"Based on these public on-demand quoted prices from AWS and IDC, we found that the IntelR GaudiR 2 has the best training performance-per-dollar, with an average advantage of 4.8x vs the NVIDIA A100-80GB, 4.2x vs. the NVIDIA A100-40GB, and 5.19x vs. the NVIDIA H100"
Adhawk, adhawk.io, has the only all day ultralight eye tracking wearable I'm aware of, all MEMS based with ultra high scan rates, 500Hz+ and research grade accuracy. For ALS u likely need something light and frictionless, wearing a hot and heavy headset all day probably doesn't work.
You'd think the cost on the largest piece of debt a family hasdoubling would have a negative effect, but not in Canada to date, extend and pretend is the name of the game. The government seems set to prevent any reckoning by instructing and monitoring the banks to lessen the impact via the body that regulates them. A week ago the Financial Consumer Agency instructed banks to take steps to assist mortgage holders "The guideline sets out how FCAC expects federally regulated financial institutions to provide tailored support to consumers with mortgages on their principal residence who are experiencing severe financial difficulty."
- negative amortization
- not charging interest on any interest not being covered
- not charging too high of rates on fixed term renewals (not sure how this one works given the cost of capital on a 5yr terms is ~ 5-5.5% -> force the banks to loose money on renewals I guess?)
So maybe the strain will slow and it will shift to non regulated debt like credit cards. It will be harder to borrow for first time buyers as banks pull back. Lots of unintended consequences I'm sure.
I imagine the hope is you can pretend rates are actually not at a 20yr high by forcing the banks to not pass on the costs, and hope they go back down again the old "money is free" days. Problem is the cost of housing staying artificially high and not letting property values comes down is just forcing the housing input to inflation to be high giving the BOC tough choices. The market is seriously fucked up, a reasonable family home in a 2nd tier city is 1M which is a cost (opportunity or real) of 130k / yr (after tax) - so better hope your family income is 200-250k. If you're 20-30 forget about it, borrow money from the bank of mom and dad.
On the plus side making 5% on a savings account is now real money -> the risk free return hasn't been this high for a long time (20 yrs?). Either rates need to come down or asset prices do, but I agree that eventually something has to give.
Laptop running what I assume is a close to final sample of the upcoming meteor lake APU on Intel 4, if you zoom in you can see it hitting ~1 it/s on stable diffusion at 512x512 on a 20 step run. I'm sure they still have work to do as the CPU seemed to running slow at ~3Ghz / drivers / maybe not the final stepping and probably a custom SD model. Based on random reddit threads this seems close to what an M2 Pro would do (after apple optimized things for SD - i think twice as good before they made those changes), similar to a 1060 / 1070. No idea how the 7940hs/780m performs on SD as that's the main competitor. Decent indication that Intel is has the thing ready for release this summer and that perf is going to be pretty good.
Woke up this morning to see YouTube had flagged one of our very boring tech demo videos for harassment. It had all the slanderous ingredients: API calls, a spreadsheet and ... Salesforce. Of course our kind "review" by YT was done in seconds - so efficient YT! and came back as BANNED. Some of my thoughts as a 4 time founder on how this isn't really unexpected and mostly okay - but really fricking annoying.
it doesn't use 150w more, if you want to be concise be concise about a facet, talk about the workload and the power consumption per workload, most computers sit idle and as commenters are pointing out AMD has 2x the power draw at idle, amd doesn't seem to respect the pl1 power limits per Ars and is less power efficient for those who care about power efficiency, run the pl1 at 125W and intel burns less and performs the same for less power draw while working, on a 2 node disadvantage. And being concise about the cores requires you to talk about the core types, 2/3 of the 24 cores are little cores, one could say intel is performing the same with 1/2 the equivalent big cores, 8 vs 16 on the top part. It's more nuanced.
check out https://arstechnica.com/gadgets/2022/10/intel-i9-13900k-and-... - they power limited both 7000 and raptor lake models and surprisingly, amd doesn't respect the power limits - at 65W it's burning 90W and intel is respecting the 65W limit, 7.868Whr vs 8.58Whr to complete the same task (h264 encoding in handbrake), so interestingly intel on a 2 node disadvantage is more power efficient if you set your power limits.
Based on a couple of the reviews it sounds like Intel is doing really well on a very old node and more or less beating AMDs part for part comparison by adding more little cores and better cache for ipc and MT gains. Power draw is better on this refresh but worse than AMDs 5nm chips, but one review showed idle power draw on the amd 7000 series at ~100w vs ~60W on the intel system at the plug which is concerning and a couple reviewers are yet to test intels claim that they can match last gen perf at ~1/4 the power draw which might actually give intel the perf/watt crown. REnder tests I saw measuring KJ drawed per task did show AMD about 15% more efficient BUT at single core the draws are the same so depends on how hard you're hammering and the workloads. I think Intel has the huge cost advantage due to cheaper MBs, memory and sounds like AMD 7000 series sales are hurting due to the large platform cost. I think Intels meteor lake parts will be the most interesting, if Intel can actually execute on time we'll see a shop who clearly can execute on an old node hopefully surprise on a modern node.
If anyone is interested I setup the bills and committee endpoints in a google sheet, makes it easy to get the data in one spot, manage pagination, etc.. You can simply copy the spreadsheet and it will copy the api connectors, just enter your api key, you can easily add other endpoints, change query params, etc.
Pagination seems to be limited to 250 on the endpoints and some of the data sets are quite large so I set a limit on pagination of 10 pages, if you want all the bills from all time, prepare to wait.
Full disclosure it uses a free addon myself and my cofounders created to easily push any api data into a google sheet. I just find it useful to play around in a spreadsheet very quickly vs setting up a db.