Soldato
Turned into the usual willy waving, AMD bashing by the usual Nvidia suspects.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
** The AMD Navi Thread **
- Thread starter subbytna
- Start date
Caporegime
In truth, it is a bit too early for a Navi thread, as we only just have had Vega launched, so I can't see much info (of any use) about Navi coming up that is useful.
Fixed.
I take it Gregster is now in the focus group with the cool kids. We need forum tags. Could I have Grumpy OBD.
Please tell me Loadsamoney Jokester and Werewolf haven't joined.
Yes but AMD are sourcing 7nm from TSMC. Vega isn't going to drop on 7nm, its going to use a refined node that its currently on.
Soldato
- Joined
- 30 Nov 2011
- Posts
- 11,380
this news, from AMD, says AMD are sticking with glofo through 2018 for GPU's, which somewhat suggests that other rumours that they were going to be making Navi starting in early 2018 for mid 2018 launch on 7nm a bit tenuous
I didn't say Navi wasn't going to be on 7nm, just that if AMD are saying they are releasing glofo sourced Vega in 2018 then a 2018 launch for Navi looks a bit slim
So it seems Navi and Vega could run alongside each other. I really hope the improved Vega lowers the price or Navi will have to be priced around 1080Ti money. Probably more.
A £250-300 Vega or Navi with 1070 performance would be just the ticket right now.
Last edited:
Yes, I can't see them superseding Vega so quickly unless it's really costing AMD a lot and they are making little profit. They won't get their money back on Vega as it's still barely off the ground (not even any AIB products yet). The only way Navi is going to come before the end of 2018 is if it's more powerful and so much cheaper to make, that AMD would make a lot more money getting rid of Vega quickly and bringing out Navi as soon as possible.
Well it would make sense to get something to take sales from the T cards and have Vega and Polaris in the line up too.
Don
Indeed.
Will reopen or start afresh when there is something of interest.
We have more Vega cards to come first as I understand.
Deleted member 651465
D
Deleted member 651465
Reopened. Yay to me.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 31,034
Why? Still zero information, and this isn't expected for 15+ months...
Caporegime
Knarf
A modular Navi or not? If they can do it with threadripper why couldn't they do it with Navi? Probably a bit TOO far ahead of its time but I think GPU's will possibly go this way. Multi GPU but acting as one GPU.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 31,034
Not. Gfx chips are a multitude more complex than CPU cores, they'd have to improve a GREAT deal on Infinity Fabric in order to do something like that. It's being speculated it's more of a second-gen 7nm chip if anything...
Caporegime
Ryzen APU's are already using Infinity Fabric to connect the GPU to the CPU.
Strapping together current (or even tweaked versions of) monolithic cores even via something like IF just won't be any better than SLI/CF - will still have to use AFR/SFR to work and all the problems with that and still need a large dedicated amount of memory per core even if they find a way to pool some of it (very non-trivial).
Advances in tech means MCM GPU designs will become a reality but that won't be anything like current GPUs and even then need at the very minimum a properly working 7nm process and more dedicated interconnects between chips on the substrate than current forms of IF.
I might be wrong but I can't see either AMD or nVidia jumping from their current architectures to MCM type implementations as it needs a fairly significant overhaul of the architecture, some experience with how that works and then using that experience to optimally implement as separate chips.
There's also the matter of latency to consider too for a multi module GPU. I think it'll happen but perhaps not for another gen anyway. As Jedi and roff said, there's a LOT more work to do to overcome these issues.
as has been said above Multi chip via some sort of infinity fabric, just isn't practical with the architecture currently in use. Of course that doesn't stop either AMD or NVidia working out a new architecture which allows it to work, with out the massive latency issue it would incur with the currant system.
Caporegime
Unless they can make the MCM approach look like one piece of unified hardware at the software level,it will require a ton of software support to work properly out of the box.
Well yes even if they get the hardware side of things to work without some massive latency issues, the software side of things would also have to work completely seamlessly, else it would be a developer nightmare and wouldn't get taken up easily. ( much like DX12
)