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

Page 102 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
DrMrLordX

DrMrLordX

Lifer
Apr 27, 2000
21,629
10,841
136
How long has it been since AMD updated their mobile dGPUs? Since notebook OEMs seem so reluctant to configure mobile iGPUs properly, I guess AMD needs to keep cranking out mobile dGPUs to earn design wins.
 
DrMrLordX

DrMrLordX

Lifer
Apr 27, 2000
21,629
10,841
136
Yotsugi said:
Vega12 was like last year but the thing wasn't exactly mass market.
Click to expand...

I never saw that in any notebooks compared to Polaris-based stuff. Just a quick glance at the notebookcheck rankings for mobile dGPUs:

Comparison of Laptop Graphics Cards

Inclusive list of all graphics cards currently available for notebooks. Approximate 3DMark benchmarks results are provided in italic.
www.notebookcheck.net www.notebookcheck.net

shows Vega mobile below the RX 580X mdGPU, and has no relative performance indicator for it. Under Vega Mobile specifically:

www.notebookcheck.net

AMD Radeon RX Vega Mobile GPU

Specifications and benchmarks of the AMD Radeon RX Vega Mobile GPU.
www.notebookcheck.net www.notebookcheck.net

They have no benchmark results for it. It may as well not exist.
 
Trumpstyle

Trumpstyle

Member
Jul 18, 2015
76
27
91
Things are looking very good for me, this has been my prediction for some time now.

Navi apu = 12CU 7nm+

Navi 14 = 24CU 7nm, 128-bit bus ggdr6
Navi 10 = 40CU 7nm, 256-bit bus ggdr6
Navi 12 = 56CU 7nm, 256-bit bus ggdr6 but 512 GB/s memory speed

Navi 21 = 80CU 7nm+, ray-tracing with HBM
Navi 23 = 102CU 7nm+, ray tracing with HBM

This is a prediction based on Yotsugis comments, redgamingtech and the Navi 10 layout.

BONUS: PS5 and next-gen Xbox will have 52CU on 7nm+ (10TF+), maybe PS5 will have 36CU with 1800mhz clockspeed we have conflicting information here.
 
S

soresu

Platinum Member
Dec 19, 2014
2,662
1,862
136
Trumpstyle said:
BONUS: PS5 and next-gen Xbox will have 52CU on 7nm+ (10TF+), maybe PS5 will have 36CU with 1800mhz clockspeed we have conflicting information here.
Click to expand...
Both stated a push for efficiency (eco friendly PR) which would imply a larger chip with lower clockspeeds, I would doubt anything higher than 1700 mhz.
 
Glo.

Glo.

Diamond Member
Apr 25, 2015
5,707
4,552
136
I wonder if Navi 14 is actually Mobile-First design. Which would be good indicator(hope...?) for Performance Per Watt of this GPU.
 
Glo.

Glo.

Diamond Member
Apr 25, 2015
5,707
4,552
136
Yotsugi said:
Perf/w is determined by whatever arbitrary point on v/f curve you choose to run your boost at.
Click to expand...
Considering that RX 5500M is supposed to have 1645 MHz clock it actually might be quite good ;).
 
M

maddie

Diamond Member
Jul 18, 2010
4,740
4,674
136
soresu said:
102 CU seems a bit much, that has to be pushing over 500 mm2 or close even at 7nm+.
Click to expand...
Hasn't 7nm+ already passed 7nm from the yield % metric? I think we should expect the larger use of EUV to drastically reset the yield curves.
 
D

darkswordsman17

Lifer
Mar 11, 2004
23,074
5,557
146
maddie said:
Hasn't 7nm+ already passed 7nm from the yield % metric? I think we should expect the larger use of EUV to drastically reset the yield curves.
Click to expand...

I highly doubt that.

soresu said:
Both stated a push for efficiency (eco friendly PR) which would imply a larger chip with lower clockspeeds, I would doubt anything higher than 1700 mhz.
Click to expand...

Yeah I'm not expecting probably more than like 1600MHz and even that in maybe boost clock situations. There's rumors that there's going to be a PS5 Pro at launch and I'd guess that'll be a model pushing higher clock speeds, but they'll be limited in what they can get.

A big part of how they're pushing for efficiency will be related to hibernation/suspend that going with a fast SSD will enable. Sony just recently talked about that.

