It seems that's going to be the only way out. To move to MoP for the Medusa Halo part. And then, AMD can initially release it with LPDDR5 and later resin the IOD for LPDDR6, without making any changes to the package, to the OEM designs.
Indeed.
Want even higher? Need dat MoP.
lmao no.It seems that's going to be the only way out. To move to MoP for the Medusa Halo part
so much effort and for what?And then, AMD can initially release it with LPDDR5 and later resin the IOD for LPDDR6, without making any changes to the package, to the OEM designs.
Going by the number of batches FW Desktop seems to sell gangbusters. 8th batch atm and still shipping in Q3 according to them. That product alone makes mockery of the narrative that Halo is rare and really expensive.Framework Desktop (halo) likely has thousands of preorders occupied until Q3-Q4
That's not public, but you can assume that it's of a similar size as for the laptops (based on some minimum feasible order size at the ODM). With the latest gen of FW Laptop 13 with AI 300 series (opened at the same time as FW Desktop) the three different configurations are at batch #4, #4 and #5 respectively currently.How big is a "batch"?
Not sure where you got the current price from, but framework webpage still lists 1699$ for bare mobo with 128GB.Prices have gone up a lot. I remember the 128GB mobo was $1700 or so. Now it's $2429.
Wonderful thing happened. I am checking from UAE and it showed me CAD prices by default!Not sure where you got the current price from, but framework webpage still lists 1699$ for bare mobo with 128GB.
https://videocardz.com/pixel/hp-zbo...e-for-preorder-prices-range-from-2240-to-3950 ignoring the price for the moment, what is the point of selling Halo chips with 16GB soldered unified RAM in a workstation laptop advertised for productivity. I think 32GB SKU should be the minimum one, unless 16GB ones exist only to up-sell the others.
It is. That does not change the fact 16GB shared with gpu is too little for any dev work imo. The chip supports up to 64GB.Isn't the 380 cut down to 128-bit memory?
I feel like there are some issues with the way this article is presenting this. All she said was “yes”. You can already buy such a mobo from framework.AMD CEO says Strix Halo will arrive on desktops soon
In a recent interview, AMD CEO Dr Lisa Su says AMD plans on bringing Strix Halo to more desktop parts. So far, it is only available in a handful of SKUs from companies like Minisforum and Frameworks.www.notebookcheck.net
If AMD starts selling motherboards with Strix Halo integrated for DIY, it certainly will open an interesting new market.
negative amount of bandwidthSeeing the wild consumer interest, AMD may be planning to release Strix Halo with quad channel DIMM slots
What if they got EXPO DDR5-8000 working on those slots?negative amount of bandwidth
yes gamer stable slop. genius.What if they got EXPO DDR5-8000 working on those slots?
Gamers are not gonna plop money on a 1TB machine. At least not the majority of them. It will be geared towards creators and LLM enthusiasts/researchers. And the stability problem is an engineering problem which they will eventually solve if they really want to make it happen.How the hell are you gonna run 1TB of gamer stable RDIMMs?
Isn't the problem solved with a soldering gun? if they could provide expandable memory, they would. Not to mention 1 TB memory at 256GB/s is not that attractive.Gamers are not gonna plop money on a 1TB machine. At least not the majority of them. It will be geared towards creators and LLM enthusiasts/researchers. And the stability problem is an engineering problem which they will eventually solve if they really want to make it happen.
More attractive than having to buy a fleet of 4090 or 5090 GPUs.Not to mention 1 TB memory at 256GB/s is not that attractive.
no it's not.More attractive than having to buy a fleet of 4090 or 5090 GPUs.
That’s like saying the current president can become smart if he tries. Nope, somethings cannot be done.Gamers are not gonna plop money on a 1TB machine. At least not the majority of them. It will be geared towards creators and LLM enthusiasts/researchers. And the stability problem is an engineering problem which they will eventually solve if they really want to make it happen.
Is that Llamabench speed? Or something else?9950x does around 4t/s for ~50gb models