Working at IBM, opinions here are my own https://xitter.cf/t.co/lc6wkxMtwy https://xitter.cf/t.co/j1cEYw9LFF
Joined August 2011
- Tweets2.9K
- Following67
- Followers901
- Likes1.9K
I replicated @lemire 's MLP testing in OpenCL, and ran that experiment on a few GPUs: clamtech.org/?dest=gpumlp
Turned out to be a bit of a rabbit hole, and I hope yall find this interesting
Sharing some thoughts on Intel's possible unified core project. Basically, I think the easiest route is a Zen 4c/5c style shrink of their P-Core. But of course, Intel has more options than that
clamtech.org/?dest=intel_uni…
Sharing another piece I wrote last year, comparing hardware AV1 encoding on Intel's Arc B580 and AMD's Hawk Point, at clamtech.org/?dest=av1hwenc
Should be a good test of image handling on the site, with sliders for quality comparisons
Sharing a piece I wrote a while ago on Zen 1, mostly to test site design with plenty of pagination, tables, and captioned images. It's a pretty complete article by itself though, and I hope yall find it a fun read: clamtech.org/?dest=zen1
Posting another fun article on Chips and Cheese, this time on chipset latency: chipsandcheese.com/p/microbe…
Testing write bandwidth on GPUs: clamtech.org/?dest=gpuwrite
Intel's B580 does a lot of really interesting things, including apparently, doing write-back L1 caching in Vulkan. Nearly everyone else has L2 as the first write-back level
Using pointer dereferences to test GPU cache latency with less overhead, without assembly: clamtech.org/?dest=gpudirect…
Here's Gemini contradicting itself with consecutive sentences. Reminder that "AI" is good at arranging words into syntactically correct sentences but doesn't care about the meaning
Lenovo laptop keyboards are good, but they beep if type too fast.
Supposedly it beeps if you press multiple keys at the same time, which I'm not doing when I'm typing (not gaming). Feels like Lenovo keys take longer to de-activate after release than on other keyboards...
If DEI programs were causing unqualified pilot hires and more accidents, there sure isn't any correlation with reality.
We've had fewer air accidents over the past few years than 20-30 years ago, despite flying more (more aircraft-miles flown)
bts.gov/content/us-air-carri…
Cross-posting from bsky.app/profile/chlamchowde…
Zen 5's L3 can let you filter performance monitoring counters by core, for up to 16 cores per CCX. That doesn't necessarily mean we'll see 16 core CCX-es, but maybe someone at AMD is thinking about larger CCX-es.
I thought GPU atomics would only be used by compute applications. But one popped up in Baldur's Gate 3.
Interesting stuff. Games have been using more compute shaders. I guess they're starting to behave a teeny bit more like compute apps too.
Trump in the debate: "They took over a big percentage of the city of Seattle"
I lived down the street from that. Protesters occupied a few blocks, a tiny % of the city. Cops spamming tear gas (which got through window vents) was worse than the protesters, who didn't bother me
I'm glad my free time coding is being put to good use :P
Also I'm pretty sure Intel used my code for their HC2024 latency slide, based on the test point spacing. (github.com/clamchowder/Micro…)
Zen 5 only getting 4 instr/cycle from the decoders is likely not a "bug" or something that'll be microcode patched. The optimization guide straight up says max throughput from a single thread is 4 instr/cycle.
But it also suggests both decoders are in use in ST mode...
Intel's doc is a great read for those wondering how CPU makers predict chip lifespan. Higher temperatures and voltages can be used to accelerate aging
Of course extrapolating back to estimate lifespan at regular temps/voltages sounds like a hard problem.
intel.com/content/www/us/en/…
Always, always make sure Steam Cloud is disabled. Otherwise, it'll delete your progress if you load the game on another computer. Then Steam Cloud will overwrite save files on your primary system, erasing game progress forever.