Wow, that was quite the interview. The questions were exceptionally good, and the answers finally gave me a good idea about tenstorrent and their place in the market
I'm short on time so just quickly checking in, but I'm totally in the market for something to _run_ models on, that is better utilized than a GPU. I'm running a local Stable Diffusion instance that people can use through my Discord bot and I would love to build a new machine with dedicated AI hardware in it 🤩
If you can sell me a PCIe card to run 34B models locally... I am all for it, and want something before this year ends. Nvidia is expensive but works. Intel (PVC-56, Gaudi2) isn't selling to consumers/developers on workstation scale. AMD I have not even looked.
10:00 "I will probably get in trouble fo saying this. Ten torrent, tens torrtent, 10 to rent..", offtop, but why this company isn't called simply Tensor-2-rent, Tensor cores to rent. or something like that. T2R. TTR, for short.
@@RobBCactive and this is exactly my point, it can be associated with file sharing, point 2 point, socket to socket, skiping the server side, direct IP, not necessarely hardware,
@@KabelkowyJoe Dictionaries help : torrent (countable noun) a sudden large or too large amount, especially one that seems to be uncontrolled: a torrent of He let out a torrent of abuse/angry words. a torrent They are worried that the flow/trickle/stream of tourists could swell into an unmanageable torrent if there are no controls. [ C ] a large amount of water that is moving quickly: Heavy rainfall turned the river into a rushing/raging torrent. Tens is for tensor used in AI, not the TENS units used to give electro-stimulation
Hmm if you have something i‘ll take it i hope it’s very good at matrix math. But i‘d like to have a nice api for a very fast C/C++ language. I‘m not using Python api‘s. Does it maybe even run MIRL? And if you get me a RiscV as a cpu with CAN-FD and say SDIO/PCIe i‘ll take it even faster. For one of our platforms we are looking for a new CPU not much AI needed there though maybe something like Denoising..
Nvidia will eat us all with the cash enough to pay $80k/year per a Cadence EDA seat to re-layout GDSII incrementally. Fused multiply-add IP does require a giant corporation to run, it's not Linux.
Chief Customer Officer... Its like saying "Customer Success Champion" but for corporate officers. 😂 Sales is turning into a bad word in corporatespeak.
Can you please sell me an ai inference machine? Something that can do 70 billion parameter large language models at 10 tokens per second. I would like good software support please. My only solution right now is the m1 ultra. I would buy that but I would like alternatives however.
![[30] Ian Interviews: Dr. Thomas Caulfield, CEO Global Foundries](http://i.ytimg.com/vi/vUQ6Wizhbpo/mqdefault.jpg)
Will you guys have a educational institution discount ? Or Student network deals ?
This series is fantastic, thank you to tenstorrent and TTP for partnering on this
Wow, that was quite the interview. The questions were exceptionally good, and the answers finally gave me a good idea about tenstorrent and their place in the market
I'm short on time so just quickly checking in, but I'm totally in the market for something to _run_ models on, that is better utilized than a GPU. I'm running a local Stable Diffusion instance that people can use through my Discord bot and I would love to build a new machine with dedicated AI hardware in it 🤩
I'm in the same boat as you, definitely want AI focused hardware
If you can sell me a PCIe card to run 34B models locally... I am all for it, and want something before this year ends.
Nvidia is expensive but works. Intel (PVC-56, Gaudi2) isn't selling to consumers/developers on workstation scale. AMD I have not even looked.
10:00 "I will probably get in trouble fo saying this. Ten torrent, tens torrtent, 10 to rent..", offtop, but why this company isn't called simply Tensor-2-rent, Tensor cores to rent. or something like that. T2R. TTR, for short.
Torrent means something think of bit torrent.
@@RobBCactive and this is exactly my point, it can be associated with file sharing, point 2 point, socket to socket, skiping the server side, direct IP, not necessarely hardware,
@@KabelkowyJoe Dictionaries help :
torrent (countable noun)
a sudden large or too large amount, especially one that seems to be uncontrolled:
a torrent of He let out a torrent of abuse/angry words.
a torrent They are worried that the flow/trickle/stream of tourists could swell into an unmanageable torrent if there are no controls.
[ C ]
a large amount of water that is moving quickly:
Heavy rainfall turned the river into a rushing/raging torrent.
Tens is for tensor used in AI, not the TENS units used to give electro-stimulation
I really want to know when the IPO is going to be for Tenstorrent
Is there a podcast feed with these interview episodes?
th-cam.com/play/PLpZKwgOOdKTpZMKLPFQ5Sb4k9JmzlWm_f.html&si=nO7wnyZlvSPRN_Fw
Hope they have a good export compliance guy.
This interview has been amazing
Hmm if you have something i‘ll take it i hope it’s very good at matrix math. But i‘d like to have a nice api for a very fast C/C++ language. I‘m not using Python api‘s. Does it maybe even run MIRL?
And if you get me a RiscV as a cpu with CAN-FD and say SDIO/PCIe i‘ll take it even faster. For one of our platforms we are looking for a new CPU not much AI needed there though maybe something like Denoising..
I can't believe Tenstorrent have a 8-wide decode RISC-V cpu..
Nvidia will eat us all with the cash enough to pay $80k/year per a Cadence EDA seat to re-layout GDSII incrementally. Fused multiply-add IP does require a giant corporation to run, it's not Linux.
"$80k/year per a Cadence EDA seat" lmao this is so true
I want one!
Please, please we need hardware!!
Chief Customer Officer... Its like saying "Customer Success Champion" but for corporate officers. 😂 Sales is turning into a bad word in corporatespeak.
Can you please sell me an ai inference machine? Something that can do 70 billion parameter large language models at 10 tokens per second.
I would like good software support please. My only solution right now is the m1 ultra. I would buy that but I would like alternatives however.
Can these guys be serious, how many years have they been teasing prosumer hardware???? the market is primed for lower than Nvidia costing solutions!
Less than two years really, and it takes at least that long to get a scalable chip into the market.
"ok, but I want it for free. agree?"
Sure, send your cv