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

[Glo.]

I will brak my rules a little bit, and I will compare... Radeon VII to RTX 2080, and we will speculate a little bit what this means for Navi GPU.

At launch, Radeon VII in Strange Brigade was around 10% faster than RTX 2080, and it was one the titles in which it was faster than Nvidia's GPU. The thing is, latest Nvidia drivers have made RTX 2080 gain 15% over Radeon VII, and RIGHT NOW, R7 is 5% SLOWER than RTX 2080. Very strong showing from Nvidia here.

Here was situation at launch, in HWUXed review Radeon GPU was 10% faster.
Lately they retested with latest drivers, and this is how it behaves:

Here, right now it is 5% slower than Nvidia's GPU.

Lets take it for Navi. Unnamed Navi GPU is demoed at Computex. We can see in the upper left corners of the demo the FPS at which the GPUs achieves. RTX 2070 hovers around 115-120 FPS, Navi GPU hover around high onetwenties, low 130ish framerate.

Hardware unboxed tested RTX 2070 at launch in Strange Brigade. In 1440p, at launch the GPU scored 101 FPS.
https://www.youtube.com/watch?v=RAcsxZfROY8

If we go by FPS gains through driver updates, based on RTX 2080 Gains, we should expect, with newest RTX 2070 drivers to make the GPU perform on similar levels to the demo. Now.

RTX 2070 at launch in Strange Brigade 1440p Ultra details, scored 101 FPS, average.
RTX 2080, at that time scored in the same resolution and details scored 125 FPS.
Radeon & was approximately 10%, which should yield 138-140 FPS in the same settings.
After driver updates, Radeon GPU was 5% slower(!) than RTX 2080. Which must've meant that the GPU scored at least 145-150 FPS, in the same res, and detail. 20-25 FPS gains compared to launch drivers.
RTX 2070 in the Computex Demo was getting 115-120 FPS. Again, around 20 FPS gains, so they are in line, with RTX 2080 gains, potentially.

All of this means, that at this moment, demoed, unnamed Navi GPUmay be 5-6% slower in Strange Brigade than Radeon VII GPU. Second thing, is that it may not be as bad as people say.

But all of this is just speculation. Food for thought. The key right now to understand is that, Strange Brigade is not that AMD biased title, and AMD was able to optimize their drivers, for this game, and make RTX 2080 faster in this title, than Radeon VII. If AMD used latest drivers for RTX 2070, which, at first glance appears to be the case, then Navi GPU is really fast, for its die size.

And all of those calculation based on Radeon VII performance exclude even the performance gains with latest drivers of Radeon VII GPU.

 

[GodisanAtheist]

Kinda feel the hype train starting up again. Thread looks a lot like the "there is no reason Polaris should be any slower than the GTX 1070" threads of yesteryear.

I'm as excited as the next guy for a new GPU, any new GPU to launch (1080ti performance for ~$350 and you can take my damn money while you're at it) but gotta keep expectations tempered to avoid unjustified disappointment.

AMD doesn't just have to put together a solid card, they have to really stick the landing on some historical soft spots: solid launch drivers and functional cooling/coolers that don't ruin a good thing.

If AMD 290's this launch I'd say they're beyond redemption...

 

[railven]

Kinda feel the hype train starting up again. Thread looks a lot like the "there is no reason Polaris should be any slower than the GTX 1070" threads of yesteryear.

With dismissive "it's a clean sheet you can't compare it to anything else" hopefully firmly behind us, we can get back to the speculation train. The talks of how many CU's is interesting. Though I can't find anything myself that hints at it, if AMD has a bigger/more enabled chip on the side, that's always a plus!

AMD doesn't just have to put together a solid card, they have to really stick the landing on some historical soft spots: solid launch drivers and functional cooling/coolers that don't ruin a good thing.

I feel like more than anything it's the launches of previous cards that tainted a lot of good will. Radeon 7 in all honestly felt like a rushed odd ball last minute decision to me. This RDNA sort of stinks of it to me too. I won't be surprised if RDNA is "GCN" with the "CN" removed thus a lot of the bloat and we're as one previous "leaker" hinted at, a design closer to NV's.

I still shake my head at the branding genius that coined "RDNA". I don't even know why it bugs me so much. It feels so ... lazy.

 

[DrMrLordX]

Navi 10 is 40 CU chip. Navi 14 is 20 CU chip. THATS IT.

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...vi-graphics-cards-at-e3.2564009/post-39829881

I know you didn't like his post originally, but it still makes a lot of sense to me. There' no evidence they they did or didn't "burn area".

If AMD did produce 40CU dice without cutting any 56CU dice at this time, then that lowers the BoM of the "2070 killer" even further.

The BOM pricing for the HBM2 however was taken from a rumor site where they just doubled the price of what memory cost for Vega64. Which is an entirely inaccurate way to determine prices.

And yet nobody came up with an alternative calculation. Somehow we knew how much the HBM2 cost for RX Vega64, but we had no idea how pricing had progressed by 2018 (Radeon Instinct Mi50). You could apply depreciation estimates and get HBM2 prices in the ballpark of $180-$200 instead of the ~$230 or so reported by "rumor sites" based on known HBM2 costs from 2017. We don't really know the rate of depreciation of HBM2 value, because unlike DRAM, HBM2 is a lower-volume product used in high-priced applications.

I'm still waiting for someone to come up with a halfway-plausible BoM estimate for either Navi10 or Navi14.

 

[DisEnchantment]

Spot on. This View attachment 6721View attachment 6722 is at least a 256mm^2 chip, if not 275mm^2, as predicted by Ryan;

which makes it 25% smaller than the other 7mm AMD Chip (Vega 20)
@ 25% less than Vega 20 (331mm^2/64CUs) = ~ 250mm^2 / 48CUs

However, @ 275mm^2 or higher, it would be a 56 CU chip
@ 15% less than Vega 20 (331mm^2/64CUs) = ~ 280mm^2 / 56CUs

Also, if this is a 56CU chip and the yields are not good, then AMD can cut it thrice; in 8CU increments (56 / 48 / 40 CU parts)

They can launch the 40 & 48 CU chips now and build up an inventory of full dies. Launch these cut down parts now, where 40 CU competes with Rx2060 and 48CU part competes with Rx2070.

Then, once they've built up a healthy inventory & Nvidia launches a Rx2070Ti, they can simply release the full 56CU chip.

255 mm2 From Mr. Reliable (so far)

https://twitter.com/x/status/1134048033098698752

 

[DisEnchantment]

If Navi has Infinity Fabric inside for linking to multiple blocks it could explain a bit on the lower Power Efficiency even after migrating to 7nm

 

[Glo.]

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...vi-graphics-cards-at-e3.2564009/post-39829881

I know you didn't like his post originally, but it still makes a lot of sense to me. There' no evidence they they did or didn't "burn area".

If AMD did produce 40CU dice without cutting any 56CU dice at this time, then that lowers the BoM of the "2070 killer" even further.

They did burn a lot of area. The point of Navi is not throwing more CUs, at the problem, but making those CUs more capable. Which consumes space. Navi is Maxwell and Pascal-like release for AMD, at the same time.

 

[Timorous]

Assuming the rework of the GPU does not impact density (which is a big assumption) then it would suggest that this GPU is around 10B transistors.

R7 has a similar number of transistors to the 2080 and performance is in the ballpark despite not all of the CUs being turned on (so some transistors are useless). The 2070 has around 10B transistors so that to me suggests that Navi could be in the same ballpark if it has similar density and perf/transistor as R7.

It could also have better perf/transistor than R7 and lower density and still hit the 2070 performance window.

I think anywhere from 90% to 110% of the 2070 is where Navi will land and I hope that it ends up being closer to 110%.

 

[Head1985]

255 mm2 From Mr. Reliable (so far)

https://twitter.com/x/status/1134048033098698752

Not much bigger than polaris.Well it should be priced similar because it is clearly polaris replacement.But because Nvdia crazy turing prices i fear AMD will follow them instead compete with them.

 

[krumme]

Not much bigger than polaris.Well it should be priced similar because it is clearly polaris replacement.But because Nvdia crazy turing prices i fear AMD will follow them instead compete with them.

Price is determined by supply demand. Besides 7nm is far more expensive.
Competitors typically have more or less same price. It would be idiotic to cut prices in half to compete.
You easily get 4 3900x for the price of one of those new gpu. 2000 usd revenue if not more. Cant say there is much motivation to cut prices to suit some enthusiast that typically want to buy another brands gpu anyway.

 

[Head1985]

Price is determined by supply demand. Besides 7nm is far more expensive.

Yes yes yes 7nm is far more expensive.But 16/14nm was also far more expensive than 28nm and 28nm was far more expensive than 40nm.

 

[lifeblood]

Unfortunately I’m now much more skeptical after Computex.

Facts:

1. AMD shows the 5700 10% faster than RTX 2070 on a single game.

2. Some pictures have surfaced of unnamed but assumed to be Navi graphic cards Asrock displayed. They have three fans.

Interpretation:

1. AMD would not show Navi vs the competition on any game, but on the game showing Navi in the best light possible. They have done this frequently in the past, it can be safely assumed they’re doing it again.

2. You know why you put three fans on a video card? Because you need them. AMD has historically overclocked their GPU’s to squeeze the max performance out of them. They have done this frequently in the past, it can be safely assumed they’re doing it again.

Conclusion:

Navi/5700 will perform on average 5% - 10% below the RTX 2070 and use lots of power to do it. We already knew it will not be feature equivalent (no RTS). The only question left in my mind (barring new evidence or convincing argument) is price. Since the RTX 2070 is (about) $500 I estimate Navi at $450.


Edited to correct 2070 current price

 

[lifeblood]

I was thinking about the “new” RDNA architecture.

Facts:

1. AMD’s older roadmaps showed Navi as a new architecture. This later changed and now Arcturus or whatever the next card is called has the new architecture.

2. A rumor surfaced that the console folks (Sony/Microsoft) wanted the new GPU to be easily backwards compatible to make porting old games to it easy

3. AMD says Navi is a new architecture, call it RDNA

Interpretation:