virtuallynathan

Google Global Cache appliances are just off-the-shelf Dell/HP hardware, same for Facebook Network Appliances.

AMD said they are not expecting to be supply constrained.

They already said it’s PCS, 1900Mhz.

None are currently at 330km, and the first shells of v2 will be at 540-570km.

Spreadsheet: https://docs.google.com/spreadsheets/d/1tuhyh6kbyDQ2D418ug_A...

If each new satellite can put down more spot beams, you can cover more cells for longer. With the spectrum they have, they could put 8 beams on a cell with 100% duty cycle, which is probably around 3.2-4Gbps.

A new document from SpaceX outlining all of the advances in brightness mitigation on Starlink v2 satellites.

It's pretty low as long as its not a rev1_preprod, the rev2_proto2 and proto4 models use 30-50W for me.

Peak time speeds are definitely a bit lower than when I was one of the only customers in the area, but it's still very good. Almost always >50Mbps, which is faster than the DSL alternative available.

They barely work better than placebo, from what I saw in my reading of the literature.

As of 2013, I calculated the YouTube catalog is ~375PB. It's probably closer to 1EB now, if not 2-5x that much.

http://www.autonomoussystem.net/2013/07/youtube-slowness.htm...

Yes, I’ve helped family/friends order, I know of at least 20 dishes delivered over the last 12 months.

Have a look at starlink.sx, it’s possible a few planes have enough sats to establish laser links.

Current dishy is 30-40W under normal operating conditions, i’d assume a bit less than double for the twice-as-large dish.

They moved from behind Google to their own network, so it wasnt exactly turning it off.

They recently added a bunch of IPv6 addresses to their GeoIP file, and announced more via BGP, I suspect its in the works. FWIW, they arent any more specifically Geo located, so this won't fix your issues.

I got a 3 or 4U chassis and hacked in Noctua fans, made it silent.

SamKnows is what many ISPs use for this. The parent comment is Sam, and he Knows.

I 100% agree with this.

Luckily we have embedded OCAs and/or peering with almost every network of significant size, so their choice of transit rarely comes into the mix.

I believe in 1 case it was pathological issues with some networking equipment, but this was a recent thing, so I haven’t looked into everything yet.

At Netflix, we found in quite a few networks IPv6 had much better QoE than IPv4. In a limited number of cases, IPv6 was very broken. It was almost always equal or better.

Sadly not much stuff supports SRv6, and I'm not aware of many deployments.

These 2 data center facilities are super small - I hope they have rapid expansion ability if they want this to succeed.

It works great.

> Once they do though that will be a significant shift for anyone where intercontinental servers form a significant part of their usage.

I have strong doubts this capacity will be used for random user traffic. It's worth much more to use it to serve oceans, poles, islands, and areas where ground stations can't be built (yet). Other capacity could easily be sold to HFT firms, etc.

I use starlink full time for a remote job, and make frequent WiFi calling, facetime, meet, teams, slack huddle, and other calls. No issues.

Have you tried >1GbE Over your cat5? I’ve done 10GbE over short runs (100-150ft), and 5GbE or 2.5 should work over longer runs…

Wifi6E is around the corner for >1GbE wifi.

It’s very quick: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3957428/

Tesla Dojo Training tile (25x D1): 565 TF FP32 / 9 PF BF16/CFP8 / 11GB SRAM / 10kW

NVIDIA DGX H100 (8x H100): 480 TF FP32 / 8 PF+ TF16 / 16 PF INT8 / 640GB HBM3 / 10kW

Dojo off-chip BW: 16 TB/s / 36TB/s off-tile

H100 off-chip BW: 3.9TB/s / 400GB/s off-DGX