied by chips on a M.2 SSD board and you'll see how much they would have to cram into it. Also, SSDs don't like super-high temperatures, so that would place an artifically low thermal ceiling on the CPU.

The second is that there'd be no practical benefit. The fastest SSDs are Optane (or 3D XPoint), which is still about 1/10th as fast as DRAM. If you look at NAND, it's even slower.

So, it's difficult to see how they would cram those dies in the CPU package, and it's not clear what benefit i would even provide. Also, SSDs tend to wear out faster than CPUs.

[f0rmat]

I really wish that i could unterstand what you said... sorry i do not.

what is ARC Executes ?

according to google something like this: "ARC executes U.S. Army Patriot Missile Battery Redeployment to Germany. BREMERHAVEN, GERMANY – Under contract with American Roll-On Roll-Off Carrier (ARC), M/V Merchant discharged a U.S. Army Patriot missile battery in Bremerhaven on January 2 after it had been used in support of ongoing NATO operations in Turkey."

https://www.arcshipping.com/news/arc-executes-u-s-army-patriot-missile-battery-redeployment-to-germany/

That ARC is the Allied Reaction Corps - it is a NATO organization. They do have M1s - but they are quite a bit different than the Apple ones you are referring to.

[qarium]

That ARC is the Allied Reaction Corps - it is a NATO organization. They do have M1s - but they are quite a bit different than the Apple ones you are referring to.

we really should benchmark the M1 tanks on phoronix.com... maybe we can profile their software stack and maybe we can tune their aiming performance speed to kill faster..

[f0rmat]

we really should benchmark the M1 tanks on phoronix.com... maybe we can profile their software stack and maybe we can tune their aiming performance speed to kill faster..

Interesting idea...phoronix.com could do a comparison of the software stack of the M1, Merkava, Leopold, T-72M, and T-90. They might also include - in a different test utilizing 150mm and up architecture - the fire control systems for the M109A6, 2S19, AS90, KRAB, Caesar, and G6. I think the only issue would be getting those software stacks. I might be wrong, but I do not think those software stacks are open source. At least, I have not found them in GitHub or anywhere else.

[name99]

This can't be true, for at least 2 reasons. First is practical -- just look at all of the space occupied by chips on a M.2 SSD board and you'll see how much they would have to cram into it. Also, SSDs don't like super-high temperatures, so that would place an artifically low thermal ceiling on the CPU.

The second is that there'd be no practical benefit. The fastest SSDs are Optane (or 3D XPoint), which is still about 1/10th as fast as DRAM. If you look at NAND, it's even slower.

So, it's difficult to see how they would cram those dies in the CPU package, and it's not clear what benefit i would even provide. Also, SSDs tend to wear out faster than CPUs.

At some point it may make sense to put an STM pRAM in the package. Maybe as IO buffer storage (which also frees up RAM).
Not exactly SSD on the SoC, but in that direction.
Mainly depends on how rapidly STM density continues to rise.

[qarium]

Interesting idea...phoronix.com could do a comparison of the software stack of the M1, Merkava, Leopold, T-72M, and T-90. They might also include - in a different test utilizing 150mm and up architecture - the fire control systems for the M109A6, 2S19, AS90, KRAB, Caesar, and G6. I think the only issue would be getting those software stacks. I might be wrong, but I do not think those software stacks are open source. At least, I have not found them in GitHub or anywhere else.

they are FOSS/FLOSS but not public the manufacturer gives the source code to the customers who buy the tanks but they never purplish it to public.

most people think opensource means public... no it it also opensource if an developer hands the source code to the customer and no one not the developer and not the customer ever make it public,

[microcode]

also i have to tell you that RISC-V-GPU is already death all the projects about RISC-V-GPU switched to IBM POWER ISA...

"all" the projects as in just one of them, which is complete vapourware? Esperanto prototyped a RISC-V based GPU and wrote a Mesa for it, seemed to do okay considering it was mostly a toy.