Does anyone know how UV indexes reported by weather apps (including forecasts) work?
I've found that some sources report lower UV index on cloudy days, whereas some seem to indicate the same index as a sunny day.
It's not clear if these sources all have a different model for UV, some being more sophisticated or conservative, and whether any of them are backed up by a UV meter.
It is a prediction of UV intensity based on specific forecast location, time of day, day of year, prevailing weather, and any anomalies affecting the ozone layer. It is calculated by NOAA.
Essentially, yes. The DGX Sparks, at 128 gigs for ~$4700 are one of the cheapest ways to get enough high-ish speed memory cobbled together to run one of the more capable open weight models to run at home/for a small biz. This switch lets you combine the memory of 2-4 of them.
They are QSFP56-DD which can be split using breakout cables. So it's 4 x 400GbE ports, which could in theory be split into as many as 32 x 50GbE ports. (Although whether that then ends up being an economical thing to do is a different matter.)
Ahh yeah, I've seen people using those. I was thinking you'd ideally want to connect them full speed, but it looks like they're 200gbps ports anyway, and I guess even if they were 400gpbs, you don't really need that much for cross-GPU traffic for MoE models doing tensor parallel.
I'm not defending the use of money; I think it's questionable. It's still a home lab by my definition: it's computer stuff that costs less than a sports car in your home.
I purchased my SSD based on price and I have a AM4 APU so I'm limited to pci-e 3, but a quick check with crystaldiskmark says I'm hitting about 3.6 GB/s on sequential reads in seq1m q8t1 ... That would saturate 25gbps (assuming you have the right access patterns). seq1m q1t1 does 2.3 GB/s on my system, close enough to 25 gbps.
By my math, pci-e 5.0 should do 4x the throughput, so if you had 4 pci-e ssds you wanted to serve over the network and you had the pci-e lanes, you could saturate 400 gbps.
I'm pretty happy with my little baby network with a handful of 10G and 2.5G ports, and I'm cheap so if I made a big upgrade it'd probably be something used that has more 10g ports plus a 40G I could DAC to the adjacent home server ... but my home servers are barely powerful enough to push 10g anyway. (Something like a DCS-7050TX-48 goes for ~ $100 and would be a big upgrade in capacity I don't use over the S3300-28X I currently run)
(tl;dw - guy connects first four DGX and then another 4 "DGX-alikes". Uses 400gbps breakout cables (you can split interfaces, TIL) so he has 2 ports on the first switch, and then gets this one, I believe. You can buy cables that are 1x400 to 4x100gig, etc.
DGX Sparks have 200Gbps interlinks for clustering. Apparently switches are the only practical way to scale such clusters beyond 2 nodes even though each Spark has 2 such ports.
Well, the problem is that you want each node to be able to directly connect to each other, so you run into a mesh situation. https://www.youtube.com/watch?v=QJqKqxQR36Y shows 8 running at 100gbps and clustered.
I understand that part, what I don't quite get is why we can't divide the layers such that each node only needs to talk to at most 2 nodes. I assume I'm misunderstanding how work is distributed between them, but haven't looked into the details yet.
I think this is the distinction between pipeline parallelism (first N layers on machine 1, next N layers on machine 2, etc.) vs tensor parallelism where all machines contribute to the same matrix multiplication operation within the same layer. Pipeline parallelism doesn't need exotic high speed networking, but also doesn't deliver the kind of speed increase that tensor parallelism provides.
I mean at lot of what a homelab is experimenting. For example I have an absurd amount of communication 56GbE Mellanox switches but never really saturate it.
One of the problems with this kind of marketing is that there are essentially no base-models available at dealerships. So the flexibility supposedly provided by making the gauge cluster an option is not actually available to the consumer. This is just a money-grab while advertising a low price.
The ads all show the fully loaded model, "starting from" $LowPrice. We need a rule that the price shown is for the fully loaded model ($HighPrice), maybe fine print "configurable as low as $LowPrice". And then they have to actually allow you to configure that base model, and stock it. Otherwise, it's just effectively false advertising when every dealership gets just one base model that they will not sell to you.
It gets even more fun than that. As a datapoint my parent wanted a manual Honda Accord. In order to be able to advertise that price, Honda had to have a grand total of 3 manual Accords cars available in all of California.
Source -- Had a broker find the car and that was the story the broker told us (they found a manual but we had to travel from San Diego to San Dimas (2 hr drive) to get it.
I would like to agree, but that's not how insurance works, at least in the U.S.
You get your catalytic converter stolen, and you will pay for it three times. The first is you paid for the original one. Then you paid for insurance to cover its theft, and likely your payments over the years have already paid for the catalytic converter a second time. And then after it's stolen, your insurance rates will raise such that you are paying for a 3rd (maybe fourth) converter, just on a monthly plan now.
Now, of course, shooting the thief is only cheaper if you don't invest much in your weapons and training, and don't face any legal scrutiny afterwards and need to hire a lawyer.
It's not irrational if you know that the insurance company has mispriced your policy, and filing a claim won't be more trouble than it's worth.
Getting a contents policy in a safe neighborhood where you know that there's little risk is likely not a good deal. Getting a contents policy when you've learned that your neighbors are getting into hard drugs is probably a good deal.
Paying $19/mo for third-party full coverage + theft on your $1,000 mobile phone is probably not a good deal even if you know you are high risk, for the simple reason that at $19, they game must be that they will find any reason to refuse to pay, and will give you a poorly refurbished phone for your troubles if you make them honor the insurance.
You should find it insulting that they are likely selling that location data, not even to the highest bidder, but to a data broker that resells that data to anyone that will buy it for any price they'll pay.
Yes, this is standard practice in the app industry. I guarantee every app outside of f-droid that has location access is tracking you 24/7 and selling the log.
It could be a typical design that the camera and LED always have power, and the LED is switchable.
Giving the camera plus LED a separate power supply means that the camera has to boot or come online, which maybe increases the dwell time. And the camera is not visible on the USB bus when powered off.
I think there's more engineering to Apple's design than it first seems.
The thing visible on the bus is the interface chip, not the sensor. Most cameras which I've seen hardwire their activity LEDs (it is impossible to grab stills or record video without lighting up the LED) do so by connecting the activity LED to the sensor power rail. With the sensor powered off, the device is still visible on the bus but you get no usable image data.
I've found that some sources report lower UV index on cloudy days, whereas some seem to indicate the same index as a sunny day.
It's not clear if these sources all have a different model for UV, some being more sophisticated or conservative, and whether any of them are backed up by a UV meter.
reply