-
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.
- Status
And Volta is using four stacks so Samsung's chips being slower isn't bandwidth problem.
For Vega with two stacks bandwidth would be mediocre for high end card.
Different factories have also different process optimizations so amount of work in porting physical chip design made for one factory to another isn't small.
It's isn't anything like getting some piece of metal machined/stamped.
One hardware site once had article about how integrated circuits are designed and then fabricated and there are awfully lots of steps in it.
And process optimized for one type of circuitry might be very bad for other kind IC.
Associate
I would get a FuryX if it's less than £200, but as it already been mentioned, they are very rare, I just missed out on one on eBay that went for £210Bless ya.
I was talking about the 980Ti, no one wants a fury.
Underboss
Latest i hear is Jan 2018 for the mainstream ones and July 2017 for the enthusiast ones
Aye. Vega Frontier Edition end of June, Radeon RX Vega is July, and Vega 11 end of the year.
Underboss
ah right, Frontier edition is going to be the fastest i guess ?
Kinda, Maybe, Not quite, Raja specifically stated their will be a faster Radeon RX version, that's better optimized for gaming ( drivers ), and has extra "goodies" he can't talk about yet.
In gaming a Quadro is around 10% slower than it's GeForce version, most just down to drivers; so I would expect something similar with Frontier Edition. Ignoring these "extra goodies".
FE seems like a card between Radeon RX, and Instinct so similar to the Titan from NVIDIA; but unlike the NV card won't have it's compute and enterprise features cut compared to Teslas.
Reviewers getting their hands on the FE should certainly test it and give us some indication of how the gaming Radeon version will compare.
Not sure if it has changed but not that long ago AMD didn't have anything like as an advanced simulation lab for their GPUs and would still need to do a fair bit of software debugging with actual working silicon compared to nVidia.
RX Vega does have its compute performance cut, FP64 is only at 1:16 just like the gaming parts.
Only for FP64, Vega still have 2:1 FP16.
Both twice more than GeForce and Titan cards, that's what I meant. The same for other features, such as HBCC, and HBM2, and all the goodies on Instinct will also be present in the RX models.
NVIDIA doesn't keep everything from Tesla in the consumer gaming line.
Which is why I stated
Associate
- Joined
- 29 Jun 2016
- Posts
- 541
Not saying your logic is wrong, however you've missed about a million possible show stoppers that AMD could have encountered.
Just looking at one facet, such as EMC, could render over a dozen significant entries for a die respin. And that's not even a functional issue!
Anyhoo, in saying all this, your assessment that HBM is holding things up sounds reasonable to me.
FP64 performance is the most important aspect for HPC use, so the main defining feature of an HPC datacenter GPU is missing.
We don't know what half-precision performance consumer Volta will bring, a good chance they will also offer double-rate FP16 but we will just have to wait and se.
The whole Vega lineup uses HBM2 so its hard to see how AMD would do anything differently. It is already clear that only the datacenter cards will have the 16GB memory, so there is a big difference.
The HBCC is the one area that I think is more orientated towards compute rather than gaming, I'm not fully convinced it will offer that much for games that can keep all resources inside VRAM, but it is probably only something that will really shine in Navi for HPC purposes.
I get what you are saying that the FE is somewhere between consumer graphics and an HPC datacenter product, but to me it is much closer to the former than the later. Just like they shoved a Fiji, dual Fiji and Polaris chips into a professional solution but all compromised in some way (all of them with only 1:16 double-precision for starters).
Don
Just skimmed through, Vega Frontier Edition? What is it?
Edit: Ah bit of info above.
Edit: Ah bit of info above.
I get what you're saying, but only the Tesla P100, and Quadro GP100 have full FP64 either, even the lower end Teslas are lacking it, never mind the majority of the Quadro cards.
Most likely why AMD is marketing Instinct primarily for Deep Learning, and streaming solutions.
They do list double precision rate is configurable. So I wonder why why FP64 is stuck at 1/16
So the FE does sit well there, although I guess it's more in line with Quadro than, while still possibly having of Instincts Rapid Packed Math, and FP 16 performance.
AMD are advertising it for Data Scientists primarily on their conferences and website.
The Frontier has a HBCC and that's going to be launched later in the month?
Yes, the 27th of June.
I expect Kaapstad at OcUK HQ, 9am launch day to buy four.
This and we need the results in detail by 12PM.
Soldato
If slightly loopy and burdened with cash then yea great I like the idea but if it wont come close to a Titan what would be the point beyond uniqueness Im not sure. Really it comes down to having some kind of excuse professionally to pick one of these up and also do some gaming.
It might have been somewhere else Im remember someone mentioning their gf had a 4x 1080 machine at their workplace, because they spent the day altering photos for advertising or something. I know Quadro is supposed to be for that really but this AMD card now will actually be placed towards such people? If not paying VAT, I guess that also helps the price
- Status