HNHacker News
TopNewBestAskShowJobs

borramakot

109 karma · joined October 8, 2016

submissionscomments
borramakot··on The Feedback Fallacy
> Adding up all the inaccurate redness ratings—“gray,” “pretty gray,” “whitish gray,” “muddy brown,” and so on—and averaging them leads us further away both from learning anything reliable about the individuals’ personal experiences of the rose and from the actual truth of how red our rose really is.

I don't understand this comment. How does averaging noisy signal, even systematically noisy signal, result in something that is noisier than any individual signal? I would have assumed the average would converge on (real signal + systematic error).

borramakot··on Building Fast Fuzzers
Yeah, there was certainly a discontinuity when it went from "turn the tree into code" and "turn it into code in a faster language" to "build a virtual machine with random opcodes". I think it's just the nomenclature change that was jarring- characterising it as random switched dispatch was more intuitive.
borramakot··on AWS Data Exchange
How does a data provider prevent someone from copying the data from their S3 bucket into a new one, then cancelling the subscription and owning the data forever?
borramakot··on Americans Now Need at Least $500k a Year to Enter Top 1%
I get the sense that those numbers might include the value of stock increasing, but that's the only site that's had numbers within 10% of the offers I actually got after negotiation.
borramakot··on New chips for machine intelligence
Reportedly, it's an insanely large (whole wafer) single chip. But, then, why is it listed with a size that's not commensurate with that?
borramakot··on Amazon AWS Outage Shows Data in the Cloud Is Not Always Safe
Is that for S3, or EBS?
borramakot··on Hiring Is Broken?
I like the idea of multiple tracks a lot, but I think you need to be hiring at sufficient scale that each track has a fairly high rate of interviews, which makes it hard to support tracks that aren't commonly chosen.

If you don't had that scale, uncommon tracks will be full of interviewers who are out of practice for your question, and poorly calibrated to evaluate the candidate.

Also, some companies might not want to allow any of these methods, since they each have their own tradeoffs/blind spots.

borramakot··on Austin Is Building a Mini Silicon Valley
It wasn't literally a ban, but it was at least a serious impediment to their operating in the way they want to operate. The TX legislature overturned Austin's regulations with superseding, much more Uber/Lyft friendly regulations, and they are back now.
borramakot··on Amazon Elastic Inference – GPU-Powered Deep Learning Inference Acceleration
Strangely, the announcement that AWS is making their own machine learning inference chip got flagged and removed.

https://aws.amazon.com/machine-learning/inferentia/

borramakot··on Convincing engineers to join your team
That seems reasonable, but the compensation ranges I've been seeing aren't 0.9-1.1 x a reference salary, they seem to range from 0.3x (crappy consulting) -3x (HFT) the compensation of a mid-range FAANG position. With that kind of variation, it's very possible that compensation differences could overwhelm a normal sized difference in culture quality.
borramakot··on Efficient Methods and Hardware for Deep Learning [pdf]
Is there a way to do nonlinearity (ReLU) in the Winograd domain?
borramakot··on Efficient Methods and Hardware for Deep Learning [pdf]
I spent a little bit of time trying to understand exactly what DeePhi provides, but couldn't find any whitepapers or anything. Do you know any documentation on which of these approximations are supported by DeePhi?
borramakot··on What it means to “disagree and commit” and how I do it (2016)
There is a principle on development: Hire and Develop the Best Leaders raise the performance bar with every hire and promotion. They recognize exceptional talent, and willingly move them throughout the organization. Leaders develop leaders and take seriously their role in coaching others. We work on behalf of our people to invent mechanisms for development like Career Choice.
borramakot··on Ask HN: Would you prefer paying $1 per 1000 API calls or $30 fixed monthly?
This seems strictly better than the price/1000 calls, since at least you only have to predict the order of magnitude of usage.
borramakot··on Why I Quit Google to Work for Myself
It doesn't sound like they meant that senior was the last possible level, just the last level to which promotion in a career should be expected through "normal" growth and development. It looks like less than 1-2% of engineers at my company (FAANG) are above "Senior".
borramakot··on Benchmarking Google’s new TPUv2
What are the Amazon rumors?
borramakot··on Cloud TPUs in Beta
Back of the envelope, a TPU costs a little more than 2x as much as a Volta on AWS P3, and delivers a little less than 2x the performance (180 TOPs for the TPU, 100 for Volta). On a raw performance/$ metric, I'm not sure the TPU is that interesting.

It might be worth it if I were willing to pay a huge amount to get back results from an experiment faster, by using lots of TPUs- distributed learning on GPUs doesn't seem easy yet.

borramakot··on Virtex UltraScale+ FPGA Augmented with Co-Packaged – Community Forums
That's Catapult, right? I have read through some of their papers. It sounds like they might be offloading some network work onto the FPGAs in the same way AWS has their custom nitro card, but I've not really been impressed with their attempts at data processing improvements (some reason for me to use it on Azure). I haven't read all of their papers, but what I have read always sounded like an after the fact justification for the FPGAs. They might show the FPGAs are better at a machine learning task than CPUs, but unless you were deciding whether to use FPGAs you already have, the real competition is GPUs and they tended not to compare to a GPU.

Do you have a recommendation for a specific data processing experiment of theirs I should check out? I really feel like I just missed a paper where they proved some real advantage over other hardware, and once I found that I'd understand. I respect the Azure teams generally and assume they know what they're doing- but can't escape the hunch that what they are doing is network acceleration, and are just releasing to the public cloud because they have these sitting around anyway.

borramakot··on Virtex UltraScale+ FPGA Augmented with Co-Packaged – Community Forums
F1's on AWS are I think ultrascale+, though I don't know how fast Amazon actually lets you drive them.
borramakot··on Virtex UltraScale+ FPGA Augmented with Co-Packaged – Community Forums
I see the appeal here, but I'm surprised a good DSP or mid-volume, old node ASIC isn't the more common solution here. Do these need such specialized processing that FPGAs make economic sense?
borramakot··on Virtex UltraScale+ FPGA Augmented with Co-Packaged – Community Forums
Thanks for these!

The first paper is actually one I've spent a significant amount of time trying to use, to the point of collaborating with one of the authors. His conclusion was that FPGAs used to be competitive with GPUs for approximated nets, but the Tesla GPUs were such a jump forward in practical network performance that it wasn't worth trying to compete outside specialized realms like binary nets.

The second paper was interesting- I can imagine why the problem they are trying to solve would be a good fit for FPGAs. However, I'm suspicious that they implemented an entirely different algorithm on the FPGA, and didn't measure the performance of that algorithm on GPUs. I'm all for using the best algorithm for the hardware, but I worry they just used an overall better algorithm on FPGAs and conflated the results.

borramakot··on Virtex UltraScale+ FPGA Augmented with Co-Packaged – Community Forums
These all seem fair- I've been mostly looking at large scale FPGAs in a data center, a la F1 or Microsoft's Catapult. I hadn't given much thought to use as low-run hardware, at which I'm sure they excel.
borramakot··on Virtex UltraScale+ FPGA Augmented with Co-Packaged – Community Forums
I hear this a lot, but every time I try to implement a specific algorithm (in crypto, compression, and ML so far), I find that a GPU practically beats the FPGA on every metric but power, especially total cost. No matter how nicely the problem seems to map to an FPGA, GPUs start from such high performance that I can't seem to beat them- the one exception so far being some genomic algorithms.

Are there any really good papers, projects, or products that show where FPGAs provide a major commercial benefit over a GPU?

borramakot··on A General Neural Network Hardware Architecture on FPGA [pdf]
Are there performance numbers for this, on e.g. resnet-50?
borramakot··on Simulating RISC-V Clusters with FPGAs on AWS
To try this, do I need the full 16XL F1 instance, or can I run 1 node on the 2XL (1 FPGA) instance?
borramakot··on Intel Announces Movidius Myriad X VPU, Featuring ‘Neural Compute Engine’
I won't defend FP24, but FP16 is a widely supported IEEE standard.

https://en.wikipedia.org/wiki/IEEE_754

borramakot··on Peter Thiel To Join Trump Transition Team
The most clear statement from him I'm aware of is "The concept of global warming was created by and for the Chinese in order to make U.S. manufacturing non-competitive." 1:15 PM - 6 Nov 2012.
borramakot··on Ask HN: Why is FizzBuzz more important than actual experience?
Out of curiosity, I tried literally FizzBuzz on a couple of new grad candidates (my job is software, but we attract a lot of computer engineers). About half were able to do it without help. I no longer use FizzBuzz, but I'll use other very trivial questions to end an interview early if I suspect a resume has been significantly exaggerated.
← PreviousPage 2 of 2