Personally I'm expecting ~15TF (probably closer to like 14.5) from 64CU in the next Xbox. Base PS5 I think will be closer to 10TF (from lower clocked 48CU, not likely fully enabled), with a Pro version that's probably fully enabled plus higher clock speeds offering ~12.5-13TF so Sony doesn't get outclassed by Microsoft as much. And then I think we might see a lower end Xbox that is closer to the base PS5 (although it might not come out til the year after), after Microsoft tweaks the lower end console that was supposed to target streaming (that I'm guessing was probably ~50% of the main console at around 7-8TF, but with the base PS5 being that close that Microsoft will choose to add some CUs and get them to be roughly equal; I'd guess it'll add ray-tracing that it likely originally lacked as well).

I just realized. There's supposedly 3 larger versions of Navi? One each for the consoles, then a halo one for the dGPU/pro market. Then the larger one will probably be dual purposed for consumer gaming, game streaming (Stadia; maybe Sony and Microsoft would choose to use the same GPU in their consoles for their streaming so that each user is getting basically a console in the cloud), and pro stuff that is rendering focused (CAD, etc). The dGPU one would be 96CU with 84 and 72 CU versions (top end one would be $999 with 4 stacks of HBM2 for 1TB/s bandwidth, then 3 stacks for ~850GB/s, and then 3 stacks for ~750GB/s). Then maybe they'll be able to sell the Microsoft GPU as a dGPU for $549 (64CU 16GB GDDR6, up to 640GB/s if it has a 320bit bus) with a 56CU version for $449. Current Navi (possibly refreshed/updated with ray-tracing bits, maybe they'd be using the PS5 GPU), would be $349/299 with 8GB GDDR6 (which will offer 512GB/s).
 
M

maddie

Diamond Member
Jul 18, 2010
4,740
4,674
136
darkswordsman17 said:
I highly doubt that.



Yeah I'm not expecting probably more than like 1600MHz and even that in maybe boost clock situations. There's rumors that there's going to be a PS5 Pro at launch and I'd guess that'll be a model pushing higher clock speeds, but they'll be limited in what they can get.

A big part of how they're pushing for efficiency will be related to hibernation/suspend that going with a fast SSD will enable. Sony just recently talked about that.

Personally I'm expecting ~15TF (probably closer to like 14.5) from 64CU in the next Xbox. Base PS5 I think will be closer to 10TF (from lower clocked 48CU, not likely fully enabled), with a Pro version that's probably fully enabled plus higher clock speeds offering ~12.5-13TF so Sony doesn't get outclassed by Microsoft as much. And then I think we might see a lower end Xbox that is closer to the base PS5 (although it might not come out til the year after), after Microsoft tweaks the lower end console that was supposed to target streaming (that I'm guessing was probably ~50% of the main console at around 7-8TF, but with the base PS5 being that close that Microsoft will choose to add some CUs and get them to be roughly equal; I'd guess it'll add ray-tracing that it likely originally lacked as well).

I just realized. There's supposedly 3 larger versions of Navi? One each for the consoles, then a halo one for the dGPU/pro market. Then the larger one will probably be dual purposed for consumer gaming, game streaming (Stadia; maybe Sony and Microsoft would choose to use the same GPU in their consoles for their streaming so that each user is getting basically a console in the cloud), and pro stuff that is rendering focused (CAD, etc). The dGPU one would be 96CU with 84 and 72 CU versions (top end one would be $999 with 4 stacks of HBM2 for 1TB/s bandwidth, then 3 stacks for ~850GB/s, and then 3 stacks for ~750GB/s). Then maybe they'll be able to sell the Microsoft GPU as a dGPU for $549 (64CU 16GB GDDR6, up to 640GB/s if it has a 320bit bus) with a 56CU version for $449. Current Navi (possibly refreshed/updated with ray-tracing bits, maybe they'd be using the PS5 GPU), would be $349/299 with 8GB GDDR6 (which will offer 512GB/s).
Click to expand...
I think you're assuming too low a yield. Look at the N7 large die and compare to 16nm. This should allow a mass market 500mm^2 die by next year.

