AMD Ryzen 7 5800X3D Continues Showing Much Potential For 3D V-Cache In Technical Computing

[phoronix]

Phoronix: AMD Ryzen 7 5800X3D Continues Showing Much Potential For 3D V-Cache In Technical Computing

As a follow-up to last week's AMD Ryzen 7 5800X3D Linux review, here are some additional Linux benchmarks of this first AMD Ryzen CPU with 3D V-Cache.

https://www.phoronix.com/vr.php?view=31091

[geearf]

Though due to AM4 socket limitations is possibly why such a flagship 3D V-Cache desktop model does not currently exist.

What do you mean by that?

[skeevy420]

So does anyone know if the 5800X3D is going to be the best & last processor for the AM4 platform in regards to overall desktop usage? Will the 5700G be the best AM4 desktop APU?

Every time I see these benchmarks it makes me even more curious about what I might replace my 4650G with if I ever do that because the 3D has very nice results. Damn near everything tested is better. For me it's hard to decide between an obvious performance advantage and a CPU with on-board graphics so two GPUs aren't necessary for 3D accelerated VMs as well as having a backup in case my GPU dies.

I really hope they have an R7 7800G3D as the high-end APU for AM5's debut.

[ermo]

Doing an AM4 8-core/16-thread Zen 3-based APU w/ 3D V-Cache + a RDNA2 GPU element would be excellent.

Just about the best low-power, SFF linux dev box w/ECC support you could imagine.

[pete910]

What do you mean by that?

I too am intrigued to know the reason to that statement !

[EvilHowl]

So does anyone know if the 5800X3D is going to be the best & last processor for the AM4 platform in regards to overall desktop usage? Will the 5700G be the best AM4 desktop APU?

Every time I see these benchmarks it makes me even more curious about what I might replace my 4650G with if I ever do that because the 3D has very nice results. Damn near everything tested is better. For me it's hard to decide between an obvious performance advantage and a CPU with on-board graphics so two GPUs aren't necessary for 3D accelerated VMs as well as having a backup in case my GPU dies.

I really hope they have an R7 7800G3D as the high-end APU for AM5's debut.

I don't think AMD is going to release more products for the AM4 lineup given that AM5 is almost here, though we don't how much time is AM4 going to be around once AM5 gets released.

I think AMD will stop producing the 5800X3D as soon as they release any 3D V-Cache CPU, but they might continue making CPUs and APUs to fill the entry-level market. I don't think they will release cheap AM5 products at launch, knowing that it's going to be a DDR5 only platform and though DDR5 prices are quickly improving, they are not very good. So it wouldn't make sense to release cheap SKUs if the end user has to pay a premium for the RAM.

So, I guess the 5800X3D will remain as the best AM4 gaming CPU and the 5700G will remain as the best AM4 APU.

[flower]

Doing an AM4 8-core/16-thread Zen 3-based APU w/ 3D V-Cache + a RDNA2 GPU element would be excellent.

Just about the best low-power, SFF linux dev box w/ECC support you could imagine.

tbh that's what i am waiting for. my nas atm uses an 4750g, looking at the ffmpeg improvements through 3D-Cache this would be interesting.

i dont care that much about rdna2 though

[V1tol]

These tests illustrate a enormous mistake Intel and AMD did by not making 3- or 4-channel memory controller in these processors. 177% performance improvement in Zstd just by adding some cache and even lowering CPU clocks. Maybe not a big deal but compression/decompression speed is very important nowadays - everything from http to zswap is compressed by different algorithms.

[schmidtbag]

These tests illustrate a enormous mistake Intel and AMD did by not making 3- or 4-channel memory controller in these processors. 177% performance improvement in Zstd just by adding some cache and even lowering CPU clocks. Maybe not a big deal but compression/decompression speed is very important nowadays - everything from http to zswap is compressed by different algorithms.

That is a good point. It would be far cheaper to just add more memory controllers rather than more cache. It helps too since you can opt for ECC. While a CPU does have a nice bulky heatsink to help protect a little bit from bit flips, a stacked V-cache is bound to be a noisy and hot environment for the transistors, so, I do question the long-term reliability.

[Danny3]

I imagine the Zstd compression speed would be even better if the kernel guys would upgrade the Zstd algorithm to the latest upstream version!