HNHacker News
TopNewBestAskShowJobs

trsohmers

1,536 karma · joined March 14, 2011

My website is trsohmers.com

2013 Thiel Fellow

2023-Present Positron AI (https://positron.ai)

2021-2023 Groq (https://groq.com)

2019-2020 Lambda Labs (http://lambdalabs.com); Managing hardware development and cloud/datacenter deployment

2013-2018 REX Computing(http://rexcomputing.com); Developed new processor architecture from scratch

@trsohmers

Follow me on twitter... @trsohmers

submissionscomments
trsohmers··on UBeam's Meredith Perry shows her wireless charging technology really works
I wasn't saying it wa snot a factor, and I totally agree with you, but the inverse square law per its exact definition does not apply here.

I think uBeam is another Theranos, but granted at a much smaller scale. Still can't believe it has raised so much money and so much press without basic questions really being answered.

trsohmers··on UBeam's Meredith Perry shows her wireless charging technology really works
Not to be defending/advocating that uBeam's tech is actually any good, but the inverse square law refers to omnidirectional transmission, while uBeam's claims are related to directional transmission of power.
trsohmers··on Samsung Foundry unveiled aggressive roadmap scaling down to 4nm
It is all marketing bullshit at this point. TSMC's 12nm process is a rebranded version of their 16FFC low power flavor, with literally no other changes other than name so that they can market as being smaller than Samsung/Global/Intel's 14nm processes. All of the foundries are just naing the newest node roughly in line with the original ITRS roadmap names, even though the smallest feature sizes on the most advanced node in mass production at the moment (Intel's 14nm, which is a good 20%-30% smaller feature sizes than Samsung's "10nm") is around 42nm (source: http://images.anandtech.com/doci/8367/14nmFeatureSize.png)
trsohmers··on Patent Application: Full body teleportation system (2006)
I definetly think the latter, and I'd ditch my body for a newer one in 30 years if I could. Would be pretty great if a transporter also did "tune up" of all of the frail biological bits as part of the copying process (without messing with the brain hopefully)

Of course, I think that this is extremely impractical so it isn't a decision that I actually have to face. In this hypothetical scenario though, I'd much rather use the "transporter" as a cloning machine... having two of me would be pretty awesome.

trsohmers··on Patent Application: Full body teleportation system (2006)
I really don't understand why people get hung up on this (with the assumption of the impossible guarantee that it is 100% safe, or "bullet proof" as you put it). I am perfectly fine with a Prestige/Star Trek transporter, as long as it is a perfect copy, I don't really care if it is the "original" meatsack or the 100th... all I would want is to make sure the original dies quickly and painlessly (preferably after it is confirmed the copy was transported safely).
trsohmers··on UploadVR sued over ‘rampant’ sexual behavior in workplace, wrongful termination
I feel I have to defend myself here to some extent (being a now former Thiel Fellow), but I can fairly safely say that the ~20 other Thiel Fellows that I would call "friends" would never do anything close to this at their companies. While there are now over 120 fellows and I do not know them all, I bet there is one or at most a small handful that could fit your description, but I would say that of any group of similar size that you were to pick from a university.

I would definitely say I am socially awkward/inept, but I don't think that necessarily correlates with being socially maladapted/unaware of societies norms. The sort of "bro" culture that exemplifies UploadVR's, Uber, and many other "hot" tech companies right now is the culture that myself and most other high-school nerds that I knew rejected... we were not the popular kids that drank, had sex, and partied. I think all of this is a symptom of technology being a much more attractive jobspace for the general public (which includes a lot of "bros") compared to just for "nerds" that it was decades ago, and your conclusion is the exact opposite of this.

Granted it is an anecdote with a relatively small sample size, but I have not seen this sort of behavior/ideas from anyone who never went to college or dropped out, but I see it from plenty of folks in their late 20s/early 30s that never grew up from their college fraternity experiences.

trsohmers··on Intel releases the last Itanium chip, the 9700
The assumption by us is that you would be recompiling from at least the LLVM IR level for each different version of the chip. This isn't a real problem for the markets we care about, as they have the source code for all of their applications. In the case of not wanting to distribute your source code, I am interested in being able to distribute emcrypted LLVM IR as a pseudo-binary, similar to how silicon IP companies deliver encrypted verilog and vhdl for integration into other chips.
trsohmers··on Intel releases the last Itanium chip, the 9700
It isn't something we have been looking at, as our current customers and target applications would have our chip as a standalone embedded processor. We are working on a prototype PCIe card with an FPGA to be able to convert our SerDes to something a PC can talk to, but it isn't really a priority unless we have someone come to us directly asking for it.

As for upstreaming, it is something that we want to do, but are not sure if it will be accepted as our chip is not generally available at the moment. It would probably be done around the same time as our next tape out (which would be in larger quantity than our first one) when we would have generally available development kits this coming winter.

I have been working on public documentation in my spare time... originally supposed to go out in March but we have had other priorities. If you watch my Stanford talk I show the basic ISA overview as well as the WIP documentation made using doxygen. I'm hoping to get stuff out on our website before our next media push in ~June.

trsohmers··on Intel releases the last Itanium chip, the 9700
LLVM is much more compartimentalized, so we can make changes to the backend without touching/breaking other parts. The GCC loader was miles ahead of the LLVM loader up until the past few releases, and we were originally using a modified version of the GCC loader before switching.

LLVM also has a bunch of awesome projects like Poly for polyhedral compilation that we are able to use that don't exist at all for GCC, or are very difficult to integrate. LLVMs lego block like architecture (it was originally designed as a tool to help build toolchains) plus things like Tablegen enabled our single "real" software engineer/CTO to do initial bring up of a backend in under a month.

trsohmers··on Intel releases the last Itanium chip, the 9700
We have played around with having "hints" in the original source code that will be preserved through the IR, but the vast majority of the optimizations and scheduling decisions we need to do are happening once it has been converted into machine code by the backend. I may be slightly off (I'm the dumb CEO/hardware architect), but I'm fairly sure 85%-90% of our compiler work has been done on the backend... I just remember being told some tweaks were made to clang due to some weird issue.
trsohmers··on Intel releases the last Itanium chip, the 9700
The two best application cases of instruction level parallelism that we have at the moment are our FFT and DGEMM, where our 64 bit instruction word contains the operations for doing a double precision fused multiply add, two 128 bit load/stores, and an ALU op for address calculation. The inner loops of these programs are able to get up to 4 instructions per cycle (5 if you count the FMA as 2 ops, which is what everyone else does but I think is cheating), and compared to a superscalar architecture, uses very little space in local memory (fitting 4 operations in only 64 bits is awesome!). Compared to the latest 14/16nm offerings from Intel and NVIDIA, we have a 25x energy efficiency advantage for our FFT while our test chip is still on 28nm.

These are the same reasons why DSPs have been VLIW since the 90s, and our initial applications overlap a lot. It has been sad to see the trend of TI and other DSP makers going down the path of adding hardware caching (that reduces determinism and energy efficiency) to try to "make it easier for programmers" when they already have a decent enough architecture, but a lack of good compiler tools to take advantage of it.

trsohmers··on Intel releases the last Itanium chip, the 9700
LLVM is the godsend that any new architecture (especially a VLIW) to compete with Intel needed. I'm the founder of a startup that has made a new VLIW processor with our toolchain built on top of LLVM... We get the front end and all of the languages it supports virtually for free (we've only had to make small tweaks to clang), and get to focus our effort on just the parts important to us, our actual target backend and the unique optimization steps. Right now we can take any code that compiles through LLVM and run it on a single one of our cores while finding a good chunk of ILP (we are quad issue, averaging close to 2.5 IPC, sometimes higher), and be beating out the performance of some ARM and DSP cores that are 10x the size. Multi core parallelism requires including our hand tuned libraries, but our true MIMD style of task and data level parallelism is a lot more flexible than SIMD/SIMT/SMP that you are stuck with on existing chips.
trsohmers··on Intel releases the last Itanium chip, the 9700
Designer of the Neo here, and owner of the California "VLIW" license plate. As you can guess, I am a die hard VLIW advocate, and a strong believer that the original promises of VLIW (Drastically simpler decode logic, implicit instruction level parallelism, virtually no control/data hazards on chip).

VLIW has gotten an extremely bad rap outside of the embedded space due to Itanium, which I strongly contend was not a VLIW in spirit (which you cans see me talk in more depth on here: https://youtu.be/ki6jVXZM2XU?t=441). Itanium introduced a ton of indeterminism by having static scheduling unfriendly things like branch prediction, variable latency caches, and still trying to have some level of support for x86. The core problem for Itanium is that it was impossible for the compiler to make good decisions in the vast majority of applications since the compiler could not know exactly where and when things in memory would be. I believe the underlying failure point has been computer architects for failed VLIW architectures not realizing that fancy bells and whistles and "cleverly complex" hardware designs is the exact opposite of what you want when the compiler needs to be able to do purely static allocation. Adding fancy dynamic pieces in hardware to account from this (seen in a lot of places with Itanium) just makes the problem worse.

