yvdriess

38 ms·

451.
by yvdriess 9y ago
> only to learn that the "Iris GPU" is not supported You can use the CPU backend, which has gotten significant optimization since ~v1.2. You can probably use the integrated GPU with the OpenCL backend, but I am not sure it wil
452.
by yvdriess 9y ago
HBM, HBM2, HBM3 etc. are industry standard and the result of precompetitive research. Intel's KNL has been using similar stacked memory (MCDRAM). NVIDIA's P100 has been using HBM2.
453.
by yvdriess 9y ago
The heat density of off-chip interconnects are low, but that does not mean non-zero power consumption.
454.
by yvdriess 9y ago
The common sentiment after the 90's is that it's hard to beat optimized x86 implementation of both. Maybe non-offloading FPGA-CPU hardware has the potential of accelerating certain language features (e.g. burn in the CLOS method l
455.
by yvdriess 9y ago
> What I can't look up is how this would work in the real world. Want safe drugs and devices? Who determines that? Somebody has done harm to another person. Who determines that? What I typically get as a response is that that person
456.
by yvdriess 9y ago
> In industry it does. Every major bank and hedge fund employs economists, and generally they have a pretty good grasp of what's going on. Of course those models are kept secret, much like trading algorithms. How useful are those mo
457.
by yvdriess 9y ago
It is almost as if reality has a left-wing bias.
458.
by yvdriess 9y ago
Indeed. Or to simplify: - At no point during the editing process do you have an invalid s-expression. - You manipulate the code datastructure, not the code text.
459.
by yvdriess 9y ago
This joke might be going to the core of the chief argument agains visual programming. After a while, you just don't see the matrix anymore.
460.
by yvdriess 9y ago
Google-image Ghost in the Shell cityscape: there is not a single ad hologram in the original (do correct me if I'm wrong though). The holograms shown are not as futuristic, they are green luminescent and pixelated, such as in the heli
461.
by yvdriess 10y ago
There is quite a gulf between inference and training, it is not just raw FLOPS. Inference is relatively simple data/control-flow, the weights/biases can be stored read-only and the entire network dataflow can be baked into the chi
462.
by yvdriess 10y ago
Answering with my own take on the subject. > What format would you propose? How is it different to instructions? It would still be instructions, but either by reworking what their operands are and introducing a destination (instruction a
463.
by yvdriess 10y ago
A benefit of introducing LISP in bioinformatics is the cultural element that tools are expected to work on the command line using simple linux streams for I/O. (As well as a systemic allergy to dependencies, preferring to do everything
464.
by yvdriess 10y ago
Titan-X is the current generation, DeepBench also shows results with Titan-X. The P100 has not been released.
465.
by yvdriess 10y ago
Intel has a Python distribution which bundles MKL as well: https://software.intel.com/en-us/intel-distribution-for-pyth...
466.
by yvdriess 10y ago
Do you have any citations to back this up? I am a user and not seeing any of this taking place.
467.
by yvdriess 10y ago
Note that ZigBee should have a lower attack surface than the Hue/LightLink extension. The regular ZigBee protocol requires a Coordinator in the mesh network, most of the LightLink protocol deals with doing away with that requirement.
468.
by yvdriess 10y ago
Do not conflate the american voting ID system with a national ID. One is an extra hurdle that opens the door for extra arbitrary barriers (pass this test before you can get an ID), the other is something everyone already has.
469.
by yvdriess 10y ago
But we already ordered a pallet of gizmo-3 cables, that new part only works with gizmo-4! Find a way to make it work.
470.
by yvdriess 10y ago
> classic clueless developer mentality Nice straw man argument. Pointing out that PHP is used for social/economic reasons is exactly saying that there are considerations outside technical merit. At the same time, Facebook is a big H
471.
by yvdriess 10y ago
> Facebook is written in PHP For social reasons, not technical ones. PHP is the language of cheap labour. To address the technical issues, Facebook's PHP code gets thoroughly translated in their backend.
472.
by yvdriess 10y ago
> It's hard to tell from the slides, but it looks like Zen is a much wider architecture than Intel, with 10 execution ports The text mentions it has to fuse the four FP ports to do a single 256-bit AVX per cycle. This is significant
473.
by yvdriess 10y ago
Actual neural network papers are far removed from the impressions you get from popular science articles. Biological neurons were some inspiration for the abstraction, but the underlying model is more a direct evolution of the maths/sta
474.
by yvdriess 10y ago
What is shown on the slides is not that innovating. If anything, they seem to have thrown the bulldozer innovations out of the window and moved the architecture closer to the sandybridge-era processors.
475.
by yvdriess 10y ago
There is a good body of research around 'Programming by Demonstration' or 'Programming by Example': https://en.wikipedia.org/wiki/Programming_by_demonstration From my own experience doing some rese
476.
by yvdriess 10y ago
Maybe you mean Dennard scaling ( https://en.wikipedia.org/wiki/Dennard_scaling ), Moore's Law is seen to still continue economically to 2021 according the the latest ITRS.
477.
by yvdriess 10y ago
The comedy here is that AMD and NVIDIA are adding it to CPU and GPU designs for the benefit of DNN training workloads; something that took off because of hardware designed for video games ;)
478.
by yvdriess 10y ago
The GP part stands for "something else than graphics". CUDA is not general purpose in the sense that a lot of types of computation and data movement results in terrible performance. You buy GPU accelerators for performance reasons
479.
by yvdriess 10y ago
Direct addressing is the preferred configuration. Only if your existing code's working set does not fit in MCDRAM does the cache configuration make sense. It might sound pedantic on my part, but 'it can act as cache' is very
480.
by yvdriess 10y ago
Nope. First, it's not L3 cache and secondly comparing 3D-stacked in-package memory (MCDRAM, HBM, HBM2) with your examples is misleading. https://en.wikipedia.org/wiki/High_Bandwidth_Memory

More ›