Question Speculation: RDNA2 + CDNA Architectures thread

[uzzi38]

All die sizes are within 5mm^2. The poster here has been right on some things in the past afaik, and to his credit was the first to saying 505mm^2 for Navi21, which other people have backed up. Even still though, take the following with a pich of salt.

Navi21 - 505mm^2

Navi22 - 340mm^2

Navi23 - 240mm^2

Source is the following post: https://www.ptt.cc/bbs/PC_Shopping/M.1588075782.A.C1E.html

 

[swilli89]

Putting together a build for a friend and he's wanting to spend about $400 on a GPU and its killing me shopping around. Yes, the 5700XT is the undisputed king at that price currently, but I'm thinking AMD drastically shakes things up at that price level when the 6000 series drops in a month.

The Navi 10 replacement, the 40 CU part rumored to be sitting at around 2.5 ghz will probably slot in the $300-400 range and it may be even smaller than Navi 10. AMD has a lot of leeway in pricing given how many chips they are getting per wafer, especially on TSMC's mature 7nm node. Could we see a repeat of the 4850/4870 and see a stunner get released at $300? Imagine a card at that price level that comes within spitting range of the 2080S?

 

[soresu]

Oyvey - it seems that Intel are so into pushing OneAPI that they are even willing to invest in getting it working on both nVidia and AMD GPU's.

That bodes extremely well for open rendering if Intel push their own OptiX GPU path/ray tracing renderer toolkit competitor for OneAPI instead of CUDA.

Given their historic investment in Embree and their increased GPU focus it doesn't seem like such a stretch.

 

[Mopetar]

Putting together a build for a friend and he's wanting to spend about $400 on a GPU and its killing me shopping around. Yes, the 5700XT is the undisputed king at that price currently, but I'm thinking AMD drastically shakes things up at that price level when the 6000 series drops in a month.

NVidia is supposed to be launching some of their cards in that range towards the end of October or in November, so there should be a lot of good competition to keep the prices in check. That the 3080 performs so close to the 3090 suggests that the 3070 and 3060 cards should have excellent performance as well since they won't hit the same issues that hold the 3090 back.

 

[blckgrffn]

Putting together a build for a friend and he's wanting to spend about $400 on a GPU and its killing me shopping around. Yes, the 5700XT is the undisputed king at that price currently, but I'm thinking AMD drastically shakes things up at that price level when the 6000 series drops in a month.

The Navi 10 replacement, the 40 CU part rumored to be sitting at around 2.5 ghz will probably slot in the $300-400 range and it may be even smaller than Navi 10. AMD has a lot of leeway in pricing given how many chips they are getting per wafer, especially on TSMC's mature 7nm node. Could we see a repeat of the 4850/4870 and see a stunner get released at $300? Imagine a card at that price level that comes within spitting range of the 2080S?

It's definitely worth waiting a month on and you can always wait for better, but this is a big generational jump we are about to get. It used to happen nearly every year it felt like, but this could be a 2-3 year point where you spend money now and enjoy it for that long. I mean, the timing is really great.

If he could save up another $100 OR find a really nice monitor to go with the new PC in the meantime, Zen 3 and RNDA 2 or RDNA 2 pressuring nvidia (fingers crossed) is like right now in the scheme of things.

 

[swilli89]

NVidia is supposed to be launching some of their cards in that range towards the end of October or in November, so there should be a lot of good competition to keep the prices in check. That the 3080 performs so close to the 3090 suggests that the 3070 and 3060 cards should have excellent performance as well since they won't hit the same issues that hold the 3090 back.

I expect the 3070 cards to be around $550 since NVIDIA seems to have an issue in production at Samsung. Hopefully, AMD doesn't repeat NVIDIA's blunder of a launch and has good inventory at launch.

Having said that I will be keeping an eye on what the 3060 ends up looking like versus the ~6600 XT or whatever its named. Imagine spending $400 on a 5700xt only to see cards 50-75% faster for the same price drop a few weeks later..

 

[Mopetar]

I expect the 3070 cards to be around $550 since NVIDIA seems to have an issue in production at Samsung. Hopefully, AMD doesn't repeat NVIDIA's blunder of a launch and has good inventory at launch.

Having said that I will be keeping an eye on what the 3060 ends up looking like versus the ~6600 XT or whatever its named. Imagine spending $400 on a 5700xt only to see cards 50-75% faster for the same price drop a few weeks later..

Wasn't the price for the 3070 already announced as $500? Sure there will be some third party cards that go for higher, but if the 3070 is at the far end of what it can achieve out of the gate there isn't a lot to gain from the premium AIB models.

I'm definitely getting a new rig this fall though. For one reason or another I've been holding off but it seems like all the stars are aligning with good new hardware options and good value for money to boot.

 

[swilli89]

Wasn't the price for the 3070 already announced as $500? Sure there will be some third party cards that go for higher, but if the 3070 is at the far end of what it can achieve out of the gate there isn't a lot to gain from the premium AIB models.

I'm definitely getting a new rig this fall though. For one reason or another I've been holding off but it seems like all the stars are aligning with good new hardware options and good value for money to boot.

Yes the 3070 was announced at $500 but scarcity might drive up real retail prices of AIB. I'm not sure there will be a FE 3070 and if there is I hope supply issues are sorted out by then.

This fall is shaping up to be an AMAZING time to build a new rig. When is the last time we saw potentially game changing CPUs and GPUs released from the major players all together in such a short time span? Myself - I simply plan to upgrade my 1070 Ti to whatever is available in November for $500 or less, and my Ryzen 2700 to a potential 5700/5800 @ $350.

 

[Guru]

RX 6900XT seems like a beast, 80cu's, 96rops, 2.2GHz boost clock, 16GB 16gbps GDDR6 memory and 256bit interface.

If its 70% faster than the RX 5700xt it would put it at RTX 3080 levels in terms of performance, but consuming less power, being build on a small node and AMD being able to adjust prices very competitively.

I expect if its as fast the the RTX 3080 to be the same price range though.

 

[swilli89]

RX 6900XT seems like a beast, 80cu's, 96rops, 2.2GHz boost clock, 16GB 16gbps GDDR6 memory and 256bit interface.

If its 70% faster than the RX 5700xt it would put it at RTX 3080 levels in terms of performance, but consuming less power, being build on a small node and AMD being able to adjust prices very competitively.

I expect if its as fast the the RTX 3080 to be the same price range though.

The top 80CU part surely must come with an HBM variant. If a much lower clockspeed 40CU Navi was combined with a 256-bit bus last gen, I can't imagine how a much faster clocked 80CU chip could make do with the same bud width simply from moving from GDDR6 to GDDR6X.

 

[blckgrffn]

The top 80CU part surely must come with an HBM variant. If a much lower clockspeed 40CU Navi was combined with a 256-bit bus last gen, I can't imagine how a much faster clocked 80CU chip could make do with the same bud width simply from moving from GDDR6 to GDDR6X.

GDDR6X isn't even a standard, how likely is AMD to be using it?

 

[GoodRevrnd]

The top 80CU part surely must come with an HBM variant. If a much lower clockspeed 40CU Navi was combined with a 256-bit bus last gen, I can't imagine how a much faster clocked 80CU chip could make do with the same bud width simply from moving from GDDR6 to GDDR6X.

I know it's just a dumb render in Fortnite, but seems like everyone forgot the rear heatsink anchor layout was highly suggestive of HBM.

 

[Stuka87]

No way the 80CU part has a 256bit memory interface, we have been through this. That is NOT enough bandwidth for a GPU of that size. It works fine on the 40CU 5700XT, but not a chip that is double the size. 16GB means its either a giant 512bit interface, or as most of us suspect, HBM2e.

 

[Karnak]

No way the 80CU part has a 256bit memory interface, we have been through this. That is NOT enough bandwidth for a GPU of that size. It works fine on the 40CU 5700XT, but not a chip that is double the size. 16GB means its either a giant 512bit interface, or as most of us suspect, HBM2e.

Or it's actually a 256bit interface but with the rumored cache.
Or it's actually a 256bit interface without the rumored cache - but with one extra stack of HBM2e. Like 14,4GT/s for GDDR6 and 3,6GT/s HBM2e (8GB + 8GB for a total of 16GB and 420Gb/s + 420Gb/s for a total of 920Gb/s bandwidth).

Who knows. I mean, less than a month til the official announcement and overall we still know close to nothing except for a few things thanks to MacOS. According to AMD it will be a halo product, so.

 

[Stuka87]

Or it's actually a 256bit interface but with the rumored cache.
Or it's actually a 256bit interface without the rumored cache - but with one extra stack of HBM2e. Like 14,4GT/s for GDDR6 and 3,6GT/s HBM2e (8GB + 8GB for a total of 16GB and 420Gb/s + 420Gb/s for a total of 920Gb/s bandwidth).

Who knows. I mean, less than a month til the official announcement and overall we still know close to nothing except for a few things thanks to MacOS. According to AMD it will be a halo product, so.

If there is two stacks of HBM2e (and there has to be for 16GB of RAM) then it would have a 512bit 2048bit interface, as each stack is 256bits 1024bits wide.

EDIT: Fixed the above text that is striked out.

 

[Karnak]

One stack of HBM2(e) is 1024bit, though there is some type of a "low cost option" iirc from Samsung with 512bit.

But I was specifically thinking of just mixing GDDR6 and HBM2e with 8GB of each for a total of 16GB. Prob won't happen , but I don't know if some sort of huge cache is more realistic at this point than just having a mix of GDDR6 and HBM2e. 256bit without anything extra won't happen on a 80CU chip. Pretty sure about this.

 

[raghu78]

RX 6900XT seems like a beast, 80cu's, 96rops, 2.2GHz boost clock, 16GB 16gbps GDDR6 memory and 256bit interface.

If its 70% faster than the RX 5700xt it would put it at RTX 3080 levels in terms of performance, but consuming less power, being build on a small node and AMD being able to adjust prices very competitively.

I expect if its as fast the the RTX 3080 to be the same price range though.

Navi 21 - 4SE, 8SA, 40 WGP / 80 CU, 128 ROPs, 256 bit GDDR6 or 2048 bit HBM2E, 16 GB. My guess is 2048 bit HBM2E and performance on par with RTX 3090 at 275w.

 

[swilli89]

I know it's just a dumb render in Fortnite, but seems like everyone forgot the rear heatsink anchor layout was highly suggestive of HBM.

I had no idea this was even a thing! I know its senseless to even worry about 256-bit whatever. This is AMD, they've been using HBM for over half a decade and I'm sure they know how to engineer a card with the requisite memory subsystem.

I suppose the real fear surrounding the 256-bit thing would be that AMD is not launching a halo card and is targeting the upper mid-range and below, ala 5700 XT. Fingers crossed this isn't the case.

 

[PhoBoChai]

I think all this concern about bus width and bandwidth is premature. AMD hasn't really ever released a GPU that is that badly bandwidth bottlenecked, outside of market segmentation reasons (5600XT).

 

[uzzi38]

Navi 21 - 4SE, 8SA, 40 WGP / 80 CU, 128 ROPs, 256 bit GDDR6 or 2048 bit HBM2E, 16 GB. My guess is 2048 bit HBM2E and performance on par with RTX 3090 at 275w.

Easily if they use HBM2E. I'm still quite worried on that, everyone's super inconsistent on that :/

Part of me is worries instead of aiming for the crown, they might settle for around 3080 perf with 256 bit bus. Which is still plenty fine and all, but considering this is the best shot they've had in a long time at the crown, I was really hoping they'd take it and go all the way.

I think all this concern about bus width and bandwidth is premature. AMD hasn't really ever released a GPU that is that badly bandwidth bottlenecked, outside of market segmentation reasons (5600XT).

Going to be honest, that was my biggest thought when I saw RDNA2 clocks. Thought there was no way they'd be clocking the GPUs that high if they were seriously bandwidth bound - it simply just didn't make sense. But god is the wait for more info killing me here.

 

[Mopetar]

I think all this concern about bus width and bandwidth is premature. AMD hasn't really ever released a GPU that is that badly bandwidth bottlenecked, outside of market segmentation reasons (5600XT).

The Polaris cards generally saw better FPS gains from a memory OC than anything else. Part of that was that a significant core clock wasn't possible for most people, but just the fact that better gains were (which tended to be linear for what most cards could manage) found with a memory overclock suggest there was more of a bottleneck there than other parts of the card, but that could have also been as a result of the clocks being pushed as far as they could go.

 

[Krteq]

6 DevIDs (also SKUs?) for Sienna Cichlid in recent commit:

0x73A0
0x73A2
0x73A3
0x73AB
0x73AE
0x73BF

Phoronix.com - Linux 5.9 Gets More Fixes For AMD RDNA2 GPUs, Promotes Navi 12

 

[Konan]

6 DevIDs (also SKUs?) for Sienna Cichlid in recent commit:

0x73A0
0x73A2
0x73A3
0x73AB
0x73AE
0x73BF

Phoronix.com - Linux 5.9 Gets More Fixes For AMD RDNA2 GPUs, Promotes Navi 12

These look like they line up with what Rogame found in February

https://twitter.com/x/status/1233392225750585345

 

[HurleyBird]

If they had churned out FEAR RTX I would have been a bit more impressed.

OMG yes. I have no idea how Monolith went backwards, not just in terms of gameplay but arguably along various graphical dimensions, with each FEAR game. And the original FEAR runs like total *** on modern systems too.

 

[BlitzWulf]

I know this guy isn't anyone's favorite "source", but its been a slow couple days, so why not?

He seems to be doubling down on the on die cache rumor and claims HBM for consumer cards is a no go .

Also claims that AMD put fake info in the MAC and Linux firmware dumps to throw people off.

 

[PhoBoChai]

The new cache is a given at this point, with the unified L1$ design in RDNA 1 opening possibilities for the changes that the dynamic CU cache patent and research paper showed. I don't think it will be a large ESRAM 128mb as rumored. Just larger L1$ and larger L2$ will do the job fine.