varelse

124 ms·

691.
by varelse 11y ago
It's a pretty big leap from lots of little parallel (and I assume 1D) FFTs to a deep learning processor. The rexcomputing chip looks like a DMA-driven systolic array with beefier data-processing units than I've ever seen before.
692.
by varelse 11y ago
Bad compilers would be my guess. And not only would it make sense for Intel to push OpenCL for Deep Learning, but it would make sense (IMO of course) for them to push OpenCL as a much better abstraction for accessing the vector units in al
693.
by varelse 11y ago
One more thing I wanted to add. As someone who often evaluates new processor architectures, n-body is anything but a real application. It's a nice demo (pretty too), but real applications (even those that are similar to n-body) have
694.
by varelse 11y ago
I would love to see AMD jump in the ring, and there's even an OpenCL port of Caffe in progress: https://github.com/BVLC/caffe/pull/2610 But its performance is less than half that of a GTX 980 running CUD
695.
by varelse 11y ago
"As for "getting my facts straight" I was using actual Titan X numbers I have seen for real applications (e.g. nbody simulation)... it only gets around 4000GFLOPs single precision compared to their advertised 7000." Nons
696.
by varelse 11y ago
Yes, NVIDIA owns Deep Learning. And they will continue to own it for at least the next 2 years. But I would be remiss not to point out that you can build one of these things yourself for less than half the price at which they're sell
697.
by varelse 11y ago
So basically you're shamelessly promoting a powerpoint processor versus shipping hardware? Pot.kettle.tapeout. I thought that was Intel's specialty. It's a very long road to a shipping chip and a lot can happen between now
698.
by varelse 11y ago
As someone who worked at Google when Google+ was released, IMO the failure was all about Vic Gundotra. 1. He's the elitist who went against Google's culture and separated the Google+ team in its own sooper sekrit headquarters with
699.
by varelse 11y ago
I'd really love to see speed measured by performance by a single collective computation of an O(n) or O(n log n) algorithm. This would emphasize the importance of balancing communication performance with computation. Not holding my b
700.
by varelse 11y ago
Does this answer your question? https://en.wikipedia.org/wiki/Convolution_theorem
701.
by varelse 11y ago
The bigger the convolution, the faster it gets because a convolution in real space is a multiplication in frequency space. It breaks even at 5x5 or so and gets dramatically better shortly thereafter. However, most of the convolutional nets
702.
by varelse 11y ago
"CPUs are super restricted by the single threaded, branching nature of the code you run on them, and this is what makes CPU performance a little more nuanced, and not directly comparable." Ironically, to really hit peak performanc
703.
by varelse 11y ago
Yes, it is true conceptually, but: 1) "Who cares" is not the sort of thing you want to say to someone who cares about performance because: 2) Sending the data to buffers for subsequent reduction was the first implementation (2009)
704.
by varelse 11y ago
That's not a problem. The application in question was a redesign of an MPI FORTRAN implementation that can produce the same outputs to within FPRE. To verify the low-level code, we just run in double-precision and make sure it matche
705.
by varelse 11y ago
There is a very active attempt to make GPUs FP-friendly called C++AMP. And for streaming transformation tasks, it's a nice abstraction. https://en.wikipedia.org/wiki/C%2B%2B_AMP However, my best algorithms are st
706.
by varelse 11y ago
I'm self-taught and while I can see the formal appeal of functional programming, I'm a low-level coder that implicitly reduces state to bare-metal registers and cache in GPUs and embedded systems. Since I'm already sweating
707.
by varelse 11y ago
GTX 580: 585M transistors GTX 680: 3.5B transistors GTX Titan: 7.1B transistors GTX 980: 5.2B transistors GTX Titan X: 8B transistors Core i7-5930k: 2.6B transistors What the data above suggests to me is that relying solely on Moore's
708.
by varelse 11y ago
2011: GTX 580, 1.5 TFLOPS 2012: GTX 680, 3.0 TFLOPS (~2.0 attainable) 2013: GTX Titan, 4.4 TFLOPS (~3.2 attainable) 2014: GTX 980, 4.6 TFLOPS 2015: GTX Titan X, 6.7 TFLOPS Looks to me like they're doubling perf roughly every 2 years. M
709.
by varelse 11y ago
If Seattle implements rent control, it will be the mother of all property investment opportunities as it turns into SF 2.0 over the next decade. Long-term renters will dance in the streets alongside property owners and everyone else will b
710.
by varelse 11y ago
I find that in the age of the open office, working remotely is the only time I'm really productive. I didn't mind working from an 8x8 cube, but my current tiny desk in a maze of desks, all alike, is the pits (cue some nimrod post
711.
by varelse 11y ago
I consulted for a couple weeks at EA about 10 years ago. On one of those days, I got 2 VPs and a director into a half-hour talk in a break area about how much they loved the game Starflight: https://en.wikipedia.org/wiki&#x
712.
by varelse 11y ago
I have PC desktops from 2009 that are still functional. In contrast, I have gone through three printers since then and in fact I no longer own one. See also mobile phones/tablets with annoying 18-24 month lifespans before they develo
713.
by varelse 11y ago
And there are a lot of cases where it doesn't work, specifically with elaborate MPI scenarios and over a network/VPN. Specifically, I do not wish to jump through hoops to enable remote profiling over heavily IT-restricted network
714.
by varelse 11y ago
Isn't it great that we can outsource our toxic e-waste? http://www.theguardian.com/global-development/2013/dec/14/to... But besides the 800 lb gorilla in the room, what worries me more than planned
715.
by varelse 11y ago
Lots of good stuff, and some not so good stuff: "Deprecated: legacy (environment variable-based) command-line profiler. Use the more capable nvprof command-line profiler instead." IMO there are way too many use cases for the comma
716.
by varelse 11y ago
So basically these guys? https://en.wikipedia.org/wiki/Keynote_DeviceAnywhere
717.
by varelse 11y ago
A) I am not a (major) outlier (or many of my friends are major outliers too). Bottom line: Senior Engineers can make $250K+ today. Burn that into your synapses. The Netflix listings already posted here demonstrate that. I know Facebooke
718.
by varelse 11y ago
Really? My starting salary as a level 1 engineer in the first dotcom boom was $115K. Now had I taken a more fun (but as it turns out in reality frequently not very fun) job in the game industry, that salary would have been ~$70K. This of
719.
by varelse 11y ago
No, they're not. Google paid me well to be the equivalent of an NCG python script kiddie. I'm the one that got bored and failed to convince them they ought to be making using of the skills that made them notice me in the first p
720.
by varelse 11y ago
I wouldn't even call this inflated in the valley anymore. There are outliers making serious money right here, right now. Such a salary pairs nicely with our lovely $1M+ median house price to insure we can never afford to buy one with

More ›