rrss

46 ms·

421.
by rrss 6y ago
The original version of the NVPTX backend was contributed by nvidia, but iiuc they abandoned it afterwards and I think all the work on it since then has been by Google.
422.
by rrss 6y ago
AMD has a tiny market share in the cloud GPU market. AWS, Google Cloud, and Azure all have compute instances with NVIDIA GPUs, and I don't think any with AMD GPUs (IIRC Azure maybe has windows instances with AMD GPUs for graphics). Acc
423.
by rrss 6y ago
1. This isn't inference only, it has the full capabilities of a normal GPU, just small and low power (and therefore much slower than normal GPUs). 2. TPUv1 is a matrix multiply ASIC that requires a host CPU to do anything. This thing i
424.
by rrss 6y ago
The hardware and platform is capable of training just fine. It's just rarely done because it is slower than training on pretty much any discrete GPU.
425.
by rrss 6y ago
Neat, so come back in a few months. Also, I haven't got a clue who John Cardillo is, but that tweet doesn't really say much other than he thinks something is a scam. The 37k currently reported at the CDC link corresponds to the de
426.
by rrss 6y ago
... but if you just ctrl-F for TSX, you probably won't find it, because of this unique web viewer.
427.
by rrss 6y ago
> COVID-19 death counts shown here may differ from other published sources, as data currently are lagged by an average of 1–2 weeks. > *Data during this period are incomplete because of the lag in time between when the death occurred
428.
by rrss 6y ago
replying to your reply via edit: > _you_ assuming that the US is a linear relation to GB may or may not be accurate. Right, this is why I called it 'naive.' My main point stands: your comment misuses the prediction that does no
429.
by rrss 6y ago
The triggers considered were 0.2 deaths / 100k population / week and 1.6 deaths / 100k population / week. For the U.S. that corresponds to ~700 deaths / week and ~5,200 deaths / week. The United States as a who
430.
by rrss 6y ago
You are misrepresenting the conclusions of that report. The report clearing distinguishes between "mitigation" and "suppression" in the second paragraph. > However, the resulting mitigated epidemic would still likely
431.
by rrss 6y ago
Looked at the report from AMD (posted at https://news.ycombinator.com/item?id=23011643 ), and it looks like this is just an error in the article. Their computing and graphics segment is up 73% YoY. > Computing and Graphic
432.
by rrss 6y ago
> AMD's computing and graphics segment, which includes revenue for both consumer CPUs and GPUs, weighed in at $1.44 billion, a 73% year-over-year increase but 13% decline on the quarter > AMD's computing and graphics segment
433.
by rrss 6y ago
See also Chapter 29 of Hamming's "The Art of Doing Science and Engineering": http://worrydream.com/refs/Hamming-TheArtOfDoingScienceAndEn... You get what you measure.
434.
by rrss 6y ago
Did Aristarchus of Samos develop a model that made any predictions that could be verified? I think that the ancients thinking about heliocentrism were mostly contemplating it as philosophy. The section you linked doesn't say anything a
435.
by rrss 6y ago
I don't think the invisible 'lines of force' really resolved anything in the minds of the 19th century scientists, but what eventually did was acceptance of Faraday's speculation that the lines of force were physically
436.
by rrss 6y ago
Would you say it's accurate to say LIGO is a big interferometer measuring changes in lengths of the arms due to gravitational waves, with some extra analysis to figure why it still works even though the light waves get stretched? Or is
437.
by rrss 6y ago
Thanks!
438.
by rrss 6y ago
My point is that your statement: > Listen to the remarks, and its clear he never said anything about injecting disinfectant directly into a patient. does not seem to be correct. That's it. > You really think he was suggesting tha
439.
by rrss 6y ago
This is what you said: > it's clear he never said anything about injecting disinfectant directly into a patient Here's what Donald Trump said: > And then I see the disinfectant where it knocks it out in a minute, one minute
440.
by rrss 6y ago
Here's the briefing: https://www.youtube.com/watch?v=PsQnfpfIa_o . Can you please point out the timestamp when Trump attempted to clarify? The closest I found was at ~31:00, some reporter asks for confirmation that ther
441.
by rrss 6y ago
Ok, I listened to Bill Bryan's prior comments again, and I still do not understand what President Trump was saying if not suggesting using light and disinfectants inside the human body. Bill Bryan was obviously not using the word &quot
442.
by rrss 6y ago
Help me understand. "So, supposing we hit the body with a tremendous, whether its ultraviolet or just very powerful light, and I think you said, that hasn't been checked but you're gonna test it. And then I said, supposing it
443.
by rrss 6y ago
There are build config options, but there's also a kernel parameter > lockdown= [SECURITY] { integrity | confidentiality } https://www.kernel.org/doc/Documentation/admin-guide/kernel-...
444.
by rrss 6y ago
cuDNN converts to NHWC for 'tensor core' operations, so if the tensors are already in NHWC you can avoid the conversion. See https://docs.nvidia.com/deeplearning/sdk/cudnn-developer-gui... EDIT: I think
445.
by rrss 6y ago
Since this is controlled via a kernel parameter, can root just unset the kernel parameter and reboot? Or are there systems where kernel parameters are also 'locked down' by some other mechanism? (these are genuine questions - I do
446.
by rrss 7y ago
That's definitely true, and the debug symbols story is better on windows (in my experience) than other platforms - AFAIK linux has no real equivalent to symbol servers.
447.
by rrss 7y ago
with cuda or opencl, you can launch a kernel and get results back in something like 10 microseconds (for trivial 'results' like setting a flag or something). Varies by hardware, platform, etc, but that's the ballpark.
448.
by rrss 7y ago
Windows has really nice tracing and WPA is pretty powerful. I've not yet seen a tool that is similarly easy to use on Linux. You can get most of the raw data from perf, but the tooling for analyzing the data isn't as good. I'
449.
by rrss 7y ago
whoops, thanks for the reminder :)
450.
by rrss 7y ago
There are still other companies doing new core designs: Huawei/HiSilicon (Taishan v110 in 2019), Marvell/Cavium (Thunder X2 in 2018), Samsung (M4 in 2019), Fujitsu (A64FX is 2019), Nvidia (carmel in 2018). If you count semi-custom

More ›