IBM
AMD MI300A: 24 Core Zen 4, 128GB HBM3, CDNA3
AMD MI300A: 24 Core Zen 4, 128GB HBM3, CDNA3
#AMD #MI300A #Core #Zen #128GB #HBM3 #CDNA3
“TechTechPotato”
The best thing about the semiconductor space is the amazing hardware that companies are dreaming of – especially now with …
source
To see the full content, share this page by clicking one of the buttons below |
I wonder if these massive amounts of HBM everyone is putting in their ML systems will at some point mean we're gonna see it return to consumer hardware as well, if manufacturing cost comes down through mass production and experience with the technology.
you and Doug's look like relatives hahah, great video btw!
Cools ruff still.
Also who's the troll from Nvidia or one of it's partners who out the giga pod ad there huh. At first I was confused if Ian was talking about the wrong brand lol but was like the video title is about AMD as well so surely he didn't make a mistake.
Can I haz ATX motherboard?
I’m still trying to figure out how MI300X boots without CPU cores.
Excellent summary, look forward to longer video review
"trying to be a second source", could you be more biased Mr Nvidia lanyard "business" guy? 8/10 top supercomputers and these analyst types think they can't figure out ai, gonna be in for a rude awakening on the 6th.
That other analyst was not good.
And Nvidia in the background ^^
What would AMD offer to Nvidia in a second-source agreement?
But can it run Crisis?
You need to get (borrow or something) 1 socket version of this system with this APU and test it with games and compare it to Nvidia 4090 and RX 7900 XTX.
If there is no video output, find workaround.
Is it hbm only ? Are there any rdimm slots?
I think AMD is gonna need to find a way to use their chips on nvidia's software…. can the AI we already see be run on ANYTHING other than nvidia?
… NVIDIA GIGA POO…..
… huh ….. that is an unfortunate choice of font…. but they did get to "crap" all over AMD with their branding in every shot so….
… oh—kay?…
Summary
– 0:00 The AMD Mi 300a was unexpectedly launched at the Gigabyte Booth, expected to be officially launched on December 6th.
– 0:08 The shown Mi 300a is a quad socket system with two sockets populated and two not, with LGA 696.
– 0:18 The system contains tons of pcie and is considered the high-end compute going into HPC systems.
– 0:26 The system includes an APU with 24 cores and 128 GB of HBM for high-performance computing.
– 0:48 Super Micro is planning to release 2U and 4U versions of the system.
– 1:21 AMD's machine learning strategy is to become a second source provider to Nvidia.
– 1:32 AMD offers chiplets and a lot more memory bandwidth compared to Nvidia's monolithic die.
– 2:38 The Mi 300a system is a four-socket system where all sockets can communicate with each other.
– 3:00 Density is a major factor in the design of these systems, due to the need for fast networking within a rack.
– 3:42 The system may come in liquid-cooled versions for even more density, as air cooling seems to be on its way out.
– 3:58 There is a push for data centers to transition to liquid cooling for more power and space efficiency.
– 4:21 Doug from Fabricated Knowledge provided insights on AMD's machine learning strategy and the future of data center cooling.
love your t-shirt
Its 192GB on the MI300X…come on guys, youre supposed to be the experts….and why does Ian look like he's about to faint?
I want to see MI300 RUNNING! I want to see what it can do….why isnt AMD trying to get this thing out there ASAP
Zen4 does not support OpenCL and Radeon Hip runtime, 11 months conting from first release. I have 7840HS 780M.
I want to see the MI300A running!
Nice video.
Had to listen without the video, the gigantic Nvidia logo behind the amd stand is just too distracting XD (seriously though, I'm surprised the organization let that pass …)
0:18 – Those fans, no, blowers, are goddamn crazy!!!