[Rumor, Tweaktown] AMD to launch next-gen Navi graphics cards at E3

Page 101 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

lifeblood

Senior member
Oct 17, 2001
867
33
91
Well there's a big one, a bigger one and the biggest one.
So small Navi, then three more "larger" beyond Navi 10? Thats a lot of work designing all those dies, but it would be incredibly exciting if it comes to fruition. Forgive me if I'm a bit cautious as that really is a lot of work for AMD. Of course now with the success of Ryzen and (to a lesser extent) RX 5700 they do have the resources to put towards it...

What hes saying is that AMD is planning Navi 14: small one, Navi 12 - big one with 4096 Shaders, bigger one: Navi 21, and I think Navi 23 If I remember correctly. About that last one I do not even dare to speculate ;).
Some clarity! Thanks you, sir.
 
  • Like
Reactions: guachi

NostaSeronx

Platinum Member
Sep 18, 2011
2,520
341
126
If Seronx is correct it might be more feasible to use single HBM2 stack for small chips, for numbers of reasons.
It is specifically aimed at low-tdp currently. Other than hot TDPs there is nothing really stopping HBM2 from being in AMD's complete lineup.

APU SoC => DDR4/HBM2 bi-modal => Hot APUs(socketed) use DDR4/Cold APUs(BGA) use HBM2.
GPU SoC => GDDR6/HBM2 bi-modal => Hot GPUs use GDDR6/Cold GPUs use HBM2.

The GPU IO-die CoWoS project at AMD is purely HBM2/HBM2E/HBM3. Which I believe is targeting CoWoS-XL2(1200~1700 mm2) with chiplets.
 
Last edited:

Yotsugi

Senior member
Oct 16, 2017
919
412
96
So small Navi, then three more "larger" beyond Navi 10?
Yes.
Thats a lot of work designing all those dies
GPUs are dumb repeatable logic, and Navi is designed to be scalable so no.
Initial uArch expenses are getting quite steep though.
Forgive me if I'm a bit cautious as that really is a lot of work for AMD
They're driving a very bullish GPU roadmap and it's not like they care about meme public opinion.
Products talk.
 
  • Like
Reactions: Glo.

Glo.

Diamond Member
Apr 25, 2015
3,178
1,513
136
I dug a little about HBM2. Low-Cost HBM2 was all related to only one thing: reducing the number of TSV's, so Yotsugi was right here.

The other side of this coin. SK-Hynix has announced HBM2E, which have up to 460 GB/s bandwidth from 1024 bit memory bus. From what I can understand, you can have 512 bit memory bus, by simply lowering the amount of TSV's, which can work both ways: you can lower the bandwidth, and complexity of the build, and the cost related to it, while at the same time increasing the bandwidth available for the GPU.

512 bit, 4 GB memory stack could, in theory, offer 230 GB/s while at the same time offering reasonable manufacturaing costs for even low-cost products.
 
Last edited:
Mar 11, 2004
19,095
1,681
126
Yes.

GPUs are dumb repeatable logic, and Navi is designed to be scalable so no.
Initial uArch expenses are getting quite steep though.

They're driving a very bullish GPU roadmap and it's not like they care about meme public opinion.
Products talk.
The reason I'm skeptical is that the per transistor costs are increasing, and larger chips exacerbate that further by reducing the number of dice you get from a wafer even further, so larger GPUs are extra expensive these days. The other issue is that Navi doesn't seem to be good for the enterprise/compute market, so they can't make that up by selling those chips in much of that market where they'd get higher prices for them, like they could in the past with larger GPUs like Vega. People have been complaining about Nvidia's pricing, I think they'd accept it even less from AMD.

I have a personal hunch that those 3 larger Navi GPUs might end up not being nearly as interesting for gamers, in that they might be custom versions that pair AMD GPU with some other IP like say Google's TPU, aimed at AI processing. I'd expect that gamers will get one of them (what was probably known as Navi 20 before), but I'm quite skeptical that they'd make 3 larger consumer GPUs.

Maybe that is changing some, as cloud gaming will have some demand, and perhaps that's why AMD would be producing 3 even bigger Navi chips. And I believe Navi would still be good for say CAD where rendering throughput would still be important, and if they include ray-tracing would add a bit more use (movie rendering for instance).
 

lifeblood

Senior member
Oct 17, 2001
867
33
91
Well then, AMD; let's get on with it. The market is overdue for a Polaris replacement. You know, something that can be had for less than $300. Preferably less than $250.
I got my first paycheck of the semester from my side job (which I work to feed my computer addiction) so I'm looking for a new video card. The 1660Ti is currently at the top of my list but I'm impatiently waiting to see what AMD has to offer.
 

Glo.

Diamond Member
Apr 25, 2015
3,178
1,513
136
I got my first paycheck of the semester from my side job (which I work to feed my computer addiction) so I'm looking for a new video card. The 1660Ti is currently at the top of my list but I'm impatiently waiting to see what AMD has to offer.
Then you might want to hold your horses for a little bit more. Not Only AMD is going to release small GPU, but Nvidia also will refresh their lower-end GPUs starting with GTX 1660.
 

lifeblood

Senior member
Oct 17, 2001
867
33
91
Then you might want to hold your horses for a little bit more. Not Only AMD is going to release small GPU, but Nvidia also will refresh their lower-end GPUs starting with GTX 1660.
Given Nvidia's recent attempts to steal AMD's thunder (releasing super models at same time as Navi) I expect they will both release their new/refreshed stuff at about the same time. At least that's what I''m hoping.
 

