Hacker News: calaphos

New comment by calaphos in "The biggest heat pumps"

calaphos — Tue, 16 Dec 2025 09:35:20 +0000

> heat-pump equipment costs roughly €500,000 per megawatt of installed capacity

Interestingly enough the price for these giant heatpumps is pretty much in line with domestic ~10kw units.

New comment by calaphos in "TPUs vs. GPUs and why Google is positioned to win AI race in the long term"

calaphos — Thu, 27 Nov 2025 22:45:02 +0000

That is comparing an all to all switched Nvlink fabric to a 3D torus for TPUs. Those are completely different network topologies with different tradeoffs.

For example the currently very popular Mixture of Experts architectures require a lot of all to all traffic (for expert parallelism) which works a lot better on the switched NVlink fabric as opposed where it doesn't need to traverse multiple links in the torus.

New comment by calaphos in "Japan's gamble to turn island of Hokkaido into global chip hub"

calaphos — Mon, 24 Nov 2025 12:34:28 +0000

It's the one exception in the semiconductor supply chain where Europe is still leading. For all other parts of the value creation Europe is either a niche player at best or completely absent, well into the actual application layer.

New comment by calaphos in "The G in GPU is for Graphics damnit"

calaphos — Mon, 06 Oct 2025 10:24:59 +0000

The hardware is heavily optimized for low precision matrix math, pretty much only used for AI.

New comment by calaphos in "The great sameness: a comic on how AI makes us more alike"

calaphos — Fri, 26 Sep 2025 11:10:55 +0000

> to OpenAI's model gpt-4o-mini

Why a model specifically distilled down for logical reasoning tasks? I would expect larger models to produce a wider variety of outputs.

New comment by calaphos in "We may not like what we become if A.I. solves loneliness"

calaphos — Sat, 02 Aug 2025 19:52:53 +0000

But these social third places have also shifted. Younger generations aren't going out as much but e.g. playing video games specifically with other close friends is very popular.

New comment by calaphos in "Batch Mode in the Gemini API: Process More for Less"

calaphos — Fri, 11 Jul 2025 07:51:05 +0000

Inference throughout scales really well with larger batch sizes (at the cost of latency) due to rising arithmetic intensity and the fact that it's almost always memory BW limited.

New comment by calaphos in "I extracted the safety filters from Apple Intelligence models"

calaphos — Mon, 07 Jul 2025 09:26:36 +0000

If it's just filtered out in the training sets, adding the information as context should work out fine - after all this is exactly how o3, Gemini 2.5 and co deal with information that is newer than their training data cutoff.

New comment by calaphos in "Lieferando.de has captured 5.7% of restaurant related domain names"

calaphos — Mon, 26 May 2025 11:58:46 +0000

Something where you're reachable for any legal purposes- in Germany this sadly remains a physical address.

There are various service which offer a 'virtual' address with digital forwarding of letters for less than 10Eur/Month, so it's not an insurmountable obstacle.

New comment by calaphos in "'I paid for the whole GPU, I am going to use the whole GPU'"

calaphos — Wed, 07 May 2025 22:52:12 +0000

There's still a throughput/latency tradeoff curve, at least for any sort of interactive models.

One of the reasons why inference providers sell batch discounts.

New comment by calaphos in "Nvidia on NixOS WSL – Ollama up 24/7 on your gaming PC"

calaphos — Fri, 11 Apr 2025 11:08:01 +0000

There's also big efficiency increases when batching multiple requests, making clouds inherently more cost effective for normal use cases.

Way better utilization of expensive hardware as well ofc.

New comment by calaphos in "The Llama 4 herd"

calaphos — Sun, 06 Apr 2025 09:29:11 +0000

Mixture of experts involves some trained router components which routes to specific experts depending on the input, but without any terms enforcing load distribution this tends to collapse during training where most information gets routed to just one or two experts.

New comment by calaphos in "Steam Networks"

calaphos — Thu, 13 Mar 2025 20:01:21 +0000

It's not quite waste heat because the cold side of thermal power plants wants to be colder than district heating temperatures for best efficiency. There is some loss in electrical efficiency compared to non cogeneration plants, but the combined efficiency is a lot higher.

New comment by calaphos in "Harder Drive: hard drives we didn't want, or need (2022) [video]"

calaphos — Sat, 31 Aug 2024 16:40:47 +0000

Used to be a common thing for storing analog signals in the past :)

https://en.m.wikipedia.org/wiki/Delay-line_memory

New comment by calaphos in "Commercial perovskite solar modules at SNEC 2024 trade show"

calaphos — Mon, 17 Jun 2024 10:02:11 +0000

Apparently Chinese mainstream silicon PV modules are already a bit cheaper at ~0.14USD/W right now.

Article doesn't talk about efficiencies but it seems production perovskite modules are slightly lower than their silicon counterparts, which will affect downstream costs like land or mounting hardware a bit.

New comment by calaphos in "Cheyenne Super Computer Auction"

calaphos — Mon, 29 Apr 2024 15:00:18 +0000

Has been really common in HPC for quite a while. I presume the higher interconnect/network of hpc favour the higher density of liquid cooling. Hardware utilization is also higher compared to normal datacenters, so the additional efficiency vs air cooling is more useful.

New comment by calaphos in "EU science advisers back call for a 'CERN for AI' to aid research"

calaphos — Thu, 18 Apr 2024 14:00:02 +0000

Poland was dealing with similar brain drain problems, but now that economic opportunities are there educated people are returning.

New comment by calaphos in "Intel Gaudi 3 AI Accelerator"

calaphos — Tue, 09 Apr 2024 19:38:12 +0000

Intels surprisingly fast 14nm processors come to mind. Born of necessity as they couldn't get their 10 and later 7nm processes working for years. Despite that Intel managed to keep up in single core performance with newer 7nm AMD chips, although at a mich higher power draw.

New comment by calaphos in "Microsoft is looking at next-generation nuclear reactors"

calaphos — Wed, 13 Dec 2023 16:04:36 +0000

If you invest a lot of money into very expensive Nvidia training hardware you certainly want to run them as close to 24/7 as possible.

Dispatchable load usually means oversizing the dispatchable consumer to get the same overall output. This is already uneconomical for even particularly energy intensive industries (e.g. aluminum smelting). I would assume server hardware is a lot more capital intensive than that.

New comment by calaphos in "MLX: NumPy like framework for Apple Silicon by Apple"

calaphos — Wed, 06 Dec 2023 12:12:19 +0000

The API and featureset so far looks like a one to one reimplementation of JAX without the jit functionality.

What does this do differently? AFAIK Jax has an experimental apple GPU backend as well.