Best GPU for running Lc0

[Solipsistic-Dreams]

I'm about to build a new rig and am trying to decide what GPU I need for competent analysis with Lc0.

So far my plan is a Ryzen 9950 CPU, 64GB ddr5 ram, and was considering the Nvidia 5070 GPU. I will be using the rig for lots of heavy workload multitasking, video editing, some gaming, and lots of chess analysis running stockfish, komodo,and Lc0.

It's a big step up in price for the 5080 or 5090 and was wondering if it would make sense to consider one of those more expensive cards.

Also, does anyone have any experience running Lc0 on one of the AMD GPUs?

[Graham Banks]

Doesn't really matter because whatever graphics card you use, it will become outdated quite quickly.

[Werewolf]

I'm about to build a new rig and am trying to decide what GPU I need for competent analysis with Lc0.

So far my plan is a Ryzen 9950 CPU, 64GB ddr5 ram, and was considering the Nvidia 5070 GPU. I will be using the rig for lots of heavy workload multitasking, video editing, some gaming, and lots of chess analysis running stockfish, komodo,and Lc0.

It's a big step up in price for the 5080 or 5090 and was wondering if it would make sense to consider one of those more expensive cards.

Also, does anyone have any experience running Lc0 on one of the AMD GPUs?

I'd do it like this:

a) decide your budget for chess.
b) decide your primary engine for chess (probably SF vs Lc0)
c) then decide where to spend your money.

If you're sure it's Stockfish, for example, I'd probably save on the GPU and consider a Threadripper. If b) is mainly Lc0 as you hint at, yes I'd consider the bigger cards. I use a 4090 and since we know how quickly GPUs increase in speed we can be reasonably confident in lifespan.

I believe the 5090 is around 25% faster than the 4090 for Lc0. Comparision between the cards was done on this forum very recently IIRC, and the TFLOPS on Tensor cores are a reasonable place to start. If you buy a 5090 I doubt you'll feel compelled to change it within 3 years unless there's an unexpected breakthrough in GPUs or your card dies from overheating.

[Solipsistic-Dreams]

I'm about to build a new rig and am trying to decide what GPU I need for competent analysis with Lc0.

So far my plan is a Ryzen 9950 CPU, 64GB ddr5 ram, and was considering the Nvidia 5070 GPU. I will be using the rig for lots of heavy workload multitasking, video editing, some gaming, and lots of chess analysis running stockfish, komodo,and Lc0.

It's a big step up in price for the 5080 or 5090 and was wondering if it would make sense to consider one of those more expensive cards.

Also, does anyone have any experience running Lc0 on one of the AMD GPUs?

I'd do it like this:

a) decide your budget for chess.
b) decide your primary engine for chess (probably SF vs Lc0)
c) then decide where to spend your money.

If you're sure it's Stockfish, for example, I'd probably save on the GPU and consider a Threadripper. If b) is mainly Lc0 as you hint at, yes I'd consider the bigger cards. I use a 4090 and since we know how quickly GPUs increase in speed we can be reasonably confident in lifespan.

I believe the 5090 is around 25% faster than the 4090 for Lc0. Comparision between the cards was done on this forum very recently IIRC, and the TFLOPS on Tensor cores are a reasonable place to start. If you buy a 5090 I doubt you'll feel compelled to change it within 3 years unless there's an unexpected breakthrough in GPUs or your card dies from overheating.

I was hoping for a more technical response, but thanks for your input anyway.

[Ras]

Moderation: Removed post by member "Solipsistic-Dreams" (charter: 2. Are not abusive in nature, 3. Do not contain personal and/or libelous attacks on others, 6. Are not bullying, or vicarious bullying). New members are expected to read the charter before posting: Charter.

One day ban for "Solipsistic-Dreams", time enough to read the charter.