Search results

I spot them often for 110 to 150 usd shipped international. There is already a thread on this of a USA deal here. They come and go.

M40 were available for half that. not sure why you would buy this one

why are you bottlenecked by storage? if you are swapping to storage during training, arent you doing something wrong?

Every time I go to Great Deals/For Sale, my purpose is to see if there are any new deals or sales. So I have to change the sort from Last Message Descending to First Message Descending (4 clicks). This is stressing me out. Possible to remember last sort?

there are few or no good reason to buy a slow 12gb vs a 4 times faster 8gb. the 24gb model might be more useful and worth the slowness, but they are never cheap.

singles are 32 bit. halves are 16 bit, doubles are 64 bit you dont need these cards for deep learning. just get a consumer RTX card. even a RTX 2060 demolishes these. and if you want to spend more, a 3060Ti is all you need. Demolishing $8000 cards.

only for doubles. who uses doubles these days tho. k80 you have back to front. its much much slower on singles.

thanks for post. btw what legitimate purpose is hashcat?

what are you doing with these bountiful pcie boards ? and why these over cheap x99 dual boards from china?

Its been a while. I hope you can find a buyer. have you considered selling them on ebay individually?

i complained about this on the previous design. The Deals and For Sale pages, its 100x more useful to sort by Start Date, not last post. I want to see the latest deal or sale. Why do I care what people are chatting about on irrelevant threads? Yet you have removed the sort option completely on...

hard to price because of storage. recommend sell that separately.

they are cheap because noone can use them because E7 motherboards are as rare as hens teeth.

older HP gear I had to use old IE to get the Java to run. did you try IE?

isnt that motherboard a Xeon-W motherboard not E5 scalable motherboard? i guess thats what you meant by foolish?

you would look at the program you are running and what AVX modes it utilises. each cpu can do a certain number of AVX per tick. if the program uses AVX-512 the Xeon scalables are twice as fast as 2nd gen EPYC per tick. if its just AVX, they will be the same per tick. older 1st gen EPYC are half...

actually cores multiplied by all-core boost frequency multiplied by operations per tick

the only people who want you to use big models are the people selling the hardware. its a scam sorry. who did the largest state of the art 33GB model recently? nvidia. hmm what a coincidence. dont be a sheep homey (unless you make money being a sheep and i cant blame u.. :p)

gpu direct and Nvlink are Nvidias way to cover up deep learnings dark secret that it doesn’t scale. it’s not embarrassingly parallel. Researchers are too busy Buying these as a stop gap and not focusing on the hard questions. this is why I say a 2080ti farm is the go for actual work these...

There are ways around these problems but most people using these aren’t inventing anything new they are just using cookie cutter tensor flow because their department got an ml budget to blow.