Hacker Newsnew | past | comments | ask | show | jobs | submit | fidotron's commentslogin

The performance here is less interesting than the apparently quasi imminent availability of real RVA23 devices. If/when that occurs it will set a baseline and things should not warrant from scratch rebuilds constantly for at least a while.

Anyone that believes semiconductor performance claims before device availability is in for a rough time.


Manufacturers don't lie about how many clock cycles Dhrystone or Coremark or SPEC takes on their design.

The first two can be checked at the verilator (or commercial equivalent) stage, and SPEC in FPGA, both long before physical chips.

The only things in doubt are:

- whether they hit the hoped-for MHz

- whether the code you want to run has the same characteristics as their benchmark

For example SG2044 GeekBench results have been appearing on the site from time to time for more than half a year now. GB as a whole is pretty much worthless for what I care about, but the Clang test within it aligns closely with what I do, including the rqtio between multi-thread and single-thread.


You are right, but I think this is different than say performance claims for the next generation by mainstream GPU or CPU manufacturers.

This is a chicken and the egg problem. You need interest in the architecture and belief in performance for device availability to start to pop, after all who would invest a lot of money in something they don't think will have a market, but you also need device availability to support performance claims.

So I think for newer products, performance claims will have to do.


Yeah, but some of us like pf and systemd. It's so confusing.

Oh come now, no one likes systemd!

I kid. Sort of. Systemd is a great startup launcher -- maybe even an improvement over rc.d. Although I confess I've never had any issues with rc.d. Systemd can be quicker, which is nice, but 15 seconds saved on a server that reboots very infrequently is not super interesting. More importantly, the rest of systemd is less compelling.


Now a comprehensive book on systemd would be really nice. Most of the stuff out there is focused on service files, but there's a LOT of other stuff in the systemd ecosystem that doesn't have much easily accessible content available.

I think I agree that there’s not a _ton_ of content out there on 80%+ of the systemd featureset.

However, the first-party docs are really quite good. Considering that some blog posts are written to compensate for a lack of docs, I wonder if there’s a causal relation there.


Yeah, another side effect is management types are now allergic to things which look like maker projects, even if done with a level of professional engineering seriousness - they are unable to distinguish between the two, so now they dismiss both.

This has been a factor in the slowdown of commercial IoT, as it is often dismissed as science fair stuff.


Yep, or the old GreenArrays GA144 or even maybe XMOS with more compiler magic.

One of the big questions here is how quickly it can switch between graphs, or if that will be like a context switch from hell. In an embedded context that's likely to become a headache way too fast, so the idea of a magic compiler fixing it so you don't have to know what it's doing sounds like a fantasy honestly.


Yep, that’s definitely the question. The article says that there are caches of recently used graphs for use in large loops. Presumably those are pretty fast to swap, but I have to imagine programming a whole new graph in isn’t fast. But maybe the E2 or E3 will have the ability to reprogram partial graphs with good AOT dataflow analysis.

Also, how would cycle-accurate assembly look like for this chip?


In fairness to Air Canada a few months ago they pleasantly surprised me. Admittedly long haul international, which does not seem to be as extremely bad as their domestic. Pearson is clearly a disaster area at this point, as that whole gold theft demonstrated.

Yeah, I've been travelling around Canada and the US for two years now for work (~30 trips) and Air Canada had a dramatic improvement somewhere around Spring/Summer 2024. From November 2023 to around April 2024 I was keeping track of the probability of actually having my checked bags arrive at the destination and they were sitting around 60%. Since around Spring 2024 it improved dramatically, plus they started offering (mediocre) beer for free in economy!

All of the "bad" examples look like they're playing on a PC with poorly set gamma curves. Play on a TV where the curves are setup properly because TV people actually care about color reproduction.

Some of us have been pointing out Intel was in a systematically impossible situation even back when they had that process advantage, now almost a decade ago.

Quite simply imagine being dropped in as CEO of Intel in 2015. Could you have prevented the malaise of today?


A fine time to cancel Larabee properly and get serious about specialized GPU hardware five years earlier.

This problem is how you spot people that have tried to do it vs those that just talk about it. Everyone ends up with batching calls back and forth because the cost is so high.

Separately the conceptual mismatch when the js has to allocate/deallocate things on the wasm side is also tedious to deal with.


Those are distinct actions though.

"Anticipation" in character animations is a source of persistent friction between game programmers and animators because if incorporated as part of user/player action it simply kills responsiveness. OTOH being able to see NPCs prepare to do something (like pulling the bow) is incredibly useful.


Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: