Jump to content

AMD Vega Roadmap


Bokishi

Recommended Posts

My guess is that AMD's insistence on using HBM has hurt it, at least in the short term.  So far, Nvidia has eschewed using HBM in its consumer-grade GPUs and it hasn't cost them at all.  As of right now, the benefits of HBM for gaming seem to be nonexistent, GDDR5 is still doing just fine.  That may, of course, change in the future.

sky_twister_suzu.gif.bca4b31c6a14735a9a4b5a279a428774.gif
🇺🇸RFK Jr 2024🇺🇸

"Any organization created out of fear must create fear to survive." - Bill Hicks

Link to comment
Share on other sites

AMD has a stake in HBM, iirc, which is why they use it. Theoretically its benefit is being able to use less of it for the same effect which ought to use less power and be cheaper even if it's more expensive per GB. They've got stung because it's new, and the main manufacturer (Hynix) hasn't been able to produce the chips they promised so they're using slower chips and trying to make up the performance in other ways- presumably, from the power draw and water cooling, overvolting. If they can get the memory up to speed Vega RXI could get a decent 'free' performance boost and power draw drop at the same time which would help competitiveness, but that's all conjecture and won't help now.

Link to comment
Share on other sites

I think HBM is the future, it's just not necessary yet.  The extra bandwith, for gaming applications, is overkill at the moment, GDDR5's bandwith is still plenty.  At some point, GDDR5 won't be enough, that point is not here yet.  Hopefully the HBM gambit pans out for AMD in the long run and they catch up to Nvidia.  As a consumer, I want the 2 companies as competitive as possible across the board.  I guess the best bit of good news for AMD is consumer-grade Volta getting delayed until 2018, as I expect Volta to mop the floor with everything that's out today.

Edited by Keyrock

sky_twister_suzu.gif.bca4b31c6a14735a9a4b5a279a428774.gif
🇺🇸RFK Jr 2024🇺🇸

"Any organization created out of fear must create fear to survive." - Bill Hicks

Link to comment
Share on other sites

My guess is that AMD's insistence on using HBM has hurt it, at least in the short term.  So far, Nvidia has eschewed using HBM in its consumer-grade GPUs and it hasn't cost them at all.  As of right now, the benefits of HBM for gaming seem to be nonexistent, GDDR5 is still doing just fine.  That may, of course, change in the future.

this is why i call it a tech demo. the features in this card go unused by today's games but they offer solutions to eternal limitations of game design (HBM most of all) so if they convince developers to take advantage of these features, you could see a big leap in the future. 

The words freedom and liberty, are diminishing the true meaning of the abstract concept they try to explain. The true nature of freedom is such, that the human mind is unable to comprehend it, so we make a cage and name it freedom in order to give a tangible meaning to what we dont understand, just as our ancestors made gods like Thor or Zeus to explain thunder.

 

-Teknoman2-

What? You thought it was a quote from some well known wise guy from the past?

 

Stupidity leads to willful ignorance - willful ignorance leads to hope - hope leads to sex - and that is how a new generation of fools is born!


We are hardcore role players... When we go to bed with a girl, we roll a D20 to see if we hit the target and a D6 to see how much penetration damage we did.

 

Modern democracy is: the sheep voting for which dog will be the shepherd's right hand.

Link to comment
Share on other sites

  • 3 months later...

Still no custom Vega cards released, but on the other hand the reference v56 are available here for a decent price at 750NZD and ~80NZD less than a 1070Ti. Since I can still sell my 580 for a decent amount a v56 card and a new monitor are on the menu for Christmas.

 

Only drawback is that 1440p is probably the natural res for a v56 card but we only have have 1 IPS freesync monitor available in the whole country apparently.

Link to comment
Share on other sites

Yeah, I'm getting a bit annoyed by that, was hoping they'd release some form of custom card. I managed to get a hold of a R9 Fury for now wich does the job well enough, but I really had wanted a Vega.

Civilization, in fact, grows more and more maudlin and hysterical; especially under democracy it tends to degenerate into a mere combat of crazes; the whole aim of practical politics is to keep the populace alarmed (and hence clamorous to be led to safety) by menacing it with an endless series of hobgoblins, most of them imaginary. - H.L. Mencken

Link to comment
Share on other sites

  • 4 months later...

Vega 20* cards have appeared in some Linux certification, so it looks liker there will be a Vega refresh this year despite it not being announced. That should at least help with efficiency, and that may indirectly help with decently clocked memory.

 

Plus eth is collapsing, so maybe I'll be able to pick up an actual card some time for less than the current $1100 asking price. Ironically Vega has probably been one of the most successful chips ever despite its flaws thanks to mining, it's basically never been in stock since launch nearly a year ago.

 

*AMD's naming conventions really are stupid, since they have Vega8, 11, 24 (Intel), 56, 64 as consumer products which are all based on Vega10 chips '14nm'; with Vega 20 being the 'tock' '7nm' die shrunk chip to Vega10's 'tick'. Ryzen naming is about as consistent.

Link to comment
Share on other sites

Yeah, consumer products aren't confirmed, but...

 

That pro/ machine learning Vega20 was announced in January. Die shrunk Vega consumer cards actually were mentioned before that, they simply haven't been mentioned since. To put it in perspective, most people expect a new nVidia series to drop in the next month or so, and it hasn't been formally announced either and there were pro versions of Vega10 before the consumer version release as well, given '7nm' is a new process and the margins for pro level cards that's an almost inevitable approach.

 

Advantages to Vega of a die shrink are far too great to pass up since it would fix nearly all its problems bar some associated with HBM; and even there reduced GPU heat will help memory clock a lot (indeed, undervolting the gpu to reduce memory throttling is perhaps the most common 'overclock' approach for Vega). Unless Navi is far further along than thought- possible, but the only 'evidence' for it is the inconsistent and speculative ps5 leaks- they'd be stuck with Vega 10/ Polaris chips for maybe the next 18 months.

Link to comment
Share on other sites

  • 9 months later...

'Consumer' 7nm Vega 2 announced at CES for Feb release. It's practically a Vega Titan as it's clearly a professional card that can be used for gaming rather than a specialist gaming card and is basically a rebadged/ low binned Instinct card. 16 GB HBM2 (lol), 60 compute units, performance about the same as a RTX2080/ 1080Ti (though significantly better in Vulkan and other heavily parallelised tasks). 700USD MSRP so it's price competitive with 2080s, but I doubt it's enough so to get much traction.

 

The 16GB of HBM2 is a shame, may be necessary for bandwidth but that's a lot of cost and complete overkill for gaming. An 8GB card with 150 off the price tag would be more competitive assuming that didn't also tank performance. An excellent release for pro types wanting a cheap but powerful option, good for Linux or if Vulkan takes off a bit more; not compelling enough otherwise.

Link to comment
Share on other sites

I think if they could make a 8GB version of the Radeon VII they would have. They may have needed to deactivate compute units to do it and it would kill the performance too much. This is a card that's been designed for work not gaming. I also have a feeling the benchmarks are highly selective, being just 3, at one resolution, at max setting, only showing averages. The launch date is what really killed this card for gaming, 2080 will have been available for a few months, more games will have RTX and DLSS.

 

I'm not sure AMD will target the high end at all. They may just put out really competitive options in rasterization targetting the 2060 and 2070.

Link to comment
Share on other sites

Yeah, with the current design 8GB of HBM2 isn't really feasible. 4 x 2GB stacks at the same bandwidth would be great but they simply don't exist, and 2 x 4GB stacks halves bandwidth. They'd really need a GDDR5x/6 option which is probably impractical for what looks like a card produced because... how else do you sell the imperfect Instinct cards? The full 64 CUs would help too as that would give it longevity/ some 8k use from the memory and a bit of performance margin over the 2080 at significantly less than a 2080Ti.

 

I wouldn't be overly surprised to see 'Vega 3' 56/64 with GDDR memory once the 7nm yields get a bit better and they get fewer defective Instincts and if there's demand, but I'm a lot less sure of that than I was that there would be a consumer 7nm card. Depends a lot on Navi's performance and how close the successor to GCN is.

Edited by Zoraptor
Link to comment
Share on other sites

  • 4 months later...

The new 7nm chips will be using the new RDNA architecture rather than GCN and will be using GDDR6 memory rather than HBM. They will support PCIE4.0. They showed a demo of an unnamed Radeon RX5700 series card beating a Nvidia RTX2700 running Strange Brigade. I don't put any stock in benchmarks from the manufacturer themselves, but it's telling that they targeted the RTX2700 and not the flagship RTX2800. I would assume, since it's a new architecture, that these are Navi and not a Vega refresh, but I can't say for sure.

Edited by Keyrock
  • Like 1

sky_twister_suzu.gif.bca4b31c6a14735a9a4b5a279a428774.gif
🇺🇸RFK Jr 2024🇺🇸

"Any organization created out of fear must create fear to survive." - Bill Hicks

Link to comment
Share on other sites

Exactly how different RDNA is from GCN is a bit of an open question up until we get a white paper but it won't be a complete departure from its predecessor. The 5700 is definitely Navi and there's decent presumed room in the specs for a 5800 and probably a 5900 later on. With no news on pricing and one benchmark it's difficult to evaluate how the 5700 will do. The Vega refresh was Vega 20 so consumer wise Radeon VII only.

(nVidia's flagship is the 2080Ti, though it is ludicrously expensive compared to previous flagship cards)

Link to comment
Share on other sites

  • 2 weeks later...

Full announcement for Navi at E3. Not exactly earth shattering, but also not awful Not great, not terrible. 5700/XT with performance a bit above the 2060/70 respectively (and a fair few Gimpworks titles were benchmarked, not just Strange Brigade), pricing at 379/449USD so equivalent to or slightly below their nVidia competitor's formal pricing. No mention of raytracing, unlike the MS conference and Scarlett yesterday.

Not much for people hoping for prices to be forced down on nVidia cards though, except perhaps the prospect of 5800/5900 later in the year.

Edited by Zoraptor
Link to comment
Share on other sites

I would describe  this announcement as tepid. Here are a couple GPUs with similar performance and similarly priced to the mid to mid/high end our competitor has had out for months. Yay?

Granted, this is about what I expected. It will take a while for Team Red to catch back up to Team Green, assuming they do catch up.

  • Like 1

sky_twister_suzu.gif.bca4b31c6a14735a9a4b5a279a428774.gif
🇺🇸RFK Jr 2024🇺🇸

"Any organization created out of fear must create fear to survive." - Bill Hicks

Link to comment
Share on other sites

A 64 CU 5900 would have performance between a 2080 and 2080Ti if the scaling is good- and maybe Navi/ RDNA does not have the old 64 CU limit either, though I'd suspect it does personally. That card doesn't exist at the moment of course, but at least theoretically AMD should be competitive at the higher end again without marketing what is basically a defective/ low binned pro card like the Radeon VII.

I'd suspect that the '6000' series/ Navi+/2/20/ Arcturus or whatever AMD ends up calling it is the one most likely to be genuinely competitive though, as it would presumably have a lot of the console developed improvements added (raytracing and memory improvements).

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...