Processors with 512-bit AVX instruction sets: finaly a break through?

[Artas1984]

Recently something caught my attention. I was checking Anandtech Core i9 9900K review and was blown away by this:

When enabling 512-bit AVX instruction set, which is present in all Skylake-X processors and in all Xeon W processors for LGA2066, the Core i7 7820X manages to smash Core i7 9900K by 450 %!!!

You would say: ye, but is just some 3D simulation, that no one uses this stuff in real everyday life.. Now wait till you read this:

http://x265.org/category/releases/

The latest version of HEVC H.265, that was released in 2018.05.21 now has support for 512-bit AVX instruction sets. You know what that means? That means if are encoding an hour long 4K video with HEVC H.265 latest codec, the Core i7 7820X MIGHT BE ABLE to encode the job 4,5 times faster than Core i9 9900K!!! Can you image spending four hours instead of one hour? Ok, it might not be that much of a difference, but surely this is finally some real good news for the Skylake-X, that got alot of bashing for various reasons last year. This, however, will not help the already very poorly accaimed Kaby Lake-X processors, as those do not have the 512-bit AVX unit.

Adding salt to injury, the Core i9 9900K has just 16 PCI-E lanes, the same amount since Sandy Bridge came to town. The performance benefit over Core i7 7820X and Ryzen 7 2700X purely comes from the clock speeds, but other than that Core i9 9900K is a result of 7 year old Intel stagnation. We had a 6 core mainstream CPU for X58 back in 2009, sure with way less MHz tp boast and much lesser IPC, but that is not the point.

We need more plugins, interfaces for programs those might utilize this 512-bit AVX set, and surely we need more benchmarks than what we have now.

 

[Vayra86]

Its not so much about performance as it is about usage of AVX512. But if the point was that 9900K isn't a fantastic buy, then yes, at its current price point I can only agree.

AVX is one of the straws Intel has to differentiate HEDT from MSDT in a meaningful way.

 

[Steevo]

AVX512 is essentially a list of predefined answers to computational challenges, in hardware, which is why its fast but will have limited functionality compared to the hardware decode in GPU's.

 

[JalleR]

I guess it is more that the X299 platform actually have advantage over the Z370/Z390 platform in some cases. (core vs core)

 

[ExV6k]

AVX is one of the straws Intel has to differentiate HEDT from MSDT in a meaningful way.

Actually, Cannon Lake supports AVX512 on the mainstream lineup (I saw a weird Core i3 8***U running around a few weeks ago, had AVX512F -and a few other AVX512 extensions- support).

Not sure if that's relevant though, intel may just be messing around / testing the process, I guess we'll have to wait and see.

 

[king of swag187]

X58 wasn't mainstream, its was the equivalent to the 7820X, the HEDT of the time. Consumers had X48/P45 and P55/H55

 

[R-T-B]

AVX512 is essentially a list of predefined answers to computational challenges, in hardware, which is why its fast but will have limited functionality compared to the hardware decode in GPU's.

That's really a massively incorrect oversimplification. It's a vector extension. It does vector math in the core. There is no answer list anywhere.

You can read what it actually does along with commands here:

https://en.wikipedia.org/wiki/Advanced_Vector_Extensions

 

[Vya Domus]

Remember that atrocious power consumption with the 9900K once overclocked ?

That's why AVX-512 is not present in mainstream parts.

There is no answer list anywhere.

I think he meant that it's a very specific hardware function "as an answer" for specific tasks.