Discussion RDNA4 + CDNA3 Architectures Thread

[DisEnchantment]

With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits

History for llvm/lib/Target/AMDGPU - llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - History for llvm/lib/Target/AMDGPU - llvm/llvm-project

github.com

Or Phoronix

More AMD "GFX940" Enablement Work Landing In LLVM - Phoronix

www.phoronix.com

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it

This is nuts, MI100/200/300 cadence is impressive.



Previous thread on CDNA2 and RDNA3 here

Question - Speculation: RDNA3 + CDNA2 Architectures Thread

Man I have been dying to make this one for a while now. First rumours for RDNA3 are here so new thread time! Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3...

forums.anandtech.com

 

[Tigerick]

RDNA4 N48 is a mid range GPU

After RDNA3 and Zen5's wrecked hype trains, people should be more cautious about RDNA4's leaked specs and hypes especially with AMD almost non-existence dual issue designs. Let me show you with the table below:

		780M	890M		7800 XT	N48	OC of N48 ?	7900 GRE	7900 XT	7900M		RTX 4070 Super	RTX4070 Ti Super	RTX 4080	Mobile RTX 4090
GPU		RDNA3	RDNA3+		N32	RDNA4		N31	N31	N31		AD104	AD103	AD103	AD103
CU/SM		12	16		60	64		80	84	72		56	66	76	76
RT		12	16		60	64 ?		80	84	72		56	66	76	76
Speed		2800 MHz	2900 MHz		2430 MHz	2900 ?	3200 ?	2245 MHz	2394 MHz	2090 MHz		2475 MHz	2610 MHz	2505 MHz	1695 MHz
IF Cache					64 MB	64 MB		64 MB	80 MB	64 MB		48 MB	48 MB	64 MB	64 MB
Memory		128-bit LPDDR5x-7500	128-bit LPDDR5x-7500		256-bit GDDR6 19.5 Gbps	256-bit GDDR6 20 Gbps	256-bit GDDR6 24 Gbps	256-bit GDDR6 18 Gbps	320-bit GDDR6 20 Gbps	256-bit GDDR6 18 Gbps		192-bit GDDR6X 21 Gbps	256-bit GDDR6X 21 Gbps	256-bit GDDR6X 22.4 Gbps	256-bit GDDR6 18 Gbps
Memory BW		120 GB/s	120 GB/s		624 GB/s	640 GB/s	768 GB/s	576 GB/s	800 GB/s	576 GB/s		504 GB/s	672.3 GB/s	716.8 GB/s	576 GB/s
+%					Base	+ 3%	+ 23%	- 8%	+ 28%	- 8%
FP32 (SI)		4.3 TF	5.94 TF		18.7 TF	23.8 TF	26.2 TF	23 TF	25.7 TF
FP32 (DI)					37.4 TF	47.6 TF	52.4 TF	46 TF	51.4 TF	38.52 TF		35.48 TF	44.1 TF	48.74 TF	32.98 TF
+%			+ 38%		Base
TPU Raster					Base	?	?	+ 10%	+ 30%	- 9%		+ 8%	+ 27%	+ 47%	+ 0%
Aztec Ruins High Tier 4K		34.6	39.7
+%			+ 15%

• 890M's graphics performance is severely constrained by memory bandwidth. Even though 890M has increased FP32 by 38%, we only seen about 15% faster in average gaming fps.
• Based on leaks, I have calculated FP32 of N48 with memory bandwidth. If 2.9GHz is the top speed of N48, the FP32 of N48 has bumped 27% compared to 7800XT even though they are having similar memory bandwidth. Based on specs and performance of 7900GRE, it seems the GPU is not memory constrained with 46 TF (DI). Theoretically, with 47.6 TF, we should be expecting N48 to perform slightly faster than 7900GRE, i.e.. around 10% faster than 7800XT.
• The calculated figure is far lower than leaks suggested which is faster than 7900XT, that's mean around 30% faster rasterization than 7800XT. Not to mentioned faster than RTX4070Ti Super....Why?
• Cause AMD marketing team is using DI to calculate FP32 and compared to RTX4070Ti Super, 47.6TF sure sound faster than 44.1TF of RTX4070Ti S and even approaching the 48.74TF of RTX4080. However, FP32 of RDNA and RTX4000 is not comparable.... Even though 7900XT is having 51.4 TF, the rasterization performance is slower than RTX4080, that's why NV is able to charge $1199 for RTX4080...
• NO, N48 is not replacing N31, AMD will keep selling 7900XT/XTX. What AMD could be replacing are 7700XT 12GB, 7800XT 16GB and 7900GRE 16GB, in which current leaks suggested there are 3 models planned: N48 12GB, N48 16GB 18Gbps and N48 16GB 20Gbps.
• Mobile RDNA4 might be another reason why AMD make it. Fire Range with 16 Zen5 cores should be launching by the end of Q3 which aligns with RDNA4 launching. If AMD manages to launch mobile N48 with 16GB 256-bit memory bus, AMD could finally challenge NV's mobile RTX4090...

 

[Mahboi]

7900 GRE	7900 XT
256-bit GDDR6 24 Gbps	256-bit GDDR6 18 Gbps
768 GB/s	576 GB/s

Those Mem speed numbers don't add up, boy.
You're tryin to scroo wit' me?

(nah seriously how is the GRE somehow having 24Gbps but the XT only 18? And how is a XT 256 bit? It's 320 bit)

 

[Tigerick]

Those Mem speed numbers don't add up, boy.
You're tryin to scroo wit' me?

(nah seriously how is the GRE somehow having 24Gbps but the XT only 18? And how is a XT 256 bit? It's 320 bit)

Ooh, the forum table is messed up, give me times to correct it

 

[Arctic Islands]

AMD has managed to dramatically shrink the L3 on Zen5 vs Zen4, I expect similar efforts for the MALL in RDNA4. N5 -> N4P is also a small shrink.

I've calculated the L3 size on Zen 5 by strix point's die shot. It's about 9.8mm^2 per 16MB slice. It is smaller than a 8x2MB slice on MCDs, which is about 15.6mm^2.
But still, SEs in N48 have to be much smaller than N31s' such that everything include I/O, codec, memory PHY and more can fit in such a small area, which I don't think it's true.

 

[Tigerick]

I've calculated the L3 size on Zen 5 by strix point's die shot. It's about 9.8mm^2 per 16MB slice. It is smaller than a 8x2MB slice on MCDs, which is about 15.6mm^2.
But still, SEs in N48 have to be much smaller than N31s' such that everything include I/O, codec, memory PHY and more can fit in such a small area, which I don't think it's true.

240 mm2 die size is older info. MLID mentioned that the die area of N48 could range from 300 - 350 mm2 which is more believable with 64MB IC.

 

[marees]

If 2.9GHz speed is the top speed of N48

Why would you assume that ?

Even chiplet RDNA 3 can boost to 3.75ghz (subject to power constraints of data transfer)

More immediate constraints are:

1. 7900 gre was constrained by memory speed
2. MLID leaked that power target for top RDNA 4 is same as 7700xt (+ 10% overclock power limits)
3. Assume a modest IPC boost

Putting all together I expect (as claimed by MLID)

1. Top n48 to launch this year (leaked)
2. 7700xt power consumption (leaked)
3. 7900xt raster give or take few %
4. 4070 ti super RT (win or lose depending on game)
5. 4070 super price (my inference)
6. 7900xt would sell for $650 since it has more VRAM & supports ROCM
7. Other RDNA 4 cards to launch strictly only after 7800xt & 7700xt are sold out

Price/performance

1. $600 = 4070 ti super
2. $500 = 4070 super
3. $400 = 4070 gddr6
4. $300 = 4060 ti

 

[blckgrffn]

Now they just need to release it and all will be well. Planning to get a top tier RDNA4 at launch if raster is really at 7900XT levels.

 

[marees]

Now they just need to release it and all will be well. Planning to get a top tier RDNA4 at launch if raster is really at 7900XT levels.