Quote:
"TSMC’s N7+ is their first process technology to adopt EUV for a few critical layers. N7+ entered mass production last quarter (Q2). TSMC says they have demonstrated similar yield to N7"

AND THIS

vlsi-2019-tsmc-n7-yield-2.jpg
 
Trumpstyle

Trumpstyle

Member
Jul 18, 2015
76
27
91
soresu said:
102 CU seems a bit much, that has to be pushing over 500 mm2 or close even at 7nm+.
Click to expand...
I actually got that one wrong, suppose to be 112CU, all cards are based on this Navi 10 layout


But wccftech latest leak about mobile navi cards indicates this navi 10 layout by AMD is just a dummy and not real. As Navi 14 should be 2 shader engines with 12CU on each and not 1 shader engine with 24CU's (wccftech says 5500m is 22CU so a cut-down navi14 with 1 shader engine).

soresu said:
Both stated a push for efficiency (eco friendly PR) which would imply a larger chip with lower clockspeeds, I would doubt anything higher than 1700 mhz.
Click to expand...
Sony was talking about standby times, we have a verified insider over at the site resetera saying both next-gen consoles will be over 10TF, 52CU on 7nm+ is the easiest way getting there, but if wccftech leak is true I think we looking at 44CU on normal 7nm clocked at 1800mhz instead.

We had an old leak in march called gonzalo showing a gaming soc with 8 zen cores clocked at 3,2ghz and a gpu clocked 1,8ghz I assumed this was 36CU but might actually be 44CU.
 
lifeblood

lifeblood

Senior member
Oct 17, 2001
999
88
91
I feel like AMD is trying to shame me into buying the 5700, or 5800 when it arrives.

Back in the days of the HD 5000 Series (2009), the 5800 Cards were for the hard core gamers while the 5700 series were for the budget gamers. The launch MSRP of the 5870 was $380 while the 5850 was $260. The 5770 was a relatively cheap $160. They also had the 5670 at $99, but those folks weren’t even allowed to call themselves gamers.

Fast forward to today and the budget gamer cards, the 5700 series, are a whopping $400 (5700 HD) and $350 (5700). If we’re lucky the low end 5600 cards (for people who don’t deserve to call themselves gamers) will only be $250 while the real “Mainstream gamer” cards, the 5800, will be heaven knows what.

I realize it’s just a name and what really matters is performance. The 5600 will be optimized for 1080p gamers while the 5700 is 1440p and the 5800 will target the 4k crowd. The logical part of my brain gets that. It’s the illogical part that is embarrassed to say I’m interested in the 5600 series.

Am I the only one?
 
fleshconsumed

fleshconsumed

Diamond Member
Feb 21, 2002
6,483
2,352
136
lifeblood said:
. . .

Am I the only one?
Click to expand...
There is no shame here, it is what it is. Most tech-aware people know we're being taken, but there is nothing we can do. I buy based on value and there is no value in video card market right now. For now I'm putting my "fun" money into photography again. If AMD/nVidia want me to shift my "fun" money back to video cards they're going to have to do better than pricing midrange at $400-500.
 
Glo.

Glo.

Diamond Member
Apr 25, 2015
5,707
4,552
136
Anyone care to speculate about the performance of navi 14 based GPUs?

RX 5600 - 20 CUs, 4 GB GDDR6 - 169$ Between GTX 1650 Ti and GTX 1660 in performance.
RX 5600 Xt - 24 CUs, 4 GB GDDR6 - 199$ for GTX 1660 Ti performance. Also 8 GB version for 229$.
 
lifeblood

lifeblood

Senior member
Oct 17, 2001
999
88
91
Glo. said:
Anyone care to speculate about the performance of navi 14 based GPUs?

RX 5600 - 20 CUs, 4 GB GDDR6 - 169$ Between GTX 1650 Ti and GTX 1660 in performance.
RX 5600 Xt - 24 CUs, 4 GB GDDR6 - 199$ for GTX 1660 Ti performance. Also 8 GB version for 229$.
Click to expand...
The leaked Benchmarks are not giving me a warm feeling. According to the leaks Navi 14 is performing anywhere from the same as the RX 570 to (slightly) faster than RX 590. The most recent rumors claim Navi 14 will be a 128bit card with 8GB GDDR6 which again makes me hesitant. 128bit bus will have a hard time keeping the GPU fed.

On the logical side, AMD is simply not so dumb as to bring out a new card that performs the same as the old one. I expect the top end release version performs better than the RX 590 while being cooler and more power efficient. Of course AMD has disappointed me before...

Anyway, here we go:

RX 5600 XT - 24 CUs, 8 GB GDDR6 - $250, GTX 1660 Ti performance. 4 GB version at $229.
RX 5600 - 22 CUs, 8 GB GDDR6 - 199$, GTX 1650 Ti performance. 4 GB version at $179.
RX 5500 - 20 CUs, 3 GB GDDR6 - $149, GTX 1650 performance
 
Status
Not open for further replies.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom