News Tachyum releases a 1,600-page performance optimization manual despite continued tape-out delays and no actual silicon

[Admin]

Tachyum's new 1,600-page document to teach software developers to optimize performance.

Tachyum releases a 1,600-page performance optimization manual despite continued tape-out delays and no actual silicon : Read more

 

[bit_user]

Tachuym's Prodigy FPGA features built-in performance counters that enable real-time monitoring and analysis of runtime events.

Intel and AMD have these baked into their production silicon.

 

[Findecanor]

If they have released it, then where is it?
I can't find it on their web site.

 

[bit_user]

If they have released it, then where is it?
I can't find it on their web site.

The article links to a press release dated 2 days ago.

https://www.tachyum.com/media/press...prodigy-performance-optimization-manual/?s=31
​

That doesn't seem to link directly to the manual, however. Perhaps they expect you to contact them and sign a partner aggrement/NDA, before you can get access to it.

 

[P.Amini]

The article links to a press release dated 2 days ago.

​

https://www.tachyum.com/media/press...prodigy-performance-optimization-manual/?s=31​

​

That doesn't seem to link directly to the manual, however. Perhaps they expect you to contact them and sign a partner aggrement/NDA, before you can get access to it.

Any thoughts on this processor?

 

[bit_user]

Any thoughts on this processor?

It's a new ISA because... why?? Their original announcement sounded much more like VLIW or EPIC, but then they pivoted towards something more like a traditional RISC.

As far as I can tell, what supposedly makes it good for AI and HPC is that each core has vector extensions and a matrix-multiply engine. You could do that with RISC-V, so why again is it a custom ISA? Even ARM now has SME (Scalable Matrix Extensions), though I'm not aware of any cores that yet implemented it.

It feels like every few years that I hear about a promising HPC processor, but it seems like they chronically underestimate the time to bring something like that to market, and once they do, the mainstream guys have caught up & usually even passed them.

The main thing they seem to have going for them is being an indigenous European project and backing from the Slovakian (?) government. However, the competition they're up against includes the European Processor Initiative, which is currently using ARM and has already planned to switch over to RISC-V.

General Purpose Processor - European Processor Initiative

www.european-processor-initiative.eu

Also, for those who don't mind buying a non-European ARM CPU, Fujitsu's Monaka should provide stiff competition in the HPC sector:

https://www.tomshardware.com/deskto...g-on-arm-based-liquid-cooler-servers-for-2027
​

Lastly, with power consumption of up to 950 W, I'm skeptical how appealing they're going to be for general-purpose cloud workloads, which they also claim to be targeting:

https://www.tomshardware.com/news/tachyum-reveals-20-exaflops-supercomputer-design
​

Don't get me wrong, though. I'm not wishing to see them fail. I wouldn't even mind being proven wrong, because I think having a greater range of computing platforms is generally a good thing. I'm just trying to be realistic about their prospects. I appreciate the monumental amount of work they've put in, this far, and it saddens me to think it might all be for nought.

Anyway, what's much more intriguing to me is NextSilicon's Maverick-2:

HPC Gets A Reconfigurable Dataflow Engine To Take On CPUs And GPUs

No matter how elegant and clever the design is for a compute engine, the difficulty and cost of moving existing – and sometimes very old – code from the

www.nextplatform.com

The idea of dynamically building dataflow pipelines sounds like it has a lot more potential to maximize silicon utilization. That should ultimately lead to better perf/W and better perf/$ than Tachyum's approach, which is basically just following what ARM, Intel, and AMD are all doing, but with slightly wider vector pipelines and more general matrix extensions than AMX currently has.

 

[froggx]

Any thoughts on this processor?

i'm extremely impressed cooking up this wall spaghetti processor has gone on this long and they still can't show us anything that stuck.

it's straight vaporware. they've set a target beyond the most remote of fantasies and aren't afraid to drop a lawsuit on whoever they get to help them as soon as things inevitably fall flat. props with how they're willing to take their already silly specs and add more cores into it every couple years, guess that's the only way to beat 3 different flavors of more specialized chips. i do admit though, i admire how they're talking about combining risc and cisc features only at least a decade or 2 (or more?) behind x86 and arm having designs leveraging the merits of both these types of design in production, cause they gotta talk about something to keep from crying. part of me actually suspects they think they struck on something new with those perf counters they mentioned. imagine if those had slipped through the cracks without anyone on contract they could extor... ahem... take to court.

i really want to see if they can end the decade neither taping out nor tapping out. i think they've got what it takes.

 

[P.Amini]

It's a new ISA because... why?? Their original announcement sounded much more like VLIW or EPIC, but then they pivoted towards something more like a traditional RISC.

As far as I can tell, what supposedly makes it good for AI and HPC is that each core has vector extensions and a matrix-multiply engine. You could do that with RISC-V, so why again is it a custom ISA? Even ARM now has SME (Scalable Matrix Extensions), though I'm not aware of any cores that yet implemented it.

It feels like every few years that I hear about a promising HPC processor, but it seems like they chronically underestimate the time to bring something like that to market, and once they do, the mainstream guys have caught up & usually even passed them.

The main thing they seem to have going for them is being an indigenous European project and backing from the Slovakian (?) government. However, the competition they're up against includes the European Processor Initiative, which is currently using ARM and has already planned to switch over to RISC-V.

