Consumer 7nm GPUs from AMD are late 2019 to early 2020

Muhammed

Senior member
Jul 8, 2009
453
199
116
Speaking to AMD's CFO Devinder Kumar:
Q: And the absolute last question on the graphics side, 7 nanometer Vega coming to the data center side of it, you've talk about that before at the end of this year. When should we expect 7 nanometer to occur on the more traditional gaming…

A: We haven’t missed that piece. I think, if you look at it from what we have stated, we have 7 nanometer data center GPU launching later this year; we are sampling the 700 CPU this second half of ’18 and then launching in 2019; after that, we'll have the client piece of it; we haven’t been specific about the timing; and graphics will be coming out later than these products.

https://www.reddit.com/r/Amd/comments/9frjoe/amd_cfo_devinder_kumar_on_7nm_products_schedule/

It will be like this: Vega 20 for HPC (late 2018) > Epyc Rome (2019) > Ryzen 3000 > Navi.

So late 2019, early 2020.
 
  • Like
Reactions: lightmanek

MrTeal

Diamond Member
Dec 7, 2003
3,554
1,658
136

ozzy702

Golden Member
Nov 1, 2011
1,151
530
136

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
Amd 2017 and 2018 gpu profit was saved by crypto.
Now datacenter for 2019.
7nm is too expensive for consumer gpu anyways so no need to make any drama about it.
Damn luck now they dont have a compettitive highend consumer arch.
The gpu execution the last 4 years have been very bad. For the last 2 years they were only saved by factors outside of their control. Now 2019 will save them too by factors they dont control.
It can only be a huge step forward with navi.
 
  • Like
Reactions: beginner99

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
At least they will have more 14/12nm capacity at GF for GPUs when ZEN 2 will be released sometime in 2019.
 

EXCellR8

Diamond Member
Sep 1, 2010
3,979
839
136
Is this around the time Intel is supposedly going to roll out discrete GPU's?

I don't have very high hopes for AMD's line of graphics after Vega honestly, but I appreciate being pleasantly surprised as much as the next guy.
 
  • Like
Reactions: psolord

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
I posted before that I'm expecting them to target holiday '19 (i.e. Nov - mid Dec) for Navi. Could slip to early 2020 though.
 

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
Is this around the time Intel is supposedly going to roll out discrete GPU's?

I don't have very high hopes for AMD's line of graphics after Vega honestly, but I appreciate being pleasantly surprised as much as the next guy.

Intel's targeting 2020, but I don't think they've narrowed it down any more than that. Not expecting much from their first try but we'll see.

Hopefully Navi ends up much more competitive than Vega. They've restructured their GPU R&D division and are able to give it some more funding, so we'll see if that pays off. Full effect of an effective restructure and increased budget probably wouldn't be seen until Navi2 or whatever comes after, but hopefully Navi gets some benefit as well.
 

EXCellR8

Diamond Member
Sep 1, 2010
3,979
839
136
I still think Vega are good cards if you're a hi-res gamer on FreeSync, but there were quite a few things that factored in to why they didn't make as big of a mark as was hoped. I won't fully dismiss Navi of course but now we have GDDR6 being introduced and who knows what else between now and 2020, HBM3??
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
I think it's unlikely but it's still possible AMD releases a Vega 20 gamer card, even if it's a Frontier Edition part.
 

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
At 7mm vega will sell truckloads on the b2b market. Efficient and a nice fit for the big players that can use the more open platform.
I think they will get very solid profit here.
The entire idea to sell limited and expensive 7nm capacity for consumers wrapped in a vega arch is crazy. Same as burning money. No need to wish for that. It will bring nothing good on the long run. We need a strong amd to restore compettition. Just look at cpu market. Its moved insane the last 1.5 year.
 
  • Like
Reactions: SteveGrabowski

prtskg

Senior member
Oct 26, 2015
261
94
101
How are you getting late 2019 or early 2020 from what is written there? Your ordering may be correct, but assigning timelines based on that is just pure speculation.
Where there is wish, there's way.
 
Mar 11, 2004
23,031
5,495
146
Yeah, OP, not sure where you're getting this from. Seeing as how Ryzen 2000 series was out mid-late April, that leaves a lot of room. Ryzen 2 seems ahead of the curve (although much of that will probably be tailored for EPYC markets early on), too so it potentially could be out earlier than the 2000 series was.

At least they will have more 14/12nm capacity at GF for GPUs when ZEN 2 will be released sometime in 2019.

For what? Even consumer GPU is going 7nm. Unless AMD starts selling RX 580s for like $100 (which I'd expect that AMD is probably going to go after the low, $150 and below market with another Polaris rebadge), I don't know what extra capacity they'd need. Its not gonna help Vega get cheap enough. I don't know maybe mobile Vega? But personally if I were AMD I'd be trying to get Ryzen 2 and bin Navi to put in laptops.

Intel's targeting 2020, but I don't think they've narrowed it down any more than that. Not expecting much from their first try but we'll see.

Hopefully Navi ends up much more competitive than Vega. They've restructured their GPU R&D division and are able to give it some more funding, so we'll see if that pays off. Full effect of an effective restructure and increased budget probably wouldn't be seen until Navi2 or whatever comes after, but hopefully Navi gets some benefit as well.

I have a hunch it'll be later 2020, and won't be surprised if it gets delayed, or for them to do something odd (like release cards to developers or something, but not sell them for awhile, just so they can get them up and working as best as they can before they sell them).

Sounds like they did that before Raja even left, so it'll be almost 2 years, I think it'll show some. I don't know that it'll matter too much in consumer dGPU (they supposedly were putting focus on Navi for the PS5). Will be interesting to see how Navi turns out and what AMD's GPU roadmap looks like going forward. I think they'll probably keep that pretty close to the vest until they're ready to get things moving (kinda like with Ryzen).

I still think Vega are good cards if you're a hi-res gamer on FreeSync, but there were quite a few things that factored in to why they didn't make as big of a mark as was hoped. I won't fully dismiss Navi of course but now we have GDDR6 being introduced and who knows what else between now and 2020, HBM3??

Consumer Vega was doomed by the HBM. It made it impossible to compete on price, and I think there's been some indication that it was partly to blame on other aspects (I see people able to undervolt the GPU a good amount, but voltage adjustment gets wonky apparently because of the HBM power). Plus going to 2048 from the 4096 of Fury really hurt (and I see people saying that Vega is likely pretty bandwidth constrained). I think there's possibly some other issues involved. It would've been nice if AMD had readied a GDDR5X version for the early part of this year. Especially if they could've maybe made some adjustment for 12nm (maybe if nothing else be able to be more efficient/clock). Although maybe for the best to just move on.

I think we see consumer cards go all GDDR6.

I think it's unlikely but it's still possible AMD releases a Vega 20 gamer card, even if it's a Frontier Edition part.

That would be nice, but I imagine it'd be so stupid expensive that even if it had competitive performance, it'd make the RTX cards seem like good value. A Frontier version might be intriguing though if the new HPC features are worthwhile, but even that would be pointless as it'd likely just take away actual HPC sales. But if they can price it decently, and the performance is there, I'd love to see it.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
For what? Even consumer GPU is going 7nm. Unless AMD starts selling RX 580s for like $100 (which I'd expect that AMD is probably going to go after the low, $150 and below market with another Polaris rebadge), I don't know what extra capacity they'd need. Its not gonna help Vega get cheap enough. I don't know maybe mobile Vega? But personally if I were AMD I'd be trying to get Ryzen 2 and bin Navi to put in laptops.

Desktop/Mobile APUs at 12nm (Picasso) and dGPUs.
7nm consumer products will not get high volume in 2019, the vast majority of the consumer products volume in 2019 will be 16/12nm for TSMC (NVIDIA) and 14/12nm from GF (AMD).
 

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
Sounds like they did that before Raja even left, so it'll be almost 2 years, I think it'll show some. I don't know that it'll matter too much in consumer dGPU (they supposedly were putting focus on Navi for the PS5). Will be interesting to see how Navi turns out and what AMD's GPU roadmap looks like going forward. I think they'll probably keep that pretty close to the vest until they're ready to get things moving (kinda like with Ryzen).

They restructured when Raja first came back and then restructured again when he left https://www.anandtech.com/show/12363/amd-reassembles-rtg-hires-new-leadership .

My understanding is that Raja had a particular vision for what RTG should be and wanted full control over the unit and a lot of autonomy when AMD hired him back from Apple. AMD obliged (this was before Lisa Su was appointed CEO). After Su became CEO there was a disagreement over how RTG should operate. Due to various factors, RTG was under-performing and AMD wanted a change in direction. Raja left, they restructured again. The increase in R&D spending only happened toward the very end of Raja's tenure at AMD after Zen was released and they knew they'd have more money rolling in.
 

Campy

Senior member
Jun 25, 2010
785
171
116
Consumer Vega was doomed by the HBM. It made it impossible to compete on price, and I think there's been some indication that it was partly to blame on other aspects (I see people able to undervolt the GPU a good amount, but voltage adjustment gets wonky apparently because of the HBM power). Plus going to 2048 from the 4096 of Fury really hurt (and I see people saying that Vega is likely pretty bandwidth constrained). I think there's possibly some other issues involved. It would've been nice if AMD had readied a GDDR5X version for the early part of this year. Especially if they could've maybe made some adjustment for 12nm (maybe if nothing else be able to be more efficient/clock). Although maybe for the best to just move on.

I think we see consumer cards go all GDDR6.

This got me thinking about a piece GN put out about a year ago https://www.gamersnexus.net/guides/3032-vega-56-cost-of-hbm2-and-necessity-to-use-it

AMD needed HBM because Vega is bandwidth hungry, and to keep power consumption reasonable. Unfortunately due in large part to the huge increases in memory pricing(the possible price fixing) HBM ended up being so expensive that AMD couldn't profit from Vega. 8GB of HBM costs about $150 for AMD and the interposer is another $25. $175 is a huge chunk of the Vega cards' retail price, and that's just for the memory solution. HBM2 is about 3x the price as the same amount of GDDR5. However it's not simple slapping GDDR5 on Vega, as that would require redesign work and ultimately might not even be feasible considering the bandwidth requirements of Vega.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
Late 2019, early/mid/late 2020 or even 2021 are just fine with me.
I'm not touching a new CPU or GPU until it's PCIE 4.0 and 7nm.
I center on computer hardware purchases approximately ever 5 years when something major happens and pricing is reasoned.
Same goes for smartphones and other electronics.
The bleeding edge is for 'take my money' consumers.
I am glad they exist as they subsidize the rest of us and keep many of us employed in the tech industry.

I'm currently sitting on a range of Pascal cards 1070s/1070tis/1080s I purchased in 2016,2017,2018.. 1080ti consumers too much power and was not worth the money. 5 Years of usage puts these GPUS out to 2021.
The only thing that causes me to purchase sooner than 5 year time horizons is if an incredible value product is released.
If there is no heightened value, I wait even longer than 5 years.

Even the enterprise that has a business case for being on the bleeding edge and billions of dollars sits on hardware for longer than the small but noisy crowd of 'take my money I don't care' consumers. So, it is interesting hearing the circles people spin themselves in to justify this habit. From an investment standpoint, it is interesting to hear the thought process of such consumers.. They are why insane premiums will always exist for 'hot new products' and the upper end. They simply don't care about price/value. Great investment returns for companies that have a good segment of consumers of such a mindset.

PCIE 4.0 / 7nm.
I have Ryzen procs from 2017 and Pascal GPUs from 2016/2017/2018. They both were amazing value/performance products and I didn't hesitate to grab them on launch. When I look back on my life, I can say that I seized great value/performance immediately and reaped the benefits until they were no longer to be had, sold the hw on ebay and a purchased a much better value/performance product. My hobbies include a range of activities centered on nature which are quite cheap. Although I work in tech, I consume far less than what I see super vocal group of consumers.

I'm set until 2021 but will open up my wallet if any corporations provides killer value sooner : Intel/AMD/Nvidia. I really don't care.

The latest Geforce20 is an absolute joke in my book. A small but vocal number of consumers disagree. I am fine with this.
AMD has a solid roadmap and execution plan. It's why they're stock has zoomed from $2 a share to $30+.
Take a $15-$20k position and you go to a present day : $225k to $300k.

This is how the game is played beyond consumption.

When you hear a consumer and markets speak, it's wise to listen.
Echos across the internet reflect that people by and large have no taste for the stupid priced GPUs Nvidia released.
The sentiment when Pascal was released which was an incredible value was much different.
When Ryzen was released, the widespread sentiment was praise as it was of value.

The consumer generally knows when they are getting taken advantage.
They also know when a great value is being provided.

I consume based on great values.
I invest when such value has a solid margin and is sold in volumes.
I invest when a company has consumers who just toss money at them mindlessly.

Working out great so far.
7nm / PCIE 4.0 for my development needs.
For goofing around, i'll likely be on older equipment
 
  • Like
Reactions: NIGELG

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
It's going to be a while, especially with the news that Global Foundries isn't doing 7 nm anymore, which means capacity is extremely limited. Anything made on 7 nm needs to be a high margin part due to limited supply and increased costs. There's no point in AMD selling a consumer GPU (that's potentially likely to get snatched up by a miner anyways) for $200 - $500, when it could be sold as a $1,000 - $2,000 professional GPU. They probably get even better margins from selling server CPUs due to being able to use multiple small dies instead of a monolithic monster die.

Hopefully AMD is using this time to come up with a new architecture. GCN was great, and the longevity of the 7000 series certainly shows how good it was. However, it's got the same kind of problem that Intel has run into where the low hanging fruit was picked clean years ago and it's hard to get significant improvements when you're refreshing the same design over and over again. Fury and Vega might have worked better if HBM were less expensive, but that's not how it turned out.
 

EXCellR8

Diamond Member
Sep 1, 2010
3,979
839
136
I'm not going to pretend like I know what it takes to run a company like AMD but I would imagine that brand new products, and/or the R&D probably cost quite a bit. There's also a fairly high level of risk involved with developing and introducing new products, especially in this day in age in technology. AMD also requires teams that work on both CPU and GPU markets whereas Intel and NVIDIA specialize in one or the other.

They did a a decent job with Vega, despite the shortcomings, but in order to make something revolutionary that actually gets bought up in droves requires some very intricate planning and possibly even a bit of luck with trending product lines. I'm sure they know what they're doing though and will release a good successor to the Vegas in a reasonable amount of time.
 
  • Like
Reactions: ryan20fun

maddie

Diamond Member
Jul 18, 2010
4,722
4,627
136
This got me thinking about a piece GN put out about a year ago https://www.gamersnexus.net/guides/3032-vega-56-cost-of-hbm2-and-necessity-to-use-it

AMD needed HBM because Vega is bandwidth hungry, and to keep power consumption reasonable. Unfortunately due in large part to the huge increases in memory pricing(the possible price fixing) HBM ended up being so expensive that AMD couldn't profit from Vega. 8GB of HBM costs about $150 for AMD and the interposer is another $25. $175 is a huge chunk of the Vega cards' retail price, and that's just for the memory solution. HBM2 is about 3x the price as the same amount of GDDR5. However it's not simple slapping GDDR5 on Vega, as that would require redesign work and ultimately might not even be feasible considering the bandwidth requirements of Vega.
Its worse than that. One of the key techs of Vega, the HBCC, needs HBM style memory as the on card cache. I guess this is one place where the designs needed for gaming and compute applications start to diverge. Both AMD and Nvidia have decided to to optimize for each market independently going forward. AMD's lack of R&D funds resulted in trying to design using the Swiss army knife approach.
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,248
136
Its worse than that. One of the key techs of Vega, the HBCC, needs HBM style memory as the on card cache. I guess this is one place where the designs needed for gaming and compute applications start to diverge. Both AMD and Nvidia have decided to to optimize for each market independently going forward. AMD's lack of R&D funds resulted in trying to design using the Swiss army knife approach.

I looked at a review with hbcc on and off and it looked like no real difference either way. Did I miss something?

https://www.gamersnexus.net/guides/3095-amd-vega-56-hbcc-gaming-benchmarks-on-vs-off?showall=1
 

maddie

Diamond Member
Jul 18, 2010
4,722
4,627
136
I looked at a review with hbcc on and off and it looked like no real difference either way. Did I miss something?

https://www.gamersnexus.net/guides/3095-amd-vega-56-hbcc-gaming-benchmarks-on-vs-off?showall=1
HBCC is a tech used to greatly enlarge the available ram for applications without having to jump through too many loops. At present AMD has an advantage here for directly accessing larger datasets than can fit in the on-card memory. The on-card memory (HBM2) is used as the cache for the larger pool of off-card memory (Ram, SSD, HDD).

I never expected that HBCC would ever allow a gain in performance for gaming. If the cards had 1-2 GB HBM2, then we would have seen a change, but 8GB is enough for most (99% ?) use cases in gaming.

HBCC is really most useful as a compute feature.
 

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
When vega is bandwith constrained its again because the ngg path is borked.
It hurts apu as much if not more as big vega.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Desktop/Mobile APUs at 12nm (Picasso) and dGPUs.

Oh how I wish AMD had been at all serious about 12nm dGPUs. It would give them a consumer filler product to put on the market now, so the cavernous void between RX Vega and Navi could be filled by something, anything.

It's stunning that AMD has allowed their consumer lineup an over-2-year gap between the launch of Vega last year, and Navi in 2019. Absolutely stunning. Talk about a wasted opportunity.

Hell if there's an HBM2 shortage, they probably could have arranged to use GF 12nm as a second source for that too. They're moving all their CPU production off 12nm anyway.

When vega is bandwith constrained its again because the ngg path is borked.
It hurts apu as much if not more as big vega.

You would think the various APU releases would be excellent test-cases in fixing ngg path. Looks like the first Vega with fixed ngg path will be Vega10, which is I guess the 7nm Vega refresh?

Also, here's hoping for 7nm Vega Frontier Edition 2 (or whatever).