lmeyerov

67 ms·

871.
by lmeyerov 5y ago
Maybe you are not familiar with how enterprise , and especially utility and gov systems work? It is often hard to not use RHEL due to compliance policies. IBMers deciding to swap in their race horse -- and simultaneously hobbling the curren
872.
by lmeyerov 5y ago
In theory yes, and we got tiny programs running this way, but could not afford to keep subsidizing megacorp politics. Nvidia punts to IBM RHEL8 docs for GPU podman, which is unusual and risky to see. We officially recommend against it for H
873.
by lmeyerov 5y ago
interactive graph ai + graph viz, such as for understanding it/sec/fraud/customer log activity
874.
by lmeyerov 5y ago
Yes, wsl2 afaict works like that and a bit improved. Instead of dual booting or a restricted dev env, I click my Linux terminal icon and jump into a full & fast Ubuntu, including UIs when helpful. Sharing with Windows is possible throug
875.
by lmeyerov 5y ago
Strong disagree. See above. IBM/RHEL seem to be the effective the stewards of Podman, and are using their monopoly-like position in enterprise OS segments to take control of the virtualization layer through it. This is similar but wors
876.
by lmeyerov 5y ago
In secure airgapped environments, very much so. We blew time setting up new offline install processes & tutorials for the Nvidia docker ecosystem for rhel8, which basically reused centos7, as most of our users took weeks/months whe
877.
by lmeyerov 5y ago
For the comparison to os x docker, the biggest slowdown for os x docker is the file system mapping. It basically breaks any reasonable performance chart: look up posts getting into alternatives like mutagen. The slowdown makes using an IO-h
878.
by lmeyerov 5y ago
podman has had repeat compatibility issues for us, and redhat has made docker installation stupidly hard in rhel8 at the policy level, which matters given the monopoly status of rhel in secure environments. It is hard for me to support the
879.
by lmeyerov 5y ago
Yep, docker in wsl2 is great, including GPU support wsl2 mode is super impressive -- none of the overhead of docker on os x
880.
by lmeyerov 5y ago
Disturbingly, their own products like azure dev portal largely push you into such a grant-universal-access-for-every-contributor path Their sec org is impressive, so these occasional massive things have been surprising
881.
by lmeyerov 5y ago
If I'm understanding right, this CPU vs. GPU bit seems apples-to-oranges for how GPU/parallel alg research would advocate optimizing this: - if you have one tiny network, no one cares, use pure python - if you have a bunch of tin
882.
by lmeyerov 5y ago
Fun to see negative downvotes here. People dislike that the dollar breaks the rules :)
883.
by lmeyerov 5y ago
Tried to find -- this was sometime 2020 / early 2021, and from a variety of sources (academic + gov), and pretty sure done by Freakonomics bc it was so weird. Interesting bc casts doubt on basically any expert wrt modern USD vs other c
884.
by lmeyerov 5y ago
Yep it is quite an arbitrage going on. Unless startup credits at the highest tier ($100K), way too expensive for most startups. For a lot of our customers, cloud is impossible now for HPC: shortages are so bad that you have to know someone
885.
by lmeyerov 5y ago
You'd think that, but data-driven economists were finding low correlations for this kind of (rational) thing specifically for the modern US dollar, which is confusing. Super great PlanetMoney/freakonomics episode on inflation ~las
886.
by lmeyerov 5y ago
Zero-sum bridge burning makes sense sometimes. But a career is rarely fully zero-sum. You likely need a reference, may have stock options dependent upon improvement, want a job offer X years later from a colleague making the new hotness, e
887.
by lmeyerov 5y ago
Leave for somewhere where you can have a strong mentor + supportive team. Not all teams are mature enough, skilled enough, and with enough allotted time to support typical junior staff: Most early career folks don't know how to figure
888.
by lmeyerov 5y ago
We've been getting increasingly involved in crypto investigation discussions, and largely: - money side is getting more anonymous, e.g., monero / tornado - ... in theory. Money crime still often using less anonymous schemes and
889.
by lmeyerov 5y ago
Sort of. A good question is what is the revenue per employee Meta is ~$2M/employee. Probably more like $1M once contractors are accounted for? But for a good person, even if the rest of the company has random issues, they can carve out
890.
by lmeyerov 5y ago
It wasn't "doesn't scale to 10-100X bigger" but "doesn't burn themselves and their teammates out 6mo from now and needlessly risk a stressful & prolonged sev0 when one of the 1-2 people who did weird things
891.
by lmeyerov 5y ago
Hard to kill because they're financially independent. We believe enough in the technology we're building (gpu-powered graph ai & visualization) and the missions we're powering (security, fraud, anti-misinfo, healthcare, e
892.
by lmeyerov 5y ago
Oof, just one of our customers pays about the same as their full revenue, and we're a cockroach team. The real thing here is they're not that surprising. A lot of companies with their kind of funding use enterprise sales to force
893.
by lmeyerov 5y ago
Part of why forcing gov employees out is hard is you have to go through hurdles like showing severe measurable underperforming, think lowest-quartile, which can be tricky, onerous, and little reward. That's why I quickly drift more tow
894.
by lmeyerov 5y ago
These programs are ultimately run by people.. So you can force them out (hard), or second-best, give their fiefdom to others and mar their record to prevent them for future big projects.
895.
by lmeyerov 5y ago
You don't do it as a specialized product company but do as a consultant of generic services where you compete by being friendly
896.
by lmeyerov 5y ago
Yes! :) Our visual graph AI tool includes a GPU-accelerated take on gephi's flows and puts on the web, including a free GPU-accelerated tier with no-code UIs, embedding & control APIs (python, js, react, arrow), and deep pydata int
897.
by lmeyerov 5y ago
There's a world of a difference of someone doing that for a Google scale thing with big teams supporting it, or say a core part of making the best wireguard product that only they can do, vs a system that could have been postgres and n
898.
by lmeyerov 5y ago
I enjoyed the post as there are invariably problems; seniority is learning how to solve more of it. Soft skills side becomes more important the higher you go and the bigger things get. The article is just one of many ways to manage up -- in
899.
by lmeyerov 5y ago
We have generally followed that for years, this is great! A few big missing areas: - node is also often the best for - 'backend for your frontend' ( batching/caching, ...) which adds a bit more complexity - structured logging
900.
by lmeyerov 5y ago
AWS+Azure (and I believe GCP) installed prev advances, and are having huge GPU shortages in general... so probably! An interesting angle here is these support partitioning even better than in the A100's. AFAICT, the cloud vendors are n

More ›