​

General Purpose Processor - European Processor Initiative

www.european-processor-initiative.eu

​

Also, for those who don't mind buying a non-European ARM CPU, Fujitsu's Monaka should provide stiff competition in the HPC sector:

​

https://www.tomshardware.com/deskto...g-on-arm-based-liquid-cooler-servers-for-2027​

​

Lastly, with power consumption of up to 950 W, I'm skeptical how appealing they're going to be for general-purpose cloud workloads, which they also claim to be targeting:

​

https://www.tomshardware.com/news/tachyum-reveals-20-exaflops-supercomputer-design​

​

Don't get me wrong, though. I'm not wishing to see them fail. I wouldn't even mind being proven wrong, because I think having a greater range of computing platforms is generally a good thing. I'm just trying to be realistic about their prospects. I appreciate the monumental amount of work they've put in, this far, and it saddens me to think it might all be for nought.

Anyway, what's much more intriguing to me is NextSilicon's Maverick-2:

​

HPC Gets A Reconfigurable Dataflow Engine To Take On CPUs And GPUs

No matter how elegant and clever the design is for a compute engine, the difficulty and cost of moving existing – and sometimes very old – code from the

www.nextplatform.com

​

The idea of dynamically building dataflow pipelines sounds like it has a lot more potential to maximize silicon utilization. That should ultimately lead to better perf/W and better perf/$ than Tachyum's approach, which is basically just following what ARM, Intel, and AMD are all doing, but with slightly wider vector pipelines and more general matrix extensions than AMX currently has.

Thanks for the answer, helpful as always. Respect.

 

[P.Amini]

i'm extremely impressed cooking up this wall spaghetti processor has gone on this long and they still can't show us anything that stuck.

it's straight vaporware. they've set a target beyond the most remote of fantasies and aren't afraid to drop a lawsuit on whoever they get to help them as soon as things inevitably fall flat. props with how they're willing to take their already silly specs and add more cores into it every couple years, guess that's the only way to beat 3 different flavors of more specialized chips. i do admit though, i admire how they're talking about combining risc and cisc features only at least a decade or 2 (or more?) behind x86 and arm having designs leveraging the merits of both these types of design in production, cause they gotta talk about something to keep from crying. part of me actually suspects they think they struck on something new with those perf counters they mentioned. imagine if those had slipped through the cracks without anyone on contract they could extor... ahem... take to court.

i really want to see if they can end the decade neither taping out nor tapping out. i think they've got what it takes.

LOL! thanks!

 

[bit_user]

it's straight vaporware. they've set a target beyond the most remote of fantasies and aren't afraid to drop a lawsuit on whoever they get to help them as soon as things inevitably fall flat.

Yeah, like the lawsuit against Cadence they licensed some IP blocks from (stuff like DDR5 and PCIe controllers, I think). They claimed the only reason their processor would miss its market window was due to their partner, but I'm nearly certain they'd have faced other setbacks if that IP had arrived on time. In other words, I suspect it was just a convenient excuse:

https://www.tomshardware.com/news/t...-prodigy-doesnt-meet-prodigious-goals-sue-you
​

Were there other such lawsuits by Tachyum?

i really want to see if they can end the decade neither taping out nor tapping out. i think they've got what it takes.

Yeah, sometimes the sunk cost fallacy can keep companies or ventures funded long past the point when they should've been cut off. That, or if the operation is allowed to fail, it would reflect poorly on certain powerful & influential people who've been backing them so far. I don't know for a fact that's what happening, here, but I wouldn't be the slightest bit surprised.

Early in my career, I worked for a startup company that reinvented itself a few times and went after wildly different markets, mainly because one of its big, early investors didn't want to admit that it was a bad investment. As far as I know, they weren't a traditional venture capital firm, BTW.

The normal thing for tech industry venture capital firms is to expect only 10% of their investments to hit it big. Being so accustomed to backing failures removes a lot of the stigma from admitting when you've got one in your fund. Then, they're pretty ruthless about cutting off and even clawing back funds from companies, basically as soon as they show signs of being non-viable.

 

[jalyst]

I feel like, every time I read all of bituser's posts (& several other posters here);
I get, like, a 'tincy bit' smarter & better-informed...

I don't always fully appreciate/understand what's conveyed, but some stuff sinks in.

It's also what I love about older-school interweb mediums, like this forum.
Of which are increasingly rarer gems, nowadays... 🙄

 

[bit_user]

I feel like, every time I read all of bituser's posts (& several other posters here);
I get, like, a 'tincy bit' smarter & better-informed...

I don't always fully appreciate/understand what's conveyed, but some stuff sinks in.

Thanks for the compliment!

I learned most of what I know about computer architecture by reading deep dives on Anandtech (defunct), RealworldTech (on hiatus, though the forums are a gem), wikichip (down???), here, and elsewhere. These days, I'd recommend ChipsAndCheese.com, if you really want to drink in the details of these sorts of things. I'm sure there are some great intro texts out there, as well as lectures on Youtube, etc.

In my experience, one of the best ways to learn things is to do a project of some sort and put that knowledge to use. So, if you really enjoy this kind of stuff, try to find ways to dabble with it!

It's also what I love about older-school interweb mediums, like this forum.
Of which are increasingly rarer gems, nowadays... 🙄

I like how we can talk without most of the ego that normally comes along with social media or real life. On here, it doesn't matter who you are or what's your background. If you have good ideas/facts/sources and know how to express them, that's pretty much all that matters. The downside of pseudonymous forums is mainly when people have an agenda or just like to troll. Thankfully, the mods do a lot to take care of such situations.