xentr_theme_editor

  • Please do not post any links until you have 3 posts as they will automatically be rejected to prevent SPAM. Many words are also blocked due to being used in SPAM Messages. Thanks!

Intel B60 and B50 Arc Pro GPU

moocow

Well-known member
Joined
Aug 8, 2011
Messages
5,804
Reaction score
2,205
Location
Vancouver, BC
xentr_thread_starter
Just before going to sleep at 3AM, I saw GN's video tearing down a Maxsun B60 x2 card with a total of 48GB of VRAM. According to GN Steve (per what Maxsun told him), each GPU is PCI-E x8 with 24GB of VRAM. But no idea how exactly the bridging works or how it would function in apps that only support 1 GPU like Invoke AI. The target price is at or under $1000 USD (huge rock of salt required).


Reading Videocardz's article, Intel actually clarified the specs (which Videocardz originally reported as a leak or rumor):
  • B60 is 24GB of VRAM, Gen 5 x8, $500 USD
  • B50 is 16GB of VRAM, Gen 5 x8, $300 USD
  • Project Battlematrix, linking up to 8 Arc Pro in a system to enable total of 192GB of VRAM
The cards are initially available by SI and may get a DIY launch in Q4. I'm curious about how this card stack up against AMD AI+ 395 with 96GB of memory allocated to the iGPU.

 
xentr_thread_starter
So I did more research on bifurcation of PCI-E lanes. In order to use this card, you really need server level hardware that can do 16x to 8x, 8x in the same slot. As far as I know, consumer level motherboard can't do this and higher end Z and X boards are 1x16 slot or split to 2x8 slots.
 
So I did more research on bifurcation of PCI-E lanes. In order to use this card, you really need server level hardware that can do 16x to 8x, 8x in the same slot. As far as I know, consumer level motherboard can't do this and higher end Z and X boards are 1x16 slot or split to 2x8 slots.
well, crap
 
xentr_thread_starter
We still don't have enough data on the 395+ to be honest. I'm secretly hoping the 395+ would out do a 4090 or 5090 in LLM and Stable Diffusion.
 
xentr_thread_starter
Bit of a thread revive. Sparkle also announced a passive B60 and a dual B60 like Maxsun. Still no pricing. Digging a bit deeper, some board seem to support bifurcation but only in terms of GPU+NVME (8x+4x) or 4x NVME (one of those NVME adapter cards) and not true 8x 8x GPU mode. So a server board seem like to easiest way to use the 2x B60 card.

 
When I seen the Maxsun dual gpu ARC card in that video I contacted Maxsun via email and they got back to me.

They told me that you could use a normal motherboard but the problem is to get the most speed out of it you need to use that card in an Intel system that has a a cpu that supports Deep Link.

For example a single gpu A Series or B Series ARC card being run with a Deep Link supported cpu will get you a 20 or 25% boost in video encoding and also certain workload stuff if the Intel cpu being used has ARC graphic's as well,the cpu's gpu and the discrete card tag team the task at hand.

From what I recall what Maxsun said about their dual gpu card in the email is that you need a cpu that use's Deep Link but the cpu's gpu does not do anything in helping the task getting completed faster,only the 2 gpu's on the dual Maxsun card work together via Deep Link to complete the task much faster.

If I still had the email I would copy and paste it here but as soon as I heard your cpu has to have Deep Link I said well I guess I ain't buying one.
 
xentr_thread_starter
OK, I went and looked up DeepLink and it's apparently a dead feature. So how is Maxsun going to support the B60 Pro dual GPU then?

I forget when Deep Link started but you'd have to use a cpu that has Deep Link and the new Core Ultra stuff has it for sure and I know some cpu's before Core Ultra have I'm just not sure which ones.

From what I read Deep Link is built into the cpu from what I got because there is no Deep Link downloads for the software which is nice that way if a guy wanted to buy 2 or 3 mobo's and cpu's as back up's for heavy encoding task Deep Link could only be taken away if your cpu died.

For the Maxsun you'll still need a cpu that has Deep Link,the difference is that your ARC graphic's in the cpu won't work with the dual gpu card to help as Deep Link can only handle 2 gpu's at a times whether it be 2 gpu's on a discrete card or a discrete card with 1 gpu and then the cpu's gpu helps as well in that instance.
 

Latest posts

Back
Top