40-45 watts idle ? 120w-130w for CPU? Stock ARM cores are such a mess.
Cheapest M4 MAX Mac Studio WITH 128GB RAM is $3499
Cheapest M3 Ultra Mac Studio with AT LEAST 128 GB RAM (256GB model, only other is 96GB) is $5599
Available Ryzen AI Max+ 395 SFF PCs with 128GB RAM
GMKTek
They are both stabbing at different sections of the market, which is admittedly quite niche at this point.
Cheapest M3 Ultra Mac Studio with AT LEAST 128 GB RAM (256GB model, only other is 96GB) is $5599
Available Ryzen AI Max+ 395 SFF PCs with 128GB RAM
GMKTek
AMD Ryzen™ AI Max+ 395 --EVO-X2 AI Mini PC $1999
Nimo's Smallest Office & Gaming AI PC - AMD Ryzen™ AI Max+ 395 (up to 5.1 GHz) - 128GB LPDDR5 8000MHz (16GB*8) - 2TB/4TB SSD With 3 Performance Modes Up to 120W $1869
Bosgame M5 AI Mini Desktop Ryzen AI Max+ 395 96GB/128GB+2TB $1699
Framework's 128GB configs start at $2099 with a 1TB drive.
That 256 (edit: should be 512) bit bandwidth comes with a cost, to the point that we can purchase 2 MAX 395 devices for the same price (or just a bit more) and also keep the same software ecosystem we are already using. Their newer, faster devices are certainly NOT going to be any cheaper.
They are both stabbing at different sections of the market, which is admittedly quite niche at this point.
Last edited:
Meteor Late
Senior member
- Dec 15, 2023
- 347
- 382
- 106
I was so looking forward to Strix Halo, but these things are insanely expensive 😭
adroc_thurston
Diamond Member
- Jul 2, 2023
- 7,958
- 10,704
- 106
stuff will get cheaper once units increment up.
It'll be even cheaper for mdsH due to adjacent market die sharing.
adroc_thurston
Diamond Member
- Jul 2, 2023
- 7,958
- 10,704
- 106
The performance and power of the CPU part seems comparable to Strix Halo.
But it's on 3nm instead of 4nm. Really...? Is the die size at least much smaller?
But it's on 3nm instead of 4nm. Really...? Is the die size at least much smaller?
adroc_thurston
Diamond Member
- Jul 2, 2023
- 7,958
- 10,704
- 106
Insanely expensive as compared to... what exactly? How cheaply do you think that you can get a pro level video card that has at least 96GB of VRAM? The NVidia RTX Pro 6000 Blackwell is north of $8000. There's a Huawei card that's a bit of an unknown quantity, but it's still almost $2000... And that's before you build the rest of the system...
They aren't obscenely expensive, they're just pricey, but for their targeted workload, they are priced quite well. I think that, were I to have the money to buy one, I'd still pass. I don't think that RDNA 3.5 is quite there yet for the GPU side. If they ever do one with RDNA 4 or 5, that'll probably be where it would be in a good place for 3-4 years of solid relevance.
That is why I quoted it, to confirm the warnings in case anyone had doubts
But they lack compute. Neither is actually supposed to replace H100
512b bandwidth?
At least Halo doesn't idle at 40W
[to be fair they did not mention the conditions for the idle power draw measurements, like if the fibers were connected]. I also wonder if STH did turn off core parking, as by default Windows will park 2nd CCD and usually it will hamper MT scores as the system tries to move as much load to the first CCD as possible. Of course it's AMD problem, but easily fixable.
But they lack compute. Neither is actually supposed to replace H100
As noted in the part of the quote that you edited out, the solution to the low compute is already in silicon and will be coming in the new Studios this summer.
Yes, the people who are able and willing to buy, power, and cool an H100 and a supporting system would never consider any of these boxes. I agree we can rule out all the people who can drop $50K or more for their local training and inference uses. An astute observation that one is.
Last edited:
adroc_thurston
Diamond Member
- Jul 2, 2023
- 7,958
- 10,704
- 106
I guess their d2d shoreline for the GPU is puny? idk.
Someone cheak STREAM numbers for the GPU blob.
Lol. The M5 Pro/Max with its new GPU architecture will crush it. I wouldn't be surprised if the M5 Pro comes with a max configuration of 128 GB RAM, and the M5 Max gets 256 GB RAM.
marees
Platinum Member
- Apr 28, 2024
- 2,092
- 2,703
- 96
Interesting that this is 3nm
Must be costlier than consumer gaming 4nm/5nm stuff then
www.servethehome.com
Must be costlier than consumer gaming 4nm/5nm stuff then
NVIDIA DGX Spark Review The GB10 Machine is so Freaking Cool
In our NVIDIA DGX Spark review, we see how these Arm and NVIDIA Blackwell powered systems with 200GbE RDMA networking are just freaking cool
ashFTW
Senior member
- Sep 21, 2020
- 327
- 248
- 126
M3 Ultra already supports 512GB. If there is an M5 Ultra, it will likely be higher since has been steadily increasing the max RAM (e.g. M2 Ultra was 192GB). I’m thinking 1TB in the next round, as long as they can source the appropriate memory chips. M5 max might go to 256.
About times; Let's see tomorrow is what I expect Microsoft going to announce...
And NV is joining as well:
Microsoft Ignite - November 18–21, 2025
Microsoft Ignite - November 18–21, 2025 in San Francisco, CA. Get the edge you need to drive impact in the era of AI. Join us to bolster your knowledge, build connections, and explore emerging technologies.
Geez, NV is sponsoring the official Microsoft Ignite backpack meanwhile AMD is sponsoring water bottle. Wonder what are NV and AMD cooking?
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter