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

[Head1985]

I was talking about overpriced navi at 400/500usd and 2070 competition.
1year later than 2070 of course.It is almost 1year since nv launched 2070.
Yes DLSS/RT is why people will buy Nv cards if navi cost same with same performance.I mean why anyone should buy navi if they can have DLSS/RT for same price instead?
More power hungry(probably because its still GCN) than 2070 1year later for same price without DLSS/RT.Again why should anyone buy navi?

If they dont undercut NV they are beyond hope.And i will call them for pricefixing.

 

[jpiniero]

If it's actually $499 for the faster model, it's going to be faster than the 2070. Maybe not by much but faster.

 

[tviceman]

It seems like excuses and apologies are in full force before any official announcements, specs, and general info is even out. Doesn't look good. If they can land between an RTX 2060 and RTX 2070 for $299-329 then that is a win in my book so long as they don't blow their power budget and go over 200 watts.

I just don't see them matching an RTX 2070 in performance with a 250mm2 die given their perf/w and perf/mm2 deficits at 14nm.

 

[rainy]

1year later than 2070 of course.It is almost 1year since nv launched 2070.

No, isn't: RTX 2070 was released a bit over 7 months ago, quite obviously time for you running faster./s
https://en.wikipedia.org/wiki/GeForce_20_series

 

[maddie]

Realizing that TSMC has already starting volume production of 7nm+, is it possible that Navi will use this process?

The consoles are almost certainly designed for 7nm+ and this would imply that if being as cost efficient as possible, AMD would have reused the sub-components circuit layouts based on the different design rules vs 7nm. What is there to lose as the yields are already equivalent to 7nm and the other advantages as power and density will be surely wanted to compete with Nvidia effectively? I'm suggesting that they need as many advantages as possible against Nvidia in contrast to intel.

1M wafers/year 7nm & 7nm+ suggests no production bottleneck for AMD.
https://digitimes.com/news/a20190523PD209.html
TSMC has aleady started volume production of chips built using 7nm N7+ technology, the foundry's first process node with EUV lithography, according to the pure-play foundry.

TSMC has improved its N7+ process manufacturing yield rate to the level on par with that of its N7 process, said the company, adding that its overall 7nm chip output will ramp up dramatically this year.

 

[Head1985]

No, isn't: RTX 2070 was released a bit over 7 months ago, quite obviously time for you running faster./s
https://en.wikipedia.org/wiki/GeForce_20_series

Navi is still quite few months away.I think paper launch will be 7.7 and hard launch month later.

 

[jpiniero]

Realizing that TSMC has already starting volume production of 7nm+, is it possible that Navi will use this process?

TSMC said they weren't expecting that much business on 7+ until next year. Navi 20 could be using it.

 

[DrMrLordX]

No, isn't: RTX 2070 was released a bit over 7 months ago, quite obviously time for you running faster./s
https://en.wikipedia.org/wiki/GeForce_20_series

If you follow the link, you'll find that the RTX 2060 was the last RTX released in January 2019. RTX 2070 was released in October 2018. More like 9 months rather than 7 months.

Even still, I'll repeat my point for good measure: you don't price your new product at the same price as something that's been on the market for awhile unless you have seriously better performance. NV wasted a fair amount of available die space on RT features that many users can do without instead of focusing on raw performance. Launching something that could beat a 2070 at a lower price is the correct move for AMD, and yet instead, they choose too high of a price.

Realizing that TSMC has already starting volume production of 7nm+, is it possible that Navi will use this process?

I would expect that next year.

TSMC said they weren't expecting that much business on 7+ until next year. Navi 20 could be using it.

Woops, you beat me to it. What you said.

 

[maddie]

TSMC said they weren't expecting that much business on 7+ until next year. Navi 20 could be using it.

..........................................................
I would expect that next year.

Woops, you beat me to it. What you said.

Did you both read the digitimes post? Volume production has already begun. This supersedes previous info.

Can you have already started volume production and also not expect much business until next year? Sounds contradictory.

 

[crazzy.heartz]

According to Sapphire Rep - $399 Navi variant is better than RTX 2060 and the $499 variant "Matches" RTX 2070.

Per their logic, if that $499 Navi SKU "matches/competes" with a RTX 2070, then it's most likely in DX12/Vulcan games that favor AMD; kind of one's that Radeon 7 has lead over RTX 2080.

Performance delta between a RTX 2060 and RTX 2070 is around 15-20%, at stock clocks.

Here, $399 variant could be the cut chip with 36 CUs & $499 is the full chip 40 CUs.

Or
if the yields are poor, $399 is the cut chip with 36 CUs & $499 is the full chip with AIO [Then release a regular full variant for $450 once yields improve]

Or
AMD could only release the full/ 40 CU variant as $399 - Air Cooled & $499 - OC'd AIO Version

$499 for a 40 CU Navi that matches a RTX 2070, which is also a full chip and features Ray Tracing, doesn't sound that enticing to me..

 

[jpiniero]

Did you both read the digitimes post? Volume production has already begun. This supersedes previous info.

Can you have already started volume production and also not expect much business until next year? Sounds contradictory.

That's what was I remember from one of their earnings calls not too long ago. Maybe it's mainly small time customers at a "volume" production level.

I suppose you could say that TSMC is way ahead of schedule, but production takes several months so for a release on 7/7 you would have needed to start in like Feb-Early March.

 

[Glo.]

According to Sapphire Rep - $399 Navi variant is better than RTX 2060 and the $499 variant "Matches" RTX 2070.

Per their logic, if that $499 Navi SKU "matches/competes" with a RTX 2070, then it's most likely in DX12/Vulcan games that favor AMD; kind of one's that Radeon 7 has lead over RTX 2080.

Performance delta between a RTX 2060 and RTX 2070 is around 15-20%, at stock clocks.

Here, $399 variant could be the cut chip with 36 CUs & $499 is the full chip 40 CUs.

Or
if the yields are poor, $399 is the cut chip with 36 CUs & $499 is the full chip with AIO [Then release a regular full variant for $450 once yields improve]

Or
AMD could only release the full/ 40 CU variant as $399 - Air Cooled & $499 - OC'd AIO Version

$499 for a 40 CU Navi that matches a RTX 2070, which is also a full chip and features Ray Tracing, doesn't sound that enticing to me..

Considering 8 shader engines with 5 CUs each you either disable the Shader Engines, or CUs.

So its either 30/35 CUs for cut down versions, if you cut down Shader Engines, or, 32 setup, if you just cut out CUs.

 

[Ajay]

Did you both read the digitimes post? Volume production has already begun. This supersedes previous info.

Can you have already started volume production and also not expect much business until next year? Sounds contradictory.

Standard term is HVM, or high volume production for high wafer throughput rates. This isn't that. As @jpiniero pointed out, TSMC doesn't expect much volume till 2020.
Oh, and I would be surprised if Navi gets spun up on 7N+ (EUV) anytime soon. 7N+ has a new PDK and requires significant re-design. Next gen Instinct GPU (2020?) is more likely designed on 7N+ from the get go. IMHO.

 

[maddie]

Standard term is HVM, or high volume production for high wafer throughput rates. This isn't that. As @jpiniero pointed out, TSMC doesn't expect much volume till 2020.
Oh, and I would be surprised if Navi gets spun up on 7N+ (EUV) anytime soon. 7N+ has a new PDK and requires significant re-design. Next gen Instinct GPU (2020?) is more likely designed on 7N+ from the get go. IMHO.

Yes, 7nm+ has a new PDK which is already being used for the consoles. That means the rops, caches, shaders, command processor, IO, etc is already done on this.

My argument is that for AMD to double the work on 7nm is a waste of money. It seems the opposite of AMD strategy of late, especially as Navi will sell a lot less than the consoles. $300M for a minor marketshare discrete GPU seems like quite a low return investment.

If AMD is going to sell at prices just below Nvidia, thus limiting market penetration, and are also using a different PDK than the consoles, then good luck to them.

 

[Stuka87]

It was 18% in february before 1660/1660TI launch..
I would add to that:
1-1year late.It is normal release product 1 year later than competition fo same money?Again i call that price fixing.But they already did it with vega so i am not sure what AMD thinking.
2-no DLSS/RT support
3-probably more power hungry(not sure)
4-if they relase 2070 performance for 2070 price then people whos waiting for navi will go out and buy 2070 because RT/DLSS or will not upgrade at all(because its overpriced like 2070)

1. You have now said 1 year late twice. It has not been a year. And tech companies release when they are ready. They don't synchronize their releases.
2: DLSS is a joke used for marketing. RT at this point in time doesn't matter. In 5 years, when mid range cards can do it, then it might matter.
3: Nothing to suggest this
4: People who are waiting for Navi are not going to go and buy NV. They are waiting because they don't want NV.

 

[crazzy.heartz]

Considering 8 shader engines with 5 CUs each you either disable the Shader Engines, or CUs.

So its either 30/35 CUs for cut down versions, if you cut down Shader Engines, or, 32 setup, if you just cut out CUs.

There are a lot of ways they can carve out this chip:

32CU


35CU


36CU


They'd most likely disable faulty CU's on a particular chip. 36CUs for the cut down part would give minimal performance hit.

Last time around, they launched the Full Polaris desktop part which was almost a replica of the chip they designed for PS4 pro, in terms of Cu count.





However, as SONY sponsored Navi's development, I doubt if they would create a 40CU only chip for the PS5; considering XBox 1X had a 44CU enabled chip. (Would most likely have a 60CU chip for the Next XBox.)

They need to clock this chip optimally, to reach desired power/performance targets on the PS5's fixed power envelop; so, the more CUs the better. AMD is not going to waste resources either creating a different design and we would most likely get a similar desktop version of that same chip.

AMD's chips have never reached performance parity with Nvidia, in the Shader/CU count. So, for this $499 chip to be at RTX 2070 performance levels, it'd most likely be a 48CU chip. Thus, a 40CU cut down part performing better than Rtx 2060 would seem more plausible.

 

[DrMrLordX]

Did you both read the digitimes post? Volume production has already begun. This supersedes previous info.

Can you have already started volume production and also not expect much business until next year? Sounds contradictory.

If AMD is using 7nm+ this year and can't beat the 2070 at a price lower than $499 then all I can say is, whiskey tango foxtrot AMD?!?

AMD has to leave room for improvement with Navi for next year, when reportedly they'll bring out Navi20 (or whatever they call it) to take on the 2080Ti and/or whatever replaces it.

Standard term is HVM, or high volume production for high wafer throughput rates. This isn't that. As @jpiniero pointed out, TSMC doesn't expect much volume till 2020.
Oh, and I would be surprised if Navi gets spun up on 7N+ (EUV) anytime soon. 7N+ has a new PDK and requires significant re-design. Next gen Instinct GPU (2020?) is more likely designed on 7N+ from the get go. IMHO.

Multiquote hosed things up, so copy-paste wins the day? Whatever. Ahem. I'll echo the sentiment from others that AMD will probably fork over the cash to respin Navi because consoles. The work is already being done. AMD also needs a process edge next year to push 56CU (or whatever) Navi over the top in 2020. They have to replace Radeon VII at the top of the stack after all.

According to Sapphire Rep - $399 Navi variant is better than RTX 2060 and the $499 variant "Matches" RTX 2070.

Clearly those prices are too high. If AMD follows that course, then RIP Navi out of the gate.

 

[beginner99]

$499 for a 40 CU Navi that matches a RTX 2070, which is also a full chip and features Ray Tracing, doesn't sound that enticing to me..

Agree. it really sounds like a "cartel". But my real suspicion is supply. they bet on x86 and hence zen2 chiplets get most of the 7nm wafers. if you can sell the few navi parts at high prices why offer them for less?

4: People who are waiting for Navi are not going to go and buy NV. They are waiting because they don't want NV.

Or people simply don't want to pay inflated prices and wait for competition to get a better deal.

 

[DrMrLordX]

. if you can sell the few navi parts at high prices why offer them for less?

Brand damage. AMD is pushing a midrange-to-upper midrange part for $499, 9 months after the launch of another similar product with more features (good, bad, or otherwise). What's the value proposition here? There is none. AMD looks awful launching at such high prices. If there really is a shortage of 7nm wafers, you launch at sane prices ($299 for the 2060 killer, $399 for the 2070-alike) and then fill in the supply gaps later when more wafers become available. Then you hit with Navi20 next year on the newer process. Now you have competition. AMD is bowing to JHH and admitting defeat before their product even hits the shelves. It's disgusting.

 

[coercitiv]

if you can sell the few navi parts at high prices why offer them for less?

If you don't have the volume you're -Redacted- either way, best approach would be to stockpile and use in laptops.

They can get away with doing this in the upper segment, as Radeon VII was seen as a halo product launch, but doing the same in the more price sensitive "mainstream" area of the market will result in immediate negative brand perception. (as @DrMrLordX already explained)

Remember this? They sang the tune, now they have to dance.



They need to have a compelling proposition with Navi, Turing prices for Turing performance won't cut it unless it comes with relaxed power figures and excellent oc potential.

Profanity is not allowed in the technical forums.

Daveybrat
AT Moderator

 

[french toast]

There are a lot of ways they can carve out this chip:

32CU
View attachment 6641

35CU
View attachment 6642

36CU
View attachment 6643

They'd most likely disable faulty CU's on a particular chip. 36CUs for the cut down part would give minimal performance hit.

Last time around, they launched the Full Polaris desktop part which was almost a replica of the chip they designed for PS4 pro, in terms of Cu count.

View attachment 6646

View attachment 6647

However, as SONY sponsored Navi's development, I doubt if they would create a 40CU only chip for the PS5; considering XBox 1X had a 44CU enabled chip. (Would most likely have a 60CU chip for the Next XBox.)

They need to clock this chip optimally, to reach desired power/performance targets on the PS5's fixed power envelop; so, the more CUs the better. AMD is not going to waste resources either creating a different design and we would most likely get a similar desktop version of that same chip.

AMD's chips have never reached performance parity with Nvidia, in the Shader/CU count. So, for this $499 chip to be at RTX 2070 performance levels, it'd most likely be a 48CU chip. Thus, a 40CU cut down part performing better than Rtx 2060 would seem more plausible.

Thank you, if this Navi 10 tops out at 40 CU and is being used to fight a 2070, then AMD just made a huge error imo.
Something like a 48CU/64Rop/8 SE die with gddr6 and lots of small improvements all round would be preferable imo.

Rx 480 had 36CU and 32 Rops, it got nowhere near a 1070, primarily because it was never designed to, it was a lower tier product.
Why the hell some 3 years later would they try to compete with a xx70 tier class gpu with only 40CU?
I smell a fish, either that or AMD is going to set themselves up for a difficult product launch..

 

[beginner99]

They need to have a compelling proposition with Navi, Turing prices for Turing performance won't cut it unless it comes with relaxed power figures and excellent oc potential.

fully agree but leaks are indicating exactly this and hence my possible explanations about supply. Supply will go up later because the big guys will then be saturated with Rome.

 

[darkswordsman17]

Since AMD has said Navi isn't until Q3, why are people talking like they expect us to get worthwhile details in the next day or so? (Yes, I know its start of Computex, but people aren't expecting an actual launch are they? Plus isn't this thread about them rumored to be launching Navi at E3 which is nearly another couple of weeks away?) I'm guessing at best we'll get some very limited demo where they lock the framerate or something so we won't be able to discern much of anything out of it. I highly doubt we get details (like CU counts, architecture details, etc) let alone prices.

I've love to be surprised (in a positive direction), but people ramping up the speculation, I think you're just going to be disappointed and be stuck waiting awhile again.

 

[JDG1980]

The slide up there right now mentions Navi... of course, it could just be mentioned with no details.

UPDATE: Lisa Su just said they're going to "preview" Navi.

 

[JDG1980]

From the keynote:
Navi will have an all-new "RDNA" architecture - Lisa Su specifically said this is different from GCN. There will still be new versions of GCN for high compute workload applications. But RDNA will be a from-scratch gaming design, basically the GPU equivalent of Zen.
1st GPU with PCIe 4 support.
Lisa Su claims 1.25x performance per clock and 1.5x performance per watt for Navi.
Radeon RX 5000 family - in honor of AMD's 50th birthday.
They're demoing Strange Brigade with the "RX 5000 series" against the RTX 2070. Announcer said the "RX 5700 GPU" beats 2070 by roughly 10% in this preview.
Available in July.
Further announcements with pricepoints, etc. will be on June 10