Avlin67

will they release great open sources models to use ?

did you try p2p enabled driver and proper nccl env vars ?

130 token per seconds on dual RTX 5090 for FP8 version

but my dual 5090 is way faster your mac

doesnt show 24, 48, and 96GB that are quite common in DDR m5 and also faster sometimes.

i have 4x48 6400RDIMM, how much it is now ?

just installed arch + hyprland. so reactive and so easy...

you never had WHEA errors... or pll issue on cpu C state transition...

remembering pushing i7 920 on dry ice with acetone by the time... also voltmod nforce 2 chipset to cranck bus clock for opteron 144. So cool !

is it the same font rendering ? if not it makes quite an akward feeling

tired of jeff glinglin everywhere...

try with 800W xeon/threadripper

where is the image ?

why not push metrics to grafana ?

you can get 8TFlops of fp64 on xeon 6980P which is 6K€ now

sorry, but looking at previous article, you computer cooling seems inadéquate for sustained loads

get a w790 motherboard with proper xeon. If it says 100° DTS it still not a good idea to run it 100° for 3hours, because it id already throtling.

what's the matter ?

so author is not aware of cpu odd cores. ok. he did not use a phenom X4 with only one core unlocked instead of 2 so you end up with a phenom x5

where is Phi 5

wrong it is approx half bandwith

xeon 6980P which now costs 6K€ instead of 17K€

I suggests using a Xeon w7 2595X so that you have 26P cores and 0E cores

ssd wear ?

i think it is the same with reddit

she do not look the same

you can still fold

discipline is all you need

AHCI with typo... oh boy

muscle size is to heat the body, more than giving strengh. human muscles are inefficient on purpose

well, at this level, it is almost a GUI why stick to terminal ?