Tachyum's Prodigy emulator achieves first boot, runs Linux and says 'hello, world'
- Reference: 1630491966
- News link: https://www.theregister.co.uk/2021/09/01/tachyums_prodigy_emulator_boots/
- Source link:
Founded in [1]2017 by a team made up of Skyera and SandForce co-founder Dr Radoslav "Rado" Danilak, Wave Computing co-founder Ken Wagner, engineer Igor Shevlyakov, and hardware architect Rod Mullendore, Tachyum is in the process of designing what it calls a "universal processor.' Dubbed [2]Prodigy , the design - originally known as [3]the "Cloud Chip" - is claimed to be able to run programs written for x86, Arm, and RISC-V architectures as well as its own native architecture, and to do so [4]10 times faster than its rivals.
In the years since its unveiling, Tachyum's claims haven't wavered - but they also haven't yet been proven, something which didn't stop the company joining an EU project to [5]build an exascale supercomputer . Earlier this year, Tachyum announced the delivery of its first [6]FPGA-based Prodigy system emulator - and now, around three months on, it has laid claim to a successful first run of a Linux operating system and a simple user-mode application.
[7]
"In the last few months the entire company, led by Krishna Thatipelli's design team, has made tremendous progress on Prodigy," said architecture manager Bill Radke in a video presentation demonstrating the company's work, "maturing our processor core to the point where we're announcing groundbreaking news: that the Prodigy Processor FPGA Prototype is now booting a Linux kernel, running a user-mode application, and shutting down successfully.
[8]
[9]
"This is one of the most critical points in proving a mature design and verification of a high-end processor. The design includes CPU cores, caches, memories, TLBs, and a console, and the simulation uses exceptions, interrupts, page-walking, and both user- and kernel-mode execution. It's not just a hardware and software design success, but also a key verification milestone."
During the demo, Radke showed the dual-board prototype platform - one board handling input/output peripherals including memory and storage and the other hosting up to four FPGA-based processor prototypes - being loaded with the Prodigy soft-core binary, then booting into a text-only Linux environment. Sadly, the demo stopped short of any user interaction: instead, it automatically executed a simple "hello world" program from user space then shut down.
Proof?
It's a far cry from proving the company's headline claims for its novel processor design: Tachyum still says its Prodigy will offer a fourfold reduction in total cost of ownership over rival processors from Intel and AMD, driven by a tenfold reduction in power draw core-for-core, while being capable of beating Nvidia's "fastest GPU" in high-performance compute and AI-related workloads, but has yet to offer any real-world proof for its expectations.
[10]'Universal Processor' startup Tachyum unveils full-system Prodigy emulator ahead of sampling later this year
[11]Prodigy chip brainiacs Tachyum hook up with Euro HPC consortium
[12]'Prodigy' chip moonshot gets hand from Arm CPU guru Prof Steve Furber
[13]The Lord saw that man was wicked and sent a flood of storage news
[14]Say a stealthy 'hi' to Gluent, Tachyum and VAST Data
Radke, though, is confident that progress continues. "Our next milestone video," he promised, "will be running a much larger set of services and user-mode applications from interactive sessions."
"Despite some hurdles in acquiring some IO IP [input/output intellectual property licences] from our suppliers, we are very pleased with the progress that has been made in bringing Prodigy to life," said Danilak, a company co-founder and chief executive officer.
[15]
"Achieving a successful Linux boot in two months after receiving the FPGA IO motherboard prototype is a huge milestone. As we turn to further debugging and testing, we look forward to being able to run larger user-mode applications on the Prodigy prototype next month. This is an exciting journey, as we continue our drive toward enabling human brain-scale AI, while significantly reducing data centre power consumption."
Tachyum has said it is ready to begin offering the FPGA emulator to customers, and that it is on-track to tape-out the first Prodigy part by the end of 2021 - though that's after missing its original schedule to begin manufacturing of physical Prodigy chips at TSMC by the end of 2019.
The first-boot demo video is available [16]here . The company had not responded to queries regarding the architecture for which the demo applications were compiled by the time of publication. ®
Get our [17]Tech Resources
[1] https://www.theregister.com/2017/02/13/gluent_tachyum_and_vast_data/
[2] https://www.theregister.com/2018/08/24/tachyum_prodigy_cpu/
[3] https://www.theregister.com/2018/02/13/tachyum_funding/
[4] https://www.theregister.com/2017/04/10/teaser_tech_alert_says_us_falling_behind_china/
[5] https://www.theregister.com/2019/02/27/tachyum_euro_hpc_club/
[6] https://www.theregister.com/2021/06/09/tachyum_prodigy_emulator/
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/servers&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YS@jwjgWqjzWbQ3A6hqFMwAAABc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/servers&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YS@jwjgWqjzWbQ3A6hqFMwAAABc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/servers&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YS@jwjgWqjzWbQ3A6hqFMwAAABc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[10] https://www.theregister.com/2021/06/09/tachyum_prodigy_emulator/
[11] https://www.theregister.com/2019/02/27/tachyum_euro_hpc_club/
[12] https://www.theregister.com/2018/07/27/tachyums_prodigy_cpu/
[13] https://www.theregister.com/2017/07/18/storage_news_roundup/
[14] https://www.theregister.com/2017/02/13/gluent_tachyum_and_vast_data/
[15] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/servers&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YS@jwjgWqjzWbQ3A6hqFMwAAABc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[16] https://www.tachyum.com/resources/presentations/
[17] https://whitepapers.theregister.com/
"'Universal processor' startup still no nearer to proving bold claims of tenfold performance gain over Chipzilla, AMD"
I'm not sure it matters. If it is open and doesn't do creepy things like the others, it should win out slowly in the end (unless it gets bought by Intel or AMD of course which is unfortunately too likely).
Or it doesn't exist at all, except to con money out of investors.
If they're really about to release the FPGA emulator to customers, then at least someone will get a look at what the instruction set looks like, and how novel it really is.
Have they patented anything? Then the patents will be published.
Or it doesn't exist at all, except to con money out of investors.
quantum computing
On the contrary, quantum computing only exists because - first - a diverse bunch of physicist thought it would be cool thing to achieve, and although they might need money to investigate it, they weren't asking "investors".
It might now be that - subsequently - a range of people use "quantum computing" to "con money out of investors", but this is not why quantum computing exists (whether in its current limited state, or in general).
Maybe I'm just too cynical, but a tenfold performance gain over Intel and AMD is a ridiculous claim unless they are trying to claim tenfold performance based on power consumption in which case I'd suspect that this outfit would claim their metric against the most power hungry Intel and AMD chips and not against their mobile offerings.
Not that another processor platform and competition is a bad thing, but this one comes across like yet another charlatan.
Transmeta
"Low power", "dynamic binary translation software layer" (although most processors translate instructions to microcode anyway): this reminds me of [1]Transmeta , and look what happened to them.
[1] https://en.wikipedia.org/wiki/Transmeta
Re: Transmeta
They should've moved from VLIW to FLIW.
Re: Transmeta
Ah, yes,where Intel payed a few hundred million dollars to shut them down by banning them from making x86 chips.
Ride
The FPGA will never be faster than a native implementation.
It's easy to take an ISA and make it 10x faster, but then when you have to implement edge cases, boundary and other checks, interrupts, any reasonable housekeeping and you'll find it is actually 10x slower.
Re: Ride
One assumes they'll bake it into real silicon once it's solid ?
It'll want to be a good simulator
To replace Keith...
Still waiting for the Atom Chip
Personally, I'm still waiting for the previous unbelievable breakthrough in CPUs to be released - [1]The Atom Chip
[1] https://www.theregister.com/2005/09/07/atom_chip_miracle_machine/
I have in my box of bits an Inmos T800 Transputer board. They were supposed to revolutionise everything. The software I was working on, well we dropped support when you needed 30 of the buggers (at £1,000 a throw) to get the same performance as a 90MHz PII.
The Intel/AMD architecture may be horribly complicated and burdened by backward-compatibility, but so long as megabucks keep getting thrown at it, it will be around for a while.
Hmm
If it runs too hot then it might be a Firestarter.
if the claims prove true then it sounds like it may be Always Outnumbered, Never Outgunned