it’s pretty clear that much of AMD’s 2024 sales will be to Companies and institutions that can’t get Nvidia chips. They are behind on everything - drivers, software, frameworks, documentation and stability. But now they have large groups of engineers working on all of this. Along with more engineers out in the field trying to make the MI300s their employers bought useful. Before the market gets saturated in mid 2025, AMD will have a clean, stable, easy-to-use software stack. It’s not likely that AMD will ever overthrow Nvidia, but they can lock in a sizable, profitable section of the market. I bet they will have at least 20 percent to Nvidia’s 65 percent by 2026.
Congratulations on your promotion, this has been a great series, looking forward to 2024’s season 2. You and Ian provide great insights…team work works 😎
Thankyou both, talk about FAST, yeah things are changing in AI/Supercomputing. Hopefully some of the nVidia competitors can gain traction and compete, as you say. Enjoyed this throughly as so much Tech News these days are still fixated on social media "stuff".
I’m hoping AMD’s and Intel’s chiplet technology will allow them to make smaller GPUs/APUs for home use. These types of devices needs to be mainstream for ai apps to flourish. I want my own Jarvis and I’ll be damned if it can be unplugged from a cloud service at someone else’s discretion.
Intel should have doubled down on Kabylake-X. It was the correct direction. Unify the entire enthusiast platform into a single socket. Z boards with dual channel and 20 lanes with support for all the CPUs and X boards with quad channel and 40 lanes with support for all the CPUs (with the 'K' SKU CPUs having the limit of being only able to use dual ch and 20 lanes). Then for the people that just want a lot of cores without the I/O, they can do that. And the xOC guys would need only a single board for all of the core counts. Shady OEMs also would no longer be able to put 'K' SKU CPUs into garbage boards. win - win - win
I want the X58 times back, where you could put unlocked Xeons on the HEDT platform (and many Xeons were unlocked at that time with a lot of clock headroom, I could get my X5675 to above 4.2 Ghz on mid-tier air cooling!).
@@cracklingice I had some hopes that Intel might enable the server Xeons of Sapphire Rapids (+ Emerald Rapids) on their HEDT platform, but haven't seen any confirmation to date if that is indeed possible (only that motherboard vendors did their validation of the HEDT boards with these Xeons). The workstation SKUs are rare and do not come down in price as much over time as the phased-out server parts. The availability of these cheap server SKUs would attract me to that platform but motherboard pricing and uncompetitiveness of the platform vs Threadripper are a huge turn-off as it stands.
@@seylaw As someone on X99 with an E5 1660 V3 and 256GB of RDIMMs I agree. The problem from their point of view though - is that it enables people to not buy new stuff - even though for myself and possibly for you as well, it allows repurposing the desktop into a home server later in it's life. This is what I intend to do with my X99 when I can finally get an upgrade. If that's Intel/AMD coming to their damned senses and not expecting me to spend $2500 or more on just the board, CPU and minimum viable amount of memory (4x16GB) - or it's in the form of used Threadripper or W790 in the future is up to them.
Jupiter in Juelich will also get a Grace Hopper "booster". It's funded by the EU and aims to get like several exaflops. But also a novel CPU - which is really interesting.
Even though I haven't watched EVERY episode, I have to say that I learned A LOT from watching them. I don't really know anything about AI but I think have more of an interest in the hardware than what people are doing with it. (although I do like to play around with the free image gen ones from time to time) And I've learned about the hardware from two wonderful Brits! :) - THANK YOU for bringing all of this to the masses, and for covering a ton of 'unknown' companies (I'm a real sucker for an underdog's triumphant story LOL) - Here's to finishing up Season One!!!! - 🥂 CHEERS!!! 🥂 - And here's to starting your SECOND SEASON!!! - 🥂 🍻 DOUBLE CHEERS!!! 🍻 🥂 (I'm ASSUMING Sally successfully convinced her Boss to allow her to continue hanging out with degenerate Ian Potato character lol :D jk Ian
We just got some Nvidia Grace benchmarks today, and while it is certainly usable, it's not really performance competitive enough against Sapphire & Emerald Rapids. Pretty disappointing results from Nvidia.
Congratulations Sally!
it’s pretty clear that much of AMD’s 2024 sales will be to Companies and institutions that can’t get Nvidia chips. They are behind on everything - drivers, software, frameworks, documentation and stability.
But now they have large groups of engineers working on all of this. Along with more engineers out in the field trying to make the MI300s their employers bought useful. Before the market gets saturated in mid 2025, AMD will have a clean, stable, easy-to-use software stack.
It’s not likely that AMD will ever overthrow Nvidia, but they can lock in a sizable, profitable section of the market. I bet they will have at least 20 percent to Nvidia’s 65 percent by 2026.
Congratulations on your promotion, this has been a great series, looking forward to 2024’s season 2. You and Ian provide great insights…team work works 😎
Congratulations to your new title/ position Sally 👍
Great Show! - Thank you going into detail about all the AI news, would be great to see a weekly round up!
Thankyou both, talk about FAST, yeah things are changing in AI/Supercomputing. Hopefully some of the nVidia competitors can gain traction and compete, as you say. Enjoyed this throughly as so much Tech News these days are still fixated on social media "stuff".
Congrats Sally
Congrats Sally. So will Tenstorrent have a ML training chip ?
I’m hoping AMD’s and Intel’s chiplet technology will allow them to make smaller GPUs/APUs for home use.
These types of devices needs to be mainstream for ai apps to flourish.
I want my own Jarvis and I’ll be damned if it can be unplugged from a cloud service at someone else’s discretion.
@44:34 Love what she says here : "...we're talking about Can Anybody take Anything from Nvidia at this Point?" and @48:53
Intel should have doubled down on Kabylake-X. It was the correct direction. Unify the entire enthusiast platform into a single socket. Z boards with dual channel and 20 lanes with support for all the CPUs and X boards with quad channel and 40 lanes with support for all the CPUs (with the 'K' SKU CPUs having the limit of being only able to use dual ch and 20 lanes). Then for the people that just want a lot of cores without the I/O, they can do that. And the xOC guys would need only a single board for all of the core counts. Shady OEMs also would no longer be able to put 'K' SKU CPUs into garbage boards.
win - win - win
Instead Intel is giving you more E-cores & LP E-cores. Then you’re going to get rentable cores in Panther Lake.
I want the X58 times back, where you could put unlocked Xeons on the HEDT platform (and many Xeons were unlocked at that time with a lot of clock headroom, I could get my X5675 to above 4.2 Ghz on mid-tier air cooling!).
@@seylaw Sadly, W790 is the closest we get to that. Intel took away Xeons on Z boards with Z170 and X boards with X299 iirc.
@@cracklingice I had some hopes that Intel might enable the server Xeons of Sapphire Rapids (+ Emerald Rapids) on their HEDT platform, but haven't seen any confirmation to date if that is indeed possible (only that motherboard vendors did their validation of the HEDT boards with these Xeons). The workstation SKUs are rare and do not come down in price as much over time as the phased-out server parts. The availability of these cheap server SKUs would attract me to that platform but motherboard pricing and uncompetitiveness of the platform vs Threadripper are a huge turn-off as it stands.
@@seylaw As someone on X99 with an E5 1660 V3 and 256GB of RDIMMs I agree. The problem from their point of view though - is that it enables people to not buy new stuff - even though for myself and possibly for you as well, it allows repurposing the desktop into a home server later in it's life. This is what I intend to do with my X99 when I can finally get an upgrade. If that's Intel/AMD coming to their damned senses and not expecting me to spend $2500 or more on just the board, CPU and minimum viable amount of memory (4x16GB) - or it's in the form of used Threadripper or W790 in the future is up to them.
Jupiter in Juelich will also get a Grace Hopper "booster". It's funded by the EU and aims to get like several exaflops. But also a novel CPU - which is really interesting.
Even though I haven't watched EVERY episode, I have to say that I learned A LOT from watching them. I don't really know anything about AI but I think have more of an interest in the hardware than what people are doing with it. (although I do like to play around with the free image gen ones from time to time) And I've learned about the hardware from two wonderful Brits! :) - THANK YOU for bringing all of this to the masses, and for covering a ton of 'unknown' companies (I'm a real sucker for an underdog's triumphant story LOL)
- Here's to finishing up Season One!!!! - 🥂 CHEERS!!! 🥂 - And here's to starting your SECOND SEASON!!! - 🥂 🍻 DOUBLE CHEERS!!! 🍻 🥂 (I'm ASSUMING Sally successfully convinced her Boss to allow her to continue hanging out with degenerate Ian Potato character lol :D jk Ian
Oh and congrats on the promo Sally!!! :)
Hey Ian. What instructions are they adding to accelerate llm ops?
We just got some Nvidia Grace benchmarks today, and while it is certainly usable, it's not really performance competitive enough against Sapphire & Emerald Rapids. Pretty disappointing results from Nvidia.
Sally with the long hair. Awesome video series as well even though i am a normie lol.
Congratulations Sally!
Congratulations Sally