Price/performance

1. $600 = 4070 ti super
2. $500 = 4070 super
3. $400 = 4070 gddr6
4. $300 = 4060 ti

Thinking back to RDNA 1 launch we would seem to have a cumulative inflation of approx 50% (8.5% annualized inflation)

1. 5700xt = $400
2. 5700 = $350
3. 5600xt = $280
4. 5500 = $200

 

[blckgrffn]

Thinking back to RDNA 1 launch we would seem to have a cumulative inflation of approx 50% (8.5% annualized inflation)

1. 5700xt = $400
2. 5700 = $350
3. 5600xt = $280
4. 5500 = $200

I am going to say that AMD/Nvidia previously underestimated how much money we'd all spend on toys and COVID+Ethereum showed them differently on top of the nice little tradewar tariff and then we added a big old cherry of inflation to the top and that's how we went from the 5700XT (which I got for $340, naturally) to the whatever it is at $600 in the same "tier".

It was a confluence of factors and some of us played our part by purchasing right through the high prices and scarcity.

 

[Tigerick]

RDNA4 N48 is a mid range GPU

Right after I posted above message's title, Golden Pig Upgrade has posted the message below:


Don't expect flagship-level performance…… If you guys don't get my calculation, well at least you should understand what Golden Pig Upgrade implied: RDNA4 is not flagship-level GPU, period. That's mean AMD will keep selling 7900XT/XTX as flagship GPU. Here comes an interesting question, how would AMD name the RDNA4 lineup if AMD still selling RX7900XT and XTX? The media and most of the forum members have jumped the gun and predicted RDNA4 would be called RX8000 series based on leaked performance. However, if the leaks are wrong about performance figure, then what would/should AMD name the RDNA4 lineup? I am inclined to believe AMD would name RDNA4 lineup to be RX7000 series like 7850XT. Of course, naming could change at the last minute, let's wait for official announcement……

Current Lineup	CU	Total Die Size		Q4 2024 - 2025	CU	Total Die Size
7900 XTX 24GB	96	529 mm2
7900 XT 20GB	84	529 mm2
7900GRE 16GB	80	529 mm2		N48 16GB	64	300 - 350 mm2
7800 XT 16GB	60	346 mm2		N48 16GB	56 ?	300 - 350 mm2
7700 XT 12GB	54	346 mm2		N48 12GB	?	300 - 350 mm2
7600XT 16GB	32	204 mm2		N44 16GB	32	?

 

[maddie]

Don't expect flagship-level performance…… If you guys don't get my calculation, well at least you should understand what Golden Pig Upgrade implied: RDNA4 is not flagship-level GPU, period. That's mean AMD will keep selling 7900XT/XTX as flagship GPU.

The problem with this reasoning is that for the next gen, the 7900XT will not be flagship level. RTX 5070 maybe?

 

[del42sa]

"Don't expect flagship-level performance……"

I would say rather : Don´t expect anything .....

 

[marees]

what would/should AMD name the RDNA4 lineup? I am inclined to believe AMD would name RDNA4 lineup to be RX7000 series like 7850XT. Of course, naming could change at the last minute, let's wait for official announcement……

Current Lineup	CU	Total Die Size	Q4 2024 - 2025	CU	Total Die Size
7900 XTX 24GB	96	529 mm2
7900 XT 20GB	84	529 mm2
7900GRE 16GB	80	529 mm2	N48 16GB	64	300 - 350 mm2
7800 XT 16GB	60	346 mm2	N48 16GB	56 ?	300 - 350 mm2
7700 XT 12GB	54	346 mm2	N48 12GB	?	300 - 350 mm2
7600XT 16GB	32	204 mm2	N44 16GB	32	?

Below names are available

7900
7850xt
7800
7750xt
7700
7650xt

 

[linkgoron]

I am inclined to believe AMD would name RDNA4 lineup to be RX7000 series like 7850XT. Of course, naming could change at the last minute, let's wait for official announcement……

Not a chance, AIB partners won't allow it, and marketing wise nobody will buy it. It's better to rebrand the 7900xtx than keep RDNA4 with the previous gen naming. I'm not sure if this ever happened with a completely new GPU family. It'll make it much more difficult to find reviews, it'll make it harder for consumers to understand what they're buying.

Even Polaris was essentially rebranded twice.

AMD need a fresh start, they don't need to release RDNA4 with RDNA3's shadow weighing on it. They also don't need the latest 7900 card lineup ruining the already broken naming tiers by competing with Nvidia's 5060 or 5060ti cards if they're planning on a RDNA5 comeback.

 

[marees]

Not a chance, AIB partners won't allow it, and marketing wise nobody will buy it. It's better to rebrand the 7900xtx than keep RDNA4 with the previous gen naming. I'm not sure if this ever happened with a completely new GPU family. It'll make it much more difficult to find reviews, it'll make it harder for consumers to understand what they're buying.

Even Polaris was essentially rebranded twice.

AMD need a fresh start, they don't need to release RDNA4 with RDNA3's shadow weighing on it. They also don't need the latest 7900 card lineup ruining the already broken naming tiers by competing with Nvidia's 5060 or 5060ti cards if they're planning on a RDNA5 comeback.

8700xt
8700
8600xt
8500xt
Would be identical to rdna 1

But I think AMD will use XTX for top card, so it could be the below:
8700xtx
8700xt
8600xt
8500xt

Maybe they could rebrand 7900xt/7900xtx as 8800xt/8800xtx

 

[soresu]

Would be identical to rdna 1

Thought RDNA1 was just a single chip?

 

[soresu]

Even Polaris was essentially rebranded twice.

To be fair it also had at least one shrink of its high end chip, in no small part due to the fact that Vega's market coverage was sketchy at best.

 

[Tuna-Fish]

But I think AMD will use XTX for top card, so it could be the below:
8700xtx
8700xt
8600xt
8500xt

Maybe they could rebrand 7900xt/7900xtx as 8800xt/8800xtx

I think they will use 8800 for the top rdna4. It should beat/match 7800, so it's justified, and bigger numbers better.

 

[marees]

So what are known rdna 4 codes?

N48 XTX
N44 XL

What else ?

https://twitter.com/x/status/1829813911115481416

 

[marees]

Thought RDNA1 was just a single chip?

you don't remember the below 😉

AMD Radeon RX Series Specification Comparison​
	AMD Radeon RX 5500 XT​	AMD Radeon RX 5700​	AMD Radeon RX 570​	AMD Radeon RX 560​
CUs	22 (1408 SPs)​	36 (2304 SPs)​	32 (2048 SPs)​	16 (1024 SPs)​
Texture Units	88​	144​	128​	64​
ROPs	32​	64​	32​	16​
Base Clock	1607MHz​	1465MHz​	1168MHz​	1175MHz​
Game Clock	1717MHz​	1625MHz​	N/A​	N/A​
Boost Clock	1845MHz​	1725MHz​	1244MHz​	1275MHz​
Throughput (FP32)	5.2 TFLOPs​	7.95 TFLOPs​	5.1 TFLOPs​	2.6 TFLOPs​
Memory Clock	14 Gbps GDDR6​	14 Gbps GDDR6​	7Gbps GDDR5​	7 Gbps GDDR5​
Memory Bus Width	128-bit​	256-bit​	256-bit​	128-bit​
VRAM	4GB/8GB​	8GB​	4GB​	4GB​
Transistor Count	6.4B​	10.3B​	5.7B​	3B​
Typical Board Power	130W​	180W​	150W​	80W​
Manufacturing Process	TSMC 7nm​	TSMC 7nm​	GloFo 14nm​	GloFo 14nm​
Architecture	RDNA (1)​	RDNA (1)​	GCN 4​	GCN 4​
GPU	Navi 14​	Navi 10​	Polaris 10​	Polaris 11​
Launch Date	12/12/2019​	07/07/2019​	04/18/2017​	05/2017​
Launch Price	$199/$169​	$349​	$169​	$99​

The AMD Radeon RX 5500 XT Review, Feat. Sapphire Pulse: Navi For 1080p

www.anandtech.com

 

[marees]

you don't remember the below 😉

​

AMD Radeon RX Series Specification Comparison​
	AMD Radeon RX 5500 XT​	AMD Radeon RX 5700​	AMD Radeon RX 570​	AMD Radeon RX 560​
CUs	22 (1408 SPs)​	36 (2304 SPs)​	32 (2048 SPs)​	16 (1024 SPs)​
Texture Units	88​	144​	128​	64​
ROPs	32​	64​	32​	16​
Base Clock	1607MHz​	1465MHz​	1168MHz​	1175MHz​
Game Clock	1717MHz​	1625MHz​	N/A​	N/A​
Boost Clock	1845MHz​	1725MHz​	1244MHz​	1275MHz​
Throughput (FP32)	5.2 TFLOPs​	7.95 TFLOPs​	5.1 TFLOPs​	2.6 TFLOPs​
Memory Clock	14 Gbps GDDR6​	14 Gbps GDDR6​	7Gbps GDDR5​	7 Gbps GDDR5​
Memory Bus Width	128-bit​	256-bit​	256-bit​	128-bit​
VRAM	4GB/8GB​	8GB​	4GB​	4GB​
Transistor Count	6.4B​	10.3B​	5.7B​	3B​
Typical Board Power	130W​	180W​	150W​	80W​
Manufacturing Process	TSMC 7nm​	TSMC 7nm​	GloFo 14nm​	GloFo 14nm​
Architecture	RDNA (1)​	RDNA (1)​	GCN 4​	GCN 4​
GPU	Navi 14​	Navi 10​	Polaris 10​	Polaris 11​
Launch Date	12/12/2019​	07/07/2019​	04/18/2017​	05/2017​
Launch Price	$199/$169​	$349​	$169​	$99​

The AMD Radeon RX 5500 XT Review, Feat. Sapphire Pulse: Navi For 1080p

www.anandtech.com

Or this one

AMD Radeon RX Series Specification Comparison​
	AMD Radeon RX 5700 XT​	AMD Radeon RX 5700​	AMD Radeon RX 590​	AMD Radeon RX 570​
Stream Processors	2560 (40 CUs)​	2304 (36 CUs)​	2304 (36 CUs)​	2048 (32 CUs)​
Texture Units	160​	144​	144​	128​
ROPs	64​	64​	32​	32​
Base Clock	1605MHz​	1465MHz​	1469MHz​	1168MHz​
Game Clock	1755MHz​	1625MHz​	N/A​	N/A​
Boost Clock	1905MHz​	1725MHz​	1545MHz​	1244MHz​
Throughput (FP32)	9.75 TFLOPs​	7.9 TFLOPs​	7.1 TFLOPs​	5.1 TFLOPs​
Memory Clock	14 Gbps GDDR6​	14 Gbps GDDR6​	8 Gbps GDDR5​	7 Gbps GDDR5​
Memory Bus Width	256-bit​	256-bit​	256-bit​	256-bit​
VRAM	8GB​	8GB​	8GB​	4GB​
Transistor Count	10.3B​	10.3B​	5.7B​	5.7B​
Typical Board Power	225W​	180W​	225W​	150W​
Manufacturing Process	TSMC 7nm​	TSMC 7nm​	GloFo/Samsung 12nm​	GloFo 14nm​
Architecture	RDNA (1)​	RDNA (1)​	GCN 4​	GCN 4​
GPU	Navi 10​	Navi 10​	Polaris 30​	Polaris 10​
Launch Date	07/07/2019​	07/07/2019​	11/15/2018​	08/04/2016​
Launch Price	$449 $399​	$379 $349​	$279​	$179​

AMD Adjusts Launch Price of Radeon RX 5700 Series: XT Down To $399, Standard Down To $349

www.anandtech.com

 

[marees]

Or this one

​

AMD Radeon RX Series Specification Comparison​
	AMD Radeon RX 5700 XT​	AMD Radeon RX 5700​	AMD Radeon RX 590​	AMD Radeon RX 570​
Stream Processors	2560 (40 CUs)​	2304 (36 CUs)​	2304 (36 CUs)​	2048 (32 CUs)​
Texture Units	160​	144​	144​	128​
ROPs	64​	64​	32​	32​
Base Clock	1605MHz​	1465MHz​	1469MHz​	1168MHz​
Game Clock	1755MHz​	1625MHz​	N/A​	N/A​
Boost Clock	1905MHz​	1725MHz​	1545MHz​	1244MHz​
Throughput (FP32)	9.75 TFLOPs​	7.9 TFLOPs​	7.1 TFLOPs​	5.1 TFLOPs​
Memory Clock	14 Gbps GDDR6​	14 Gbps GDDR6​	8 Gbps GDDR5​	7 Gbps GDDR5​
Memory Bus Width	256-bit​	256-bit​	256-bit​	256-bit​
VRAM	8GB​	8GB​	8GB​	4GB​
Transistor Count	10.3B​	10.3B​	5.7B​	5.7B​
Typical Board Power	225W​	180W​	225W​	150W​
Manufacturing Process	TSMC 7nm​	TSMC 7nm​	GloFo/Samsung 12nm​	GloFo 14nm​
Architecture	RDNA (1)​	RDNA (1)​	GCN 4​	GCN 4​
GPU	Navi 10​	Navi 10​	Polaris 30​	Polaris 10​
Launch Date	07/07/2019​	07/07/2019​	11/15/2018​	08/04/2016​
Launch Price	$449 $399​	$379 $349​	$279​	$179​

AMD Adjusts Launch Price of Radeon RX 5700 Series: XT Down To $399, Standard Down To $349

www.anandtech.com

And this one

AMD Radeon RX Series Specification Comparison​
	AMD Radeon RX 5600 XT​	AMD Radeon RX 5700​	AMD Radeon RX 5500 XT​	AMD Radeon RX 590​
CUs	36 (2304 SPs)​	36 (2304 SPs)​	22 (1408 SPs)​	36 (2304 SPs)​
Texture Units	144​	144​	88​	144​
ROPs	64​	64​	32​	32​
Base Clock	1247MHz?​	1465MHz​	1607MHz​	1469MHz​
Game Clock	1375MHz​	1625MHz​	1717MHz​	N/A​
Boost Clock	1560MHz​	1725MHz​	1845MHz​	1545MHz​
Throughput (FP32)	7.2 TFLOPs​	7.95 TFLOPs​	5.2 TFLOPs​	7.1 TFLOPs​
Memory Clock	12 Gbps GDDR6​	14 Gbps GDDR6​	14 Gbps GDDR6​	8 Gbps GDDR5​
Memory Bus Width	192-bit​	256-bit​	128-bit​	256-bit​
VRAM	6GB​	8GB​	4GB/8GB​	8GB​
Transistor Count	10.3B​	10.3B​	6.4B​	5.7B​
Typical Board Power	150W​	180W​	130W​	225W​
Manufacturing Process	TSMC 7nm​	TSMC 7nm​	TSMC 7nm​	GloFo/Samsung 12nm​
Architecture	RDNA (1)​	RDNA (1)​	RDNA (1)​	GCN 4​
GPU	Navi 10​	Navi 10​	Navi 14​	Polaris 30​
Launch Date	01/21/2020​	07/07/2019​	12/12/2019​	11/15/2018​
Launch Price	$279​	$349​	$199/$169​	$279​

The AMD Radeon RX 5600 XT Review, Feat. Sapphire Pulse: A New Challenger For Mainstream Gaming

www.anandtech.com

 

[linkgoron]

Thought RDNA1 was just a single chip?

RDNA1 had 5700XT et al. (Navi 10) and 5500XT et al. (Navi 14)

To be fair it also had at least one shrink of its high end chip, in no small part due to the fact that Vega's market coverage was sketchy at best.

You're just making my point stronger. Even a relatively minor 12nm "shrink" got a new lineup (RX 580 was maybe 10% faster than RX 480). They could have easily done 485 instead. Grenada which was essentially Hawaii also got the R9 390 rebrand.

 

[soresu]

Hopefully with N31 being such a conservative CU increase we will see at least 1.33-1.5x more CUs on the RDNA5 hi end, possibly higher still for the enthusiast SKU if chiplets makes a super hi CU count possible without exploding overheads.

 

[soresu]

You're just making my point stronger. Even a relatively minor 12nm "shrink" got a new lineup (RX 580 was maybe 10% faster than RX 480). They could have easily done 485 instead. Grenada which was essentially Hawaii also got the R9 390 rebrand.

12nm came with RX 590/Polaris 30.