-
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.
AMD VEGA confirmed for 2017 H1
- Thread starter Doom112
- Start date
- Status
Just a tad guessing or using random info that hasn't been proven i.e fanboy talkAmd had a chance to bring something new out for the start of the year but looks like they can't manage it yet.
Personally not bothered gave up on AMD a long long time ago when got burned on a ATI Rage Fury MAXX lol.
burned on a ATI Rage Fury MAXX
Me too ..and there's me thinking i was the only one ....Damn card....
So Vega is a huge chip, and if its true that it can process 2x the triangles/clock than Pascal TX than it is really more like Volta competitor, it will be faster than the Pascal cards in theory (ofc it depends on clock speeds, etc)
It looks whule NV opted for the Pascal which is more of a Maxwell refresh, AMD opted to pull off a year with just Polaris, and do a huge change.
Most articles call it the biggest change since GCN arrived.
It looks whule NV opted for the Pascal which is more of a Maxwell refresh, AMD opted to pull off a year with just Polaris, and do a huge change.
Most articles call it the biggest change since GCN arrived.
I was expecting it to beat Titan XP and be looking to compete with Volta back in August once we found out Vega's release date, so nothing has changed for me. I will have been disappointed if it did not do that anyway.
But at least I explained how I came to that conclusion. Flopper just comes out with funny random things
Nearly as funny to me as those who think big Vega will just be competing with a 1070 and not be able to provide 1080 performance
Seriously those who actually think that Vega will not be able to compete with a 1080 is just deluded. I'm sure that if AMD created a 4096 shader Polaris card with GDDR5X it would be able to compete with the GTX 1080, in DX12 games at least.
I think Vega will be around 1080Ti performance at launch and then reach Titan XP levels of performance later on with driver optimisations. Developers will need to be given time to properly optimise for Vega's new architecture, especially in DX12 games.
Well if the clocks are good, and the changes are working, then its really more like Volta competitor.
But lets just wait and see. AMD is trying to do one big step now, instead of two small, we shall see if they succed. If it works for them and it is beating the top Pascal then the one year on Polaris pays off, as from tge chaser (NV releases, then AMD cathes up) they will become the leader (or at least it will be a status quo in terms of they beat each others with every generation)
Last edited:
If it's a Volta competitor I am hella glad I skipped Pascal. Paying through the nose for a year with no competition.
Nah. I am glad I got my card when I did. I expect to make very little loss when selling it. By the time Vega is out I will have had 9-12 month of fun with my 1070 for a two figure sum, which is nothing really. I have seen people here who were late to sell their cards by a few weeks and lost on around £500 due to devaluation
Associate
- Joined
- 11 Jul 2016
- Posts
- 138
"Meanwhile, a since-deleted leak in the Ashes of the Singularity database in early December showed a GPU with the Device ID “687F:C1” surpassing many GTX 1080s in benchmark results. Here’s the twist: The Device ID shown in the frame rate overlay during AMD’s recent Vega preview with Doom confirmed that Vega 10 is indeed 687F:C1."
http://www.pcworld.com/article/3153...w-about-amds-cutting-edge-graphics-cards.html
These guys seem to think it's 10, big Vega is 11 no?
Soldato
- Joined
- 28 May 2007
- Posts
- 10,204
I heard the naming scheme is whatever comes first like with Polaris the big chip came first so was 10 and the smaller 11. If small Vega is coming first it will be 10 and the big version 11. I can't remember where i read it.
Associate
- Joined
- 11 Jul 2016
- Posts
- 138
I'm under the same impression RealDeal, pretty sure I read the same but can't remember where
Haha, that would be truly fun to see if actually AMD revealed that all along their demo Vega card was the smaller Vega, like when they first showed off Polaris 11 rather than Polaris 10.
If smaller Vega could match the GTX 1080 then Nvidia will be in trouble without any Volta cards this year. Even refreshed 14nm Titan X with 2.0 GHz stock clock priced at £600 might not save them.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
People are kind of missing this point, this isn't late to compete with Pascal, they've skipped a new architecture for 2016, to bring out a Volta competitor earlier. They will now beat Volta to market by some margin and with more time spent making a HBM2 specific architecture.
Some things I mentioned before that got ignored by most.
HBM wasn't used for architectural OR bandwidth reasons on Fury X, it was used both to prove HBM1, to get production going, to show it can work, to sort out the supply chain for production of interposer chips and to get some personal feedback on how it works. This was all stuff that was effectively game changing for the industry and AMD in particular. If HBM failed, if the supply chain wasn't working, if anyone screwed up and the chips couldn't be produced then it sets back the introduction of interposers and the like by a few years. It will also take a couple years anyway to go from small scale production of one line of interposer/gpu/memory combining to a fully fledged facility that can handle the entire product stack from not just AMD but many many more customers. So it was always going to take time going from HBM Fury X in low volume to midrange + high end from AMD + entire product ranges from multiple other companies who start using HBM2.
Fury X didn't need 512GB/s nor did it use 512GB/s, internally it seemed limited to around 360-380GB/s. Still the 4GB of HBM1 likely saved around 30W over doing gddr5 at 380GB/s which made it a cooler and better card all around. EDIT_ for the record there is quite a lot of discussion going on over testing 1080/1080ti cards and finding they have no where near the full bandwidth available to them either. I'm just pointing out, internal bandwidth and architecture is the important part, not the actual number being supplied in theory by the chips externally.
Read up on chip architecture, bandwidth is pretty much the fundamental design choice that decides a chip design. Everything else is done to work within a specific bandwidth level. That means more cache, more logic, more prediction etc on a cpu to get more cpu performance out of that amount of bandwidth.
HBM2 is a complete game changer, entirely, but NOT for Fury X because the architecture was not designed for that amount of bandwidth.
I also said this next point at the time, the difficulty of hitting HBM2 is you have an existing architecture that is based around relatively low bandwidth, with it's peak efficiency probably around 250-300GB/s as the 7970/290x were designed around. If you go HBM1-2 as they did, you want an architecture designed to use 512GB/s.... but if you design that architecture, it won't scale down particularly well to use gddr5 except on significantly lower chips.
This meant AMD essentially committed to the next big architecture(Vega) being a completely different architecture designed around 512GB/s bandwidth, or like 50-60% more than GDDR5 based architectures were designed for.
They could have made an entirely new architecture based on GDDr5 for 2016, then a completely new architecture again for 2017 when HBM2 was ready in large volume.... or they could just save their cash, put out a minor update to GCN(with some of that work having been done for consoles anyway, so cheap) for 2016 to make a good lower price card and rather than spend a couple years on a short lived architecture for 2016, focus on their next 'next' gen architecture for 2017.
Yeah, AMD conceded the high end for 8-9 months, but that saves time wasting time and money releasing what is a dead end inbetween architecture and just focus on a design based around much higher bandwidth. Half the slides focus on HBM2 and a revolutionary cache hierarchy in the new architecture.
Last edited:
Permabanned
- Joined
- 23 Apr 2014
- Posts
- 23,552
- Location
- Hertfordshire
Even refreshed 14nm Titan X with 2.0 GHz stock clock priced at £600 might not save them.
Save them from what?
- Status