Glo.

Diamond Member
Apr 25, 2015
3,178
1,513
136
Given Nvidia's recent attempts to steal AMD's thunder (releasing super models at same time as Navi) I expect they will both release their new/refreshed stuff at about the same time. At least that's what I''m hoping.
Considering that Nvidia is readying GTX 1660 Super, with 6 GB's of GDDR6, they know perfectly well how Navi 14 performs ;).
 

Yotsugi

Senior member
Oct 16, 2017
919
412
96
Considering that Nvidia is readying GTX 1660 Super, with 6 GB's of GDDR6, they know perfectly well how Navi 14 performs ;).
I still wonder why not just good old pricecuts.
If 1660S is 1660 pricing, then 1660Ti is ded (outside of like mobile).
 

Glo.

Diamond Member
Apr 25, 2015
3,178
1,513
136
I still wonder why not just good old pricecuts.
If 1660S is 1660 pricing, then 1660Ti is ded (outside of like mobile).
We have to wait for Navi 14 full pricing and performance, because AMD can undercut Nvidia, and offer, lets say, performance between 1660 and 1660 Ti, with 4 GBs of VRAM for 199$.

Which would make both Nvidia GPUs dead.
 

lifeblood

Senior member
Oct 17, 2001
867
33
91
We have to wait for Navi 14 full pricing and performance, because AMD can undercut Nvidia, and offer, lets say, performance between 1660 and 1660 Ti, with 4 GBs of VRAM for 199$.

Which would make both Nvidia GPUs dead.
Nvidia can’t do price cuts because their ego won’t let them (and anyone who claims ego has nothing to do with company decisions have obviously not sat in very many boardrooms of tech companies). They (Nvidia) has to have what they feel is the better product at/near that price range.

While I’m sure AMD will offer a 4 GB version of Navi 14 it’s pretty clear from the leaks that it will have 8 GB. I don’t think AMD wants to seriously try to undercut Nvidia on price either. They (Lisa Su) said they were no longer going to be the “value” company. Of course right after they said that they did undercut Nvidia on price with Navi 10, but I still believe there is a certain level of truth to that statement. And why not? Navi 10 is a very competitive product. Other than some early driver bugs it appears to be an excellent performer.

Additionally, I don't think AMD can undercut Nvidia. The 1660/1660Ti is not saddled with Tensor cores so it’s not as big (ie, expensive) as RTX. Both appear to use GDDR6 so no price savings there. Navi 14 is on a smaller process but 1660 is on a (assumed) less expensive process (per wafer). I think AMD is going to have to fight this one on performance alone.

(Edited for grammar)
 
  • Like
Reactions: prtskg

soresu

Senior member
Dec 19, 2014
220
35
91
Considering that Nvidia is readying GTX 1660 Super, with 6 GB's of GDDR6, they know perfectly well how Navi 14 performs ;).
Doesn't take a genius to infer the performance of a 24 CU chip based on Navi 10.
 

VirtualLarry

Lifer
Aug 25, 2001
44,787
3,825
126
Additionally, I don't think AMD can undercut Nvidia. The 1660/1660Ti is not saddled with Tensor cores so it’s not as big (ie, expensive) as RTX. Both appear to use GDDR6 so no price savings there. Navi 14 is on a smaller process but 1660 is on a (assumed) less expensive process (per wafer). I think AMD is going to have to fight this one on performance alone.
No, the GTX 1660 (as it is currently) is 6GB GDDR5, whereas, GTX 1660 ti is 6GB GDDR6. The GDDR6 cards. so I hear, have much better bandwidth, and are better for mining. If I had known that a GTX 1660 GDDR6 / Super version was coming out, I would have waited. I literally bought my MSI Ventus GTX 1660 ti like 3-4 days before I read that announcement (rumor?) in this thread. Darn it. (I mean, I know "smaller Navi" was coming out, and I kind of figured that NVidia might round out their "Super" lineup, at the lower end as well, but I had no idea of the time-frame that they might be released in.)

Hopefully, the GTX 1660 Super is only $200, maybe $220. Would be worth $220 if it had 8GB of VRAM somehow.
 

Bulldog21

Junior Member
Jan 21, 2019
8
3
41
No, the GTX 1660 (as it is currently) is 6GB GDDR5, whereas, GTX 1660 ti is 6GB GDDR6. The GDDR6 cards. so I hear, have much better bandwidth, and are better for mining. If I had known that a GTX 1660 GDDR6 / Super version was coming out, I would have waited. I literally bought my MSI Ventus GTX 1660 ti like 3-4 days before I read that announcement (rumor?) in this thread. Darn it. (I mean, I know "smaller Navi" was coming out, and I kind of figured that NVidia might round out their "Super" lineup, at the lower end as well, but I had no idea of the time-frame that they might be released in.)

Hopefully, the GTX 1660 Super is only $200, maybe $220. Would be worth $220 if it had 8GB of VRAM somehow.
Couldn't have said it better
 

ASK THE COMMUNITY