|
dumb question but if I try to buy a game on steam with bitcoin, it talks about supplementing the base cost with an additional fee (tip?), and I understand that upping the fee makes the transaction go thru faster? how the heck does that work
|
# ¿ Jun 24, 2017 06:08 |
|
|
# ¿ May 1, 2024 02:07 |
|
Thanks for that. Based on "For the median transaction size of 226 bytes, this results in a fee of 81,360 satoshis", am I right in saying that I'd need to "tip" 2 bucks and change to get my transaction thru a < 35 minute time window. lol
|
# ¿ Jun 24, 2017 06:21 |
|
makes you wonder how many opportunistic sysadmins or students are mining using their corporate or university CUDA racks on the down-low
|
# ¿ Jun 24, 2017 10:21 |
|
yeah it's a dumb bubble with kids spending a couple grand on video cards to get rich but at the same time, i love that we're living in Neal Stephenson's world now.
|
# ¿ Jun 24, 2017 22:57 |
|
Yeah I'm going to have to see the receipts on a 50% annual failure rate. Jury-rigged Geforce CUDA desktops might fail more often than Teslas mounted in a data centre setup but that's not really an apples to apples comparison. And AIB Geforce boards have better cooling design than Teslas.
|
# ¿ Jun 25, 2017 02:10 |
|
For a single model, you can parallelize by either having each GPU hold the entire model but train on different batches of data at a time, or having each GPU hold a separate chunk of the model and flow the same batch of data through. How easy it is to do this depends on the ML framework you're using. This is lot easier if you're doing model averaging or stacking, you can just have each GPU work on a separate model.
|
# ¿ Jun 28, 2017 02:32 |
|
|
# ¿ May 1, 2024 02:07 |
|
Paul MaudDib posted:Do people find that risers are actually reliable? I tried them once, I did all the usual stuff (turn PCIe speed down to 1.0x, enable 4G+ addressing, etc), and I was still getting maybe a 48-hour MTBF with 4 or 5 GPUs on a board. Honestly it seems like it makes more sense to go with fewer but faster GPUs and to not use risers. I actually use 16x-to-16x risers for 4x 1080 Tis for Machine Learning stuff and they work perfectly fine and are a lot more intensive than mining. By all accounts, the 16x-to-1x stuff works well for mining. I got my risers from a B&M place that sells the lower end cables for miners and they're doing brisk business. shrike82 fucked around with this message at 09:39 on Nov 29, 2017 |
# ¿ Nov 29, 2017 09:36 |