You both deserve such a large audience. Please do whatever you need to grow. There is a true lack of indepth heady content out there. Trust me, we want and need it.
I actually prefer this non-scripted part to the scripted one. Maybe the future shows could be a mixture of both - one of you presents the topic (scripted part) and then you both discuss the topic (non-scripted part). But, anyhow, I'm eager to learn about this rapidly developing field.
I do have to say Sally Ward-Foxton is a very good co-host. I will watch it later sadly, I do need some sleep and previous 2 nights I haven't slept. I will gladly watch it. And I do hope you might do a special episode (whether this podcast or your own style video) on MI300 when it is released. Because it seems like it can be a very big deal in AI (and HPC), especially now that people realized that CPU's are also useful for AI (which is something that I only noticed getting to mainstream people about a year ago at best. There are still many people who think that for AI all you need is tons of GPU's and CPU's are there only to steer GPU's). Because technologically MI300 will be something very, very interesting.
100% on the comments about tendencies towards mumbo jumbo of the marketing and press on this stuff. Hope you guys can continue to address the little guys' conveniences in this tech. In looking at - and being very excited about RISCV for instance - it would be very interesting to see how the architecture's design may, or may not, improve convenience of creating future versions of chips from the past and present, and not just in BIG AI. (ie: ease of designing chips for PC, phone, tablet, consumer-level) Is that going to get easier or harder for hopeful, independent operations like Pine64 and Purism? You have a fantastic sanity and down-to-earth-ness.
I dig. Nerds who know what the f they're talking about always catch my attention. Hope to see much more as this insanity keeps on slipping into the future.
Once you start creating content on RUclips, you understand just how important concise videos are. RUclips is your daddy, and they are the ones that control your videos reach and if you don't get metrics like high watch time percentage it ruins you. Being concise/watch time is the most important thing on RUclips. You can't just upload whatever you want, irregardless of its performance. A viewer can switch at any time to a different content so you are always competing with every other piece of content and if your video performs poorly not only is it hurt your current video it can hurt your future ones
Your stance on china itself is odd.. are you for the ccp or against them? Do you support the ban or not? The old proven wrong argument that 'you will cause them to come up w their own' is stale and not relevant
Sorry, had to reupload due to a low pitched hum. :) Links to podcast sites in the description!
You both deserve such a large audience. Please do whatever you need to grow. There is a true lack of indepth heady content out there. Trust me, we want and need it.
I actually prefer this non-scripted part to the scripted one. Maybe the future shows could be a mixture of both - one of you presents the topic (scripted part) and then you both discuss the topic (non-scripted part). But, anyhow, I'm eager to learn about this rapidly developing field.
I do have to say Sally Ward-Foxton is a very good co-host. I will watch it later sadly, I do need some sleep and previous 2 nights I haven't slept. I will gladly watch it. And I do hope you might do a special episode (whether this podcast or your own style video) on MI300 when it is released. Because it seems like it can be a very big deal in AI (and HPC), especially now that people realized that CPU's are also useful for AI (which is something that I only noticed getting to mainstream people about a year ago at best. There are still many people who think that for AI all you need is tons of GPU's and CPU's are there only to steer GPU's). Because technologically MI300 will be something very, very interesting.
@Ian Haven't heard of Gustafson's posits for a while. I hope that will change and will get implemented in hardware.
Fantastic job guys! These episodes were incredibly well done, looking forward to the next one!
Loved the content and Sally accent, hope to hear more than more on this topic.
Excellent work and episode. Please do one on Tesla's Dojo chips
100% on the comments about tendencies towards mumbo jumbo of the marketing and press on this stuff. Hope you guys can continue to address the little guys' conveniences in this tech. In looking at - and being very excited about RISCV for instance - it would be very interesting to see how the architecture's design may, or may not, improve convenience of creating future versions of chips from the past and present, and not just in BIG AI. (ie: ease of designing chips for PC, phone, tablet, consumer-level) Is that going to get easier or harder for hopeful, independent operations like Pine64 and Purism?
You have a fantastic sanity and down-to-earth-ness.
This podcast is excellent in sonmany ways. I really do love it, forget the formal version just make longer informal
Nice. I'm wondering what AMD will bring to the table in consumer space.
I dig. Nerds who know what the f they're talking about always catch my attention. Hope to see much more as this insanity keeps on slipping into the future.
I prefer this format, I found my eyes glazing over with the scripted episode, so feedback for the show, keep it casual
We want high perf, low cost, cuda compatible low precision inference cards. Realtime stable diffusion is the next big thing
This seems a lot more natural than the actual podcast. Maybe next time do away with the scripted monologues to camera and just have a chat
Once you start creating content on RUclips, you understand just how important concise videos are. RUclips is your daddy, and they are the ones that control your videos reach and if you don't get metrics like high watch time percentage it ruins you. Being concise/watch time is the most important thing on RUclips. You can't just upload whatever you want, irregardless of its performance. A viewer can switch at any time to a different content so you are always competing with every other piece of content and if your video performs poorly not only is it hurt your current video it can hurt your future ones
We still need to increase ai training 10 times. It takes a few days for openai to train chatgpt.
Waiting for Ian to "Pretty Woman" a large AI company once TechTechPotato makes it big 😛
Would like to see a systems Intel Habana Gaudi2-based HGX versus Nvidia DGX H100/640G discussion.
Wait: You recorded that in 2022? How long is your pipeline?
21:00 😂
how is an int1 network different from desinging a FPGA ??
nvidas current gen is the H100
We literally say that
@@TechTechPotato in the scripted video, it was mentioned that the A100 was Nvidia's current gen
As usual i just listen in because i love tech but i have no idea what is going on lol. Forcing chinese company to lower performance is bullshit lol
Musk trolled me for years, and when I asked what he wanted, he said: “Nothing”
I want to step on his toe.
Your stance on china itself is odd.. are you for the ccp or against them? Do you support the ban or not? The old proven wrong argument that 'you will cause them to come up w their own' is stale and not relevant