tntn

28 ms·

211.
by tntn 8y ago
> By having its own silicone Will this mistake never go away?
212.
by tntn 8y ago
Cache coherency is not the same as memory consistency models. This article is talking about cache coherence, and you are talking about memory consistency. As an example, both ARM and x86 have coherent caches, but neither guarantees sequenti
213.
by tntn 8y ago
Sounds like EE.
214.
by tntn 8y ago
Wow, you sure got them good. Thanks for rooting out the financial motivations of fast.ai, a nonprofit research lab that publishes courses online for free. /s
215.
by tntn 8y ago
> When I see the average driver competence on the roads, I don't see most of those people successfully commuting in a flying anything over the course of their lifetime. Ah! But AI, you see! They'll fly themselves!
216.
by tntn 8y ago
This is a BS cop out to avoid having to think about ethics. Some tech is inherently good or bad. If you create platforms that spread misinformation and rage much more effectively than information and reason, you shouldn't be able to wa
217.
by tntn 8y ago
Good magic system, but I found the ending to be disappointing. Literal deus ex machina.
218.
by tntn 8y ago
Only somewhat related,but I've tried to learn ocaml before but have never been able to figure out when to put what kind of delimiters where. Can anyone recommend a resource that explains the usage of the double comma and the like?
219.
by tntn 8y ago
One other point not mentioned in other comments: some work was presented at GTC regarding using tensor cores for a low precision solution followed by iterative refinement to a fp64-equivalent solution. IIRC, 2-4x speed up for fp64 dense sys
220.
by tntn 8y ago
> not of the good kind, like e.g. becoming a doctor I would estimate that around 90% of the people I know who want to become a doctor (around 30 people) are motivated solely by the desire to make lotsa money and be able to drive an expen
221.
by tntn 8y ago
it seems somewhat expected that someone might say a "sensitive word" in a meeting intended to discuss "sensitive words." Firing the guy for saying a word they were talking about seems strange - must that meeting be full
222.
by tntn 8y ago
Loads of questions here that seem to be based on only reading the abstract. Most of these are addressed in the paper.
223.
by tntn 8y ago
Why? When rent/mortgage is easily upwards $35k and taxes are upwards of another $30k, that doesn't leave the employer with a ridiculous amount of money.
224.
by tntn 8y ago
That post doesn't specify whether the training was fine with single or half precision floating point. Given that it's on the website of an AMD GPU cloud provider, I tend to think they probably used fp32 since it will make vega loo
225.
by tntn 8y ago
This idea of refusing to use a 3d imaging system because it's theoretically possible without is sort of bogus. No one refuses to use GPS because it's possible to use a map or sextant like humans can. No one refuses to use radar be
226.
by tntn 8y ago
Summit node CPUs can access GPU memory coherently and unified memory allows for a single pointer across all processors. On most systems that involves page faults migrating pages, but summit has something called ATS that allows GPU to direct
227.
by tntn 8y ago
RHEL 7.4. https://www.olcf.ornl.gov/for-users/system-user-guides/summi...
228.
by tntn 8y ago
TPUv3 pods have a stated performance of 100 peta "flops," which is half precision operations on hardware designed specifically for machine learning. The 200 petaflops number of Summit is referring to general purpose double precisi
229.
by tntn 8y ago
As a point of interest, there has been some cool work done recently regarding accelerating high precision dense system solvers using tensor cores + iterative refinement. See http://on-demand.gputechconf.com/gtc/2018&#x2
230.
ORNL Launches Summit Supercomputer(hpcwire.com)
2 points by tntn 8y ago|0 comments
231.
by tntn 8y ago
I sort of understand this sentiment, but Summit in particular is fascinating technologically. Half as many nodes as Titan, way higher power efficiency, NVLink2 interconnect between CPU and GPU, ATS unified memory between all GPUs plus CPU,
232.
by tntn 8y ago
No, it's not. Google's data centers would probably perform like crap on the kind of workloads run on supercomputers.
233.
by tntn 8y ago
For some context, the U.S. foremost supercomputer up to now was Titan, which had a theoretical peak of 27 petaflops at 8.2 MW. Summit has significant higher peak performance at slightly higher power. It also has significantly fewer nodes, w
234.
by tntn 8y ago
Like it or not, it is quite a bit of a competition.
235.
by tntn 8y ago
What? Summit had been in development for years, and is ridiculously efficient for a machine with so much performance. It's a major milestone on the path to exascale computing.
236.
by tntn 8y ago
It's not inherently wrecked, but it's often wrecked anyway. For example, the BART extension to south Fremont cost around $150 million per mile and took 3 years longer than planned. CA can't seem to maintain roads, and when it
237.
by tntn 8y ago
Whether or not it's obvious in the map, it seems to be true. > By race/ethnicity, non-Hispanic whites and American Indian/Alaska Natives had the highest rates of suicide, with rates for both groups showing notable increase
238.
by tntn 8y ago
> Not that pre-gps long range missiles are any better. They were much, much better. EDIT: It appears the V2 in testing had a 4.5km CEP, while he minuteman III has a 200 m CEP.
239.
by tntn 8y ago
The typical number for German combatants is closer to 50k, so I'm curious what your source is for 5k.
240.
by tntn 8y ago
The price may have been symbolic for universities, but AT&T licensed Unix commercially for many tens of thousands of dollars.

More ›