Recent content by Cixelyn

Sounds good -- will DM in a bit! Though I wasn't really planning on getting two...

Snagged 1x. Thanks!

We managed to find a deal on blower 3090s so I guess I'll have a chance to revisit our earlier benchmarks now that CUDA 11.2 has been out for a while.

From page 13 of the user manual: The switch is actually mandatory since you need the ConnectX-6 cards directly linked to the GPU on the same PCIe Root in order for GPUDirect RDMA to work for large multi-node deployments. You'll notice that each GPU also has a direct path to one of the NICs on...

I could definitely see an A100 having higher latency in specific cases; especially if the pinning isn't right, and data has to traverse both a PCIe Switch as well as an Infinity Fabric link before hitting the target GPU. Just due to the nature of 3090 systems, they're much more likely to be...

Yeah I think these boards need a fair bit of cooling, even moreso than the normal servesr. We had a E300-9D-4C8TP that slowly died on us (likely due to heat death). One of the 10Gbase-T NICs failed first after about a year. And then a few months later the entire motherboard died. Ended up...

Yeah in our experience the scaling breakpoints are: - 1 GPU: easiest, you don't need to write any special model or data parallel code and stuff "just works" as long as you have enough VRAM. - 4 GPUs: no longer fits in a single EATX chassis _and_ you've maxed out an entire 15A standard USA...

The Original TF-based SG2 repo is what we were using to test, config-f with batchsize 32. Chassis fans are all maxed. Rack also has a rear door with a giant maxed active fan. We have one system w/ pairs nvlinked, and one without, but it didn't make a big enough difference for our particular...

lol I wish. The stock situation hasn't improved at all in the past 4 months.The whole parts shortage + scalper situation is driving me nuts; currently trying to get our hands on AMD 5000 series for workstation builds and can't :confused: Wasn't in consideration due to chassis constraints. Our...

We've had issues attempting to use 3090s in distributed training. I tweeted some graphs here. Not sure if this is driver handicaping or not. Granted this was in October of last year, so the driver situation might have improved a bit, but at the time the downclocking issues was severe enough...

We use a UCoustic 9210i in our office, and we're pretty happy with the sound dampening. Instead of our office sounding previously sounding like a torrent of screeching harpies, it now just sounds like there's a large always-on commercial AC unit in the corner. Can't do infinite sound...

Yeah, honestly I think these projection articles are the best content. It's one thing to regurgitate benchmark numbers, but it's another to actually editorialize a little bit and draw some projections and conclusions. Especially given your industry experience -- I find this sort of content...

At least in the US, it's much more difficult to get M6 in the average hardware store vs. 10-32 or 12-24. On one hand, metric is better down with imperial, but on the other hand the convenience of being able to get 10-32 on a moment's notice from the mom-and-pop hardware store down the street is...

Yeah hmm unfortunately gonna guess that Onyx is going to be killed in favor of Cumulus in a year or two. I liked it too, but it doesn't make sense to fund development of two separate switch OSes esp. when Mellanox already has a SKU + deep integration for Cumulus-based Spectrum switches.

Ooh, that video is nice. Please report back with thoughts if you end up purchasing one of the new POE Microscanners! :D