barebones/light RX 570 for $99

[kracker]

Pretty sure nothing else can come close to this one for the price...

(light aka only one DVI-D output, came in a brown cardboard box, literally nothing fancy)

https://www.amazon.com/Sapphire-Rade.../dp/B07MCDNQX2


Getting ~500ish GHz/days on 90M exponent 76 to 77 bits right now on mfakto,
For gpuowl, 80M/4608K FFT - 4.24 ms/iter
70M/4096K FFT - 3.53 ms/iter

[Mark Rose]

That's compelling for gpuowl!

[xx005fs]

Dang they PL'ed it to 75W i assume as there's no external 6 pin connector. This would be great for building a high dense PRP testing rig as not a lot of cooling will be required. For 99$ this is an absolute steal. If the card is tweakable (which I assume it will cuz it's AMD), then undervolt it to push for higher clock as the efficiency is already stunning and undervolt will only push that efficiency even higher.

[kracker]

Originally Posted by xx005fs

Dang they PL'ed it to 75W i assume as there's no external 6 pin connector. This would be great for building a high dense PRP testing rig as not a lot of cooling will be required. For 99$ this is an absolute steal. If the card is tweakable (which I assume it will cuz it's AMD), then undervolt it to push for higher clock as the efficiency is already stunning and undervolt will only push that efficiency even higher.

There is a single 8 (6+2) power connector. And yes... it is as tweakable as any other card (core clock, memory clock, power limits (need to push this one up from the beginning since running mfakto will hit the 130W power limit easily and it'll start dropping the core clock... right now it's using around 150W.)

[xx005fs]

Originally Posted by kracker

There is a single 8 (6+2) power connector. And yes... it is as tweakable as any other card (core clock, memory clock, power limits (need to push this one up from the beginning since running mfakto will hit the 130W power limit easily and it'll start dropping the core clock... right now it's using around 150W.)

Interesting... the screenshot on amazon shows no power connector, weird... I think radeon cards are generally more fit towards PRP or LL (if implemented on OpenCL properly unlike cllucas which is very slow compared to PRP on GPUOWL) than TF because the int32 perf is very weak compared to turing/volta, but the double precision is quite good along with memory bw.