The main improvement we have made with the Neo architecture is we have hard (exact cycle count) guarantees on all memory movements on and off the chip, which actually gives the compiler the information it needs to make good decisions and very condensed code. I think the jury is out on the applicability for "general purpose" code on VLIWs... it is not something we at REX really care about at the moment, and I think RISC-V is a great solution with improvements over x86 and ARM.

trsohmers··on How enthusiasts designed a powerful desktop PC with an ARM processor
Pretty sure he is thinking of Elbrus, but AFAIK nothing actually happened from that being announced back in 2004... Elbrus still exists and is making chips. They were basically the Russian Transmeta.
trsohmers··on Debian port for RISC-V 64-bit
Hack the Planet!
trsohmers··on A Look inside Russian 28nm MIPS CPU – Baikal-T1
> "... since FinFET mask costs are 10x what they used to be in bulk CMOS"

Not true, it is closer to 2 to 3x, and it is (slowly) getting cheaper. While it will take a long time (~3 years) for 14/16nm FinFET processes to reach price parity with what 28nm is now, it is dropping in price faster than when the 28nm generation came out due to the massive volumes of Apple, NVIDIA, etc.

It is looking like the 28nm generation will be a mainstay node, with continued investments by the major pure play fabs to keep bringing costs lower.

trsohmers··on A Look inside Russian 28nm MIPS CPU – Baikal-T1
It's a flip chip design, so the heatspreader and thermal paste does not touch the part that you are seeing in these images. The areas that I think you are referring to (the big "blob" areas) are computer place and routed standard cells (primarily transistors) that the PNR tool brute forces efficient placement under certain constraints. While it does not produce the most pretty or clean design, it does a good enough job in optimizing for speed and area for most designs.
trsohmers··on The “high-level CPU” challenge (2008)
Our current plans are for very simple SIMD modes that reutilize the same hardware to maximize Area and power efficiency. Right now, we can separately load/store the upper half and lower half of a 64 bit register with 32 bits of data, and while we did not have the time to implement it on this silicon, the plan for the future is to have a mode switch to allow the user to use the same set of instructions/hardware/registers to do double, single, or half precision floating point operations.

The other thing we are looking forward to directly test/compare are unums, specifically the new "type 3" ones known as Posits, which are useful (for some definitions of useful) all the way down to 4 bits, and have a greater dynamic range plus greater precision than IEEE floats while using let's bits and theoretically lower area/power on a chip.

trsohmers··on The “high-level CPU” challenge (2008)
On your edit: RISC and VLIW are not mutually exclusive. I would say we are a RISC architecture (Load/store based/everything happens on registers, fixed instruction word size, shallow pipeline) that happens to have a much simpler instruction decoder & higher instruction level parallelism since our compiler guarantees that the instruction bundle (the 64 bit Very Long Instruction Word containing 4 "syllables" of instructions itself) will only give instructions that will not conflict with anything it is given with or in the pipeline.

Other than we are faster (20% higher clock plus 2x higher IPC) and more efficient (Epiphany is single precision only, where we are twice as efficient as them, we also support FP64), there are three quick points:

1. We actually have double the local scratchpad of Epiphany per core, and our SPM banking and register port scheme actually enables us to operate all four functional units within a core simultaneously, while also having data go in and out of the core on the Network on Chip. With Epiphany, you are very limited in what instructions can run together primarily due to port conflicts... the biggest difficulty is that you can't do any sort of control instructions with anything else.

2. As far as the Network on Chip goes, we are able to guarantee all latencies through a strict deterministic static priority routing scheme. Epiphany had 3 levels of it's Network on Chip, one for stores, one for read requests (8x slower than stores), and one for off chip communications. We have a (patent pending) way of simplifying all of this greatly while reducing latency and having greater bit efficiency.

3. Off chip memory bandwidth is extremely important to us. Even on our test chip, we have 4x higher bandwidth than the Epiphany IV, plus lower latency... our chip to chip and memory interface also uses the exact same protocol as our NoC, simplifying things even further.

There are a handful of smaller things, though my biggest gripe with Epiphany has always been the lack of bandwidth both on and especially off chip. If you are targeting DSP and similar applications like both Epiphany and we are, you really really need to have the ability to saturate your networks and match compute capabilities with it.

trsohmers··on The “high-level CPU” challenge (2008)
1: Around the 30:40 mark in the video I transition to the live hardware demo, where I learned to never do live demos. While it did not work there in person, I do show pictures from it working earlier in the day along with the results from our 1024-point in place FFT assembly test. It was written and tested Monday and Tuesday of last week (finished the night before the presentation), so it is not fully optimized, but we are getting a very good 25 double precision GFLOPs/watt. As a comparison, Intel and NVIDIA's advertised FP64 numbers are around 8 to 12 GFLOPs/watt while they are on a more advanced process node (They are on 14/16nm, while our test silicon was on 28nm).

We're in the process of getting cross platform benchmarks (HPL, HPCG, FFTW, Coremark, etc) up and running, but I'm hoping we'll be posting results shortly.

2. Since Linux 4.2, there has been a mainline port for STM32 which does not have a MMU, so porting linux is technically possible, though not on our priority list. Something like uCLinux would probably be easier, but not as useful. I have no doubt our cores would be able to handle it, it is just that our current customers already expect their software to be running on bare metal.

trsohmers··on The “high-level CPU” challenge (2008)
Boy, do I have a new processor architecture for you. For the past 3 years my company (REX Computing), has been working on a new architecture that not only covers both of your points, but removes hardware managed caching entirely and replaces it with software managed (physically addressed) scratchpad memory. By removing all of the unnecessary logic (MMU, TLB, everything required for virtual addressing, prefetching, coherency, etc) associated with the onchip memory system, we can fit more memory on the chip itself, have lower latency to that memory, and use a lot less power. As we fully expose the memory system to our software tools, we can make very good decisions at compile time that replicate most of the features you get in having a hardware managed caching system.

We have silicon back, and are now working with early customers in showing a 10 to 25x energy efficiency improvement for high performance workloads.

I gave a talk at Stanford last week that covers the hardware architecture and shows off our development hardware: https://www.youtube.com/watch?v=ki6jVXZM2XU

trsohmers··on The “high-level CPU” challenge (2008)
Well unums in particular (especially the new type 3 "posits" ones) are very interesting and practical for hardware implementation. John Gustafson, the creator of unums, gave the first talk on type 3 unums at Stanford earlier this month, and there are already multiple hardware implementation efforts (one of which I am indirectly overseeing). While many people have recently hyped up low precision floating point for things like machine learning, Posits provide greater dynamic range AND greater accuracy for very few bits (as low as four, but with good ML starting 8 to 10 bits). On top of all of that, it costs less area and power in silicon than IEEE hardware.

Check out the talk: https://www.youtube.com/watch?v=aP0Y1uAA-2Y

trsohmers··on The REX Neo Architecture: An energy efficient new processor architecture
The website has not been updated in over a year, while this talk was from this past Wednesday. The video shows our hardware in a temporarily broken state, and goes over what has happened over the last year of development leading to our first test silicon.

We will be relaunching the site early next month with developer documentation, demos of our test chip, and a look at the future 256 core production chip.

trsohmers··on The REX Neo Architecture: An energy efficient new processor architecture [video]
Hey, Thomas Sohmers (the presenter) here. Feel free to ask any questions... it is a pretty long talk, and I recommend watching at 2x speed if you have the ears for it.
trsohmers··on Stanford Seminar: Beyond Floating Point: Next Generation Computer Arithmetic
Abstract on the Stanford website, as well as link to slides: http://web.stanford.edu/class/ee380/Abstracts/170201.html

First 50 minutes or so is an overview on John Gustafson's unum's, followed by the introduction of "type 3 unums" that have two subtypes: Posits (discussed here), and Valids. In short, Posits have a greater mathematical accuracy AND dynamic range of floats, while in many cases also using fewer bits in the representation.

At around the 53 minute mark, the presentation switches over to Dr. Isaac Yonemoto (HN username: dnautics) who gives a live demonstration of a Julia implementation of Posits, showing it preforming favorably for both a FFT and neural network compared to low precision floats.

trsohmers··on Why are people from the future not time traveling to our period? (2012)
Are you sure we're not in 1985A? There's a big Biff Tanen hotel right down the street from the whitehouse...
trsohmers··on Peter Thiel considering bid for California governor
It's a true story, and both Thiel and Musk have been saying it the same way for over 15 years. https://www.youtube.com/watch?v=mOI8GWoMF4M
trsohmers··on Mill Computing in 2017
We're still alive, received our first silicon back in October, and will have news coming out soon.
trsohmers··on HiFive1: A RISC-V-based, Open-Source, Arduino-Compatible Development Kit
The frequency limit for bit banging is usually limited by the GPIO pins (well, their drivers) rather than the core clock. I use some ~250MHz PIC32MZ's with their GPIOs capping out at around 50MHz, and even the raspberry pi with a core clock of ~1GHz can only bit bang with its GPIOs up to ~60MHz without running into problems. As for LVDS, it takes up more area and power, and you usually need to license the IP block rather than it being available for free from the standard cell library provider.
trsohmers··on Mobile Holoportation
There are no NVIDIA parts in the Hololens... It uses a ULV Intel processor with integrated graphics, along with what Microsoft calls a "holographic processing unit" which is built by Microsoft, but utilizes slightly customized Tensilica cores for DSP offload.
← PreviousPage 4 of 12Next →