Soldato
Agreed
If they can pull it out the bag I may be tempted to switch back to AMD for the first time in years
-
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 7nm GPU News and Rumours 2018/2019
- Thread starter Beren
- Start date
Agreed
If they can pull it out the bag I may be tempted to switch back to AMD for the first time in years
Soldato
I think the fact AMD have waited until seeing 20xx card perf and then announced this, suggests they could/should be competitive.
Unless I'm mistaken this is just the same old news about Vega 20 for HPC use starting to ship end of the year, volume Q1/Q2 next year. This has nothing to do with gaming.
The photo Su is holding up with the vega 20 shows 4 stacks of HBM2, a consumer part would only use 2 like Vega 10.
The photo Su is holding up with the vega 20 shows 4 stacks of HBM2, a consumer part would only use 2 like Vega 10.
Caporegime
Waited? I think they have nothing really (hence why Raja left) and will be hoping to have something out to compete ASAFP. I would love to see AMD competing, as well as Intel but I feel we are some time off from that unfortunately.
Six desktop graphics cards use HBM. Four of those use four stacks each.
Not sure what got people hyped. If AMD say poor "famous mathematician" don't let yourself become hysterical.
I love how the press grabs a rumour and circulates it so much it sounds legitimate...
Other than a random ChipHell post, which doesn't seem to actually say a lot - Has anyone heard anything about the 12nm Polaris refresh? If I were TSMC I would have been pitching hard for any new GPU release so it makes a reasonable amount of sense to me - although I thought this chip was going to come about 9 months ago.
There appears to be two guesses - a straight re-spin of the 480/580 Polaris to a 680? with 10% higher performance but the same ram. There is also a rumour that they might be using the Xbox1X layout which would give a few more CU's.
There was Rumour in June that a GDDR 6 card was coming, but I doubt it is this one - more likely that will be Navi in Q2 2019.
So, has anyone seen anything around that is not directly coming back to a single uncorroborated chiphell source about a new AMD card?
Other than a random ChipHell post, which doesn't seem to actually say a lot - Has anyone heard anything about the 12nm Polaris refresh? If I were TSMC I would have been pitching hard for any new GPU release so it makes a reasonable amount of sense to me - although I thought this chip was going to come about 9 months ago.
There appears to be two guesses - a straight re-spin of the 480/580 Polaris to a 680? with 10% higher performance but the same ram. There is also a rumour that they might be using the Xbox1X layout which would give a few more CU's.
There was Rumour in June that a GDDR 6 card was coming, but I doubt it is this one - more likely that will be Navi in Q2 2019.
So, has anyone seen anything around that is not directly coming back to a single uncorroborated chiphell source about a new AMD card?
It would be interesting to see a TSMC 12FF respin of Polaris - IMO a good bit of potential performance was lost using GF 14nm never mind any optimisations from a newer node.
Soldato
I suggested on another thread that Polaris 30 on GloFo's 12nm process is largely a box-ticking exercise to fulfil their wafer agreement with them. AMD send a relatively cheap and token project to GloFo to uphold their legal minimum and get their toes wet implementing GDDR6 ready for Navi Q3 next year.
Associate
- Joined
- 29 Jun 2016
- Posts
- 541
+1
AMD appeared to do a similar thing with Zen, focusing on a clean slate design instead of constantly trying to adapt designs that were never intended for the purpose. Don't get me wrong, the incremental improvements of Bulldozer did end up having quite big gains, but the changes were small for the amount of time that passed.
Taking Lisa Su's words about applying the Zen approach to GPU's, it makes perfect sense they will hold the middle/lower tiers of graphics with Polaris, and attack the higher end and feature rich segment (RTX, VR[LMS, etc]) in later 2019 when they can do so in an economically sensible way.
I have no doubt she is not happy with the returns of Vega, as it must be really tricky to manufacture with the silicon interposer and problematic HBM.
I also have no doubt HBM is the future, where it just needs several evolutions to iron out the kinks in manufacture. My reasoning is that DDR6 is really pushing the boundaries of copper/PCBs. Eventually all the clever tricks we use to pre-emphasize, train links running at 10+GHz will not be sufficient.
A Polaris refresh wouldn't surprise me at all, particularly as GloFo are no longer pushing 7nm instead focusing on volumes of 14nm/12nm. Makes sense for GloFo if they can be 100% loaded with more profitable processes.
Soldato
Pretty sure Lisa would be delighted with the absolute killing they made selling RX Vega to miners, and rather thankful of that particular stroke of luck. And I'm not sure where you get the idea that HBM is "problematic" or Vega is difficult to produce. It's inherently more complex to produce yes, and there's little chance of salvaging a part if something goes wrong, but increased complexity doesn't necessarily equate to increased difficulty. Plus it's not like HBM exists on a small run of a single, niche product so the production processes are unrefined.
Don
It's unlikely to be anything Xbox1X "Scorpio" based as that was a TSMC 16nm part. Although it would be interesting to see a "big" shader count 384-bit bus part again (xbox1x has 2560 usable CU from a 2816 CU part vs 2304 for the RX580)
https://www.techpowerup.com/gpu-specs/xbox-one-x-gpu.c2977
https://en.wikichip.org/wiki/microsoft/scorpio_engine
Last edited:
What I want to happen:
March 2019 - AMD launch Navi GPU - flagship model performs as good as a 2080ti in rasterized graphics. Costs £599. Doesnt have any raytracing nonsense.
AMD win back big GPU market share over next two years. Nvidia forced to reduce prices late 2019.
2020-beyond - AMD launch new GPU with raytracing support when its actually viable (ie, not 1080p/30) Each cost over a grand for flagship models!
What I dont want to happen:
No Navi till Q4 2019. Flagship performance barely equal to a 2080. Terrible raytracing support. Similar price. Nvidia laughing all the way to the bank?
March 2019 - AMD launch Navi GPU - flagship model performs as good as a 2080ti in rasterized graphics. Costs £599. Doesnt have any raytracing nonsense.
AMD win back big GPU market share over next two years. Nvidia forced to reduce prices late 2019.
2020-beyond - AMD launch new GPU with raytracing support when its actually viable (ie, not 1080p/30) Each cost over a grand for flagship models!
What I dont want to happen:
No Navi till Q4 2019. Flagship performance barely equal to a 2080. Terrible raytracing support. Similar price. Nvidia laughing all the way to the bank?
Sadly I suspect its more likely to be the latter than the former.
Don
"Arcturus to succeed Navi"
https://videocardz.com/78367/amd-arcturus-to-succeed-navi
Also more importantly, codenames won't be reused as product names ala Vega
https://videocardz.com/78367/amd-arcturus-to-succeed-navi
Also more importantly, codenames won't be reused as product names ala Vega
Soldato
What you don't want to happen, @Calo, is partly what IS going to happen. But Navi isn't going to have dedicated RT, it's too soon. Arcturus maybe.
And even if Navi can only match RTX2080 in core performance, I don't see that as a given Nvidia will be "laughing all the way to the bank", because nobody is buying the 2080, and AMD won't price their top card the same as a competitor's 2nd tier that nobody is buying to begin with.
Here's my wishful thinking...
March 2019 - AMD launch RX600 series: 12nm Polaris 30 with GDDR6. All cards beat their direct 10 series equivalent by 15-20%, but cost 20% to 30% less. The 1080 Ti is still unchallenged directly, but the RX680 nips at its heels (the secret here is this would put a heavily overclocked RX680 within performance range of the RTX 2080 for about half the price).
This generates revenue in the mainstream and mid-range market, sweeping up market share. Nvidia won't drop prices, because Mvidia mindshare won't acknowledge the RTX 2080 doesn't offer that much more raster performance than the RX680 at 50% more expensive.
September 2019 - Navi 10 with HBM2 released for workstations
October 2019 - Navi 10 with GDDR6 released for gamers, beats the 1080 Ti (insultingly, lower Navi priced only £50 cheaper than 1080 Ti, upper Navi costs about £100 more than 1080 Ti, but that's still hundreds of pounds cheaper than the top RTX cards). Nvidia respond by lowering 1080 Ti prices, AMD counter with price drops of their own (the gouging on launch intentionally sets up the ability to fight a price war and still retain some profit margins).
March 2020 - "Big" Navi 20 with HBM2 hits data centers, succeeding Vega 20. MIGHT launch a Navi Ti/Navi Titan esque thing with GDDR6 for gamers.
September 2020 - Arcturus launches with its own 1st generation RT cores and MLSS (Machine Learning Super Sampling
)
And even if Navi can only match RTX2080 in core performance, I don't see that as a given Nvidia will be "laughing all the way to the bank", because nobody is buying the 2080, and AMD won't price their top card the same as a competitor's 2nd tier that nobody is buying to begin with.
Here's my wishful thinking...
March 2019 - AMD launch RX600 series: 12nm Polaris 30 with GDDR6. All cards beat their direct 10 series equivalent by 15-20%, but cost 20% to 30% less. The 1080 Ti is still unchallenged directly, but the RX680 nips at its heels (the secret here is this would put a heavily overclocked RX680 within performance range of the RTX 2080 for about half the price).
This generates revenue in the mainstream and mid-range market, sweeping up market share. Nvidia won't drop prices, because Mvidia mindshare won't acknowledge the RTX 2080 doesn't offer that much more raster performance than the RX680 at 50% more expensive.
September 2019 - Navi 10 with HBM2 released for workstations
October 2019 - Navi 10 with GDDR6 released for gamers, beats the 1080 Ti (insultingly, lower Navi priced only £50 cheaper than 1080 Ti, upper Navi costs about £100 more than 1080 Ti, but that's still hundreds of pounds cheaper than the top RTX cards). Nvidia respond by lowering 1080 Ti prices, AMD counter with price drops of their own (the gouging on launch intentionally sets up the ability to fight a price war and still retain some profit margins).
March 2020 - "Big" Navi 20 with HBM2 hits data centers, succeeding Vega 20. MIGHT launch a Navi Ti/Navi Titan esque thing with GDDR6 for gamers.
September 2020 - Arcturus launches with its own 1st generation RT cores and MLSS (Machine Learning Super Sampling
)
Last edited:
This probably can happen.
September 2019 - Navi 10 with HBM2 released for workstations
October 2019 - Navi 10 with GDDR6 released for gamers, beats the 1080 Ti (insultingly, lower Navi priced only £50 cheaper than 1080 Ti, upper Navi costs about £100 more than 1080 Ti, but that's still hundreds of pounds cheaper than the top RTX cards). Nvidia respond by lowering 1080 Ti prices, AMD counter with price drops of their own (the gouging on launch intentionally sets up the ability to fight a price war and still retain some profit margins).
March 2020 - "Big" Navi 20 with HBM2 hits data centers, succeeding Vega 20. MIGHT launch a Navi Ti/Navi Titan esque thing with GDDR6 for gamers.
September 2020 - Arcturus launches with its own 1st generation RT cores and MLSS (Machine Learning Super Sampling
These no.. Navi 10 with HBM2 and GDDR6 support won't happen. And probably it will be HBM3..
If the "RX780" Navi card lands at a GTX 1080 level and sells for £300, and they launch it next summer/autumn, that would be a smash hit and what I'm hoping for. Certainly not expecting high-end Navi next year from the rumours so far, which is a bit sad but eh, whatcha gonna do. It would also make sense on another level if you consider that for consoles, currently an RX 480 gets you on the highest console level with ease. The speculation is that PS5/etc will be around a V64 level, so an "RX 780" that would be equivalent to that would fit right in.
To me the most important timing is if they launch near or before CP2077 release as that's what a lot of people are waiting for, especially to upgrade.
To me the most important timing is if they launch near or before CP2077 release as that's what a lot of people are waiting for, especially to upgrade.
It doesn't?
When doesn't it?
Also I very much doubt a Polaris card can get anywhere near the 1080Ti. If it were possible they would have done it before now. The difference in process from "14nm" to "12nm" doesn't suddenly give Polaris enough headroom to gain 50% more perf.
You yourself say 15-20% gains, but you must surely know that wouldn't get a 580 anywhere near a 1080 Ti. It may be beating a 1060 today, but the difference between a 1080 and a 1080 Ti is >30% as it is...
Last edited: