• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Discussion AMD SoC Halo series GPU discussion

Page 28 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
It’s kinda breaks the 64GB Mac mini M4 Pro market that’s costs $1999, you get 128GB with framework.

This might be a shock, but people buy Macs, because they want a Mac, not because it's a better value than an x86 PC.

The Value on the PC side isn't really there unless the big memory pool for AI is required, which is probably why AMD was pushing the AI message hard...
 
Well, I think this puts to bed the concern over STX-Halo pricing. Seems like Framework might have a breakout hit with their new machine. Hope everything goes well for them, I've always appreciated the work they are trying to do, even if I am yet to be a paying customer.
It's intriguing because it is a very non-framework product but still I think they have a much better support track record than e.g. Minisforum.
 
This might be a shock, but people buy Macs, because they want a Mac, not because it's a better value than an x86 PC.
Not the people he is talking about. They're buying Macs specifically for ML crap right now. But I wonder if that market will still exist for STH by the time this ships in Q3.
 
Apple will still own the high end. Especially if the M4 Ultra with 1TB/s 256GB RAM model comes around WWDC for $5000.

No one is beating that. But the $2000 area now got comp
 
The Value on the PC side isn't really there unless the big memory pool for AI is required, which is probably why AMD was pushing the AI message hard...
Even for AI the proposition is iffy. You can run large models, but slowly. Theoretical max is something like 3-4 tokens/s for a 72b model at a decent quant.
 
This might be a shock, but people buy Macs, because they want a Mac, not because it's a better value than an x86 PC.

hackintosh on x86 works way better than windows emulation on a mac

halo desktop = mac mini killer, basically no point buying mac desktop anymore if you know the alternative

laptops are a different story bc macbooks are superior - screen battery etc, except macOS lol
 
Framework desktop is very cool, but the gaming advertising just doesn’t make sense the more I think about it. It’s just not good value for price/performance. Killer mini workstation though.
 
hackintosh on x86 works way better than windows emulation on a mac

halo desktop = mac mini killer, basically no point buying mac desktop anymore if you know the alternative

laptops are a different story bc macbooks are superior - screen battery etc, except macOS lol

Extremely unlikely you get a hackintosh running on this. It's mainly running on older HW. Pretty sure it doesn't work on anything newer than RDNA 2 GPUs, and this is RDNA 3.5.
 
Last edited:

oh my god "pro developers" at hacker news will never "get" hardware and systems 🤪😂

the distinction between devs and sysadmins from the 90s will always remain

all comments are "so weird! desktop and non-upgradable ram?? this is not framework..."
and the worst "why not ATX size?????" 🤪🤪🤪

Extremely unlikely you get a hackintosh running on this. It's mainly running on older HW. Pretty sure it doesn't work on anything newer than RDNA 2 GPUs, and this is RDNA 3.5.

if there's demand it will definitely get updated, currently still an odd hobby project
 
Sooo... I guess when it will be available, I'll get a silent psu from HDPLEX, a Sparrow MQ4 case and have a killer quiet system. And the price is even lower than I thought. For a desktop cpu, I'd have had to pay another 3-400€ for a MB, 200€ for ram, so this cpu is... Cheap??

Wow...

Edit: - didn't know the collab was so deep. Frank Azor coming on stage? Nice! Framework resetted their road map to make the desktop. It has collaboration with AMD, Coolermaster, Noctua.


Scroll to 12:00 if timestamp isn't working
 
Last edited:
Actually on one hand it's impressive they have released that, on the other hand I would have expected more customization from framework, but probably it's out of the question given the scale they operate at. What I would like to see is the possibility to:
  1. Have more memory options for 8 core SKU
  2. Provide more PSU options.
  3. Provide a way to use the internal PCIe slot with their own case, like HP or Lenovo are doing.
I know that 2 and 3 can be solved by doing a custom build, nevertheless as an option it would be nice. But I was really fighting with myself should I preorder or not, but given the time before shipment and lack of info at this stage (idle power draw, how noisy their setup is, what are the bios customization options) I guess I will wait to know if I would be better of with a custom build or by using their case/psu.

Still, it was a pleasant surprise.
 
if there's demand it will definitely get updated, currently still an odd hobby project

No, it won't.

It's depends on getting OS and drivers from Apple, and Apple is NOT releasing new drivers for x86 Macs. Even old drivers are disappearing from newer releases, such that most involved in Hackintosh recognize it's demise. The writing's been on the wall for Hackintosh since Apple announced ARM Macs. A couple events are pushing the demise closer:

The imminent loss of non-T2 support:


The loss of old Wifi/BT driver support


Right now, to keep a hackintosh running, it has to be on older HW, and stuck at older OS versions.

So there is NO chance at all that this APU will work as a Hackintosh.
 
What was it? There will not be Strix Halo soldered into desktop motherboards? 😉

Anyway. Since the topic of Memory on Package appeared, let me clarify something for you all.

DO NOT EXPECT MoP to appear for designs like Strix Point and Strix Halo.

Something above that, that targets premium high end and ultra premium designs, Like MX Max chips - that is completely different story. That may be the ONLY way to increase memory bandwidth for large iGPUs, even up to 80 CUs.

I do not expect those before Zen 7, or whatever the next gen architecture will be called.

Secondly. In the world of AI, and ipadisation of Apps, and Microsoft making any PC an Xbox - expect that memory requirements will skyrocket. Which will make large RAM pools viable, and to have a use case, even on mainstream.
 
Secondly. In the world of AI, and ipadisation of Apps, and Microsoft making any PC an Xbox - expect that memory requirements will skyrocket. Which will make large RAM pools viable, and to have a use case, even on mainstream.
I expect to see some insane quality texture packs that go beyond 32GB VRAM and maybe a few that specifically target 128GB of Strix Halo. AMD continues their tradition of breaking RAM requirement barriers, starting with the fall of the 4GB barrier with AMD64.

Which reminds me, next time anyone says "AMD what AMD who?", I will show them this:

1740580410275.png
 
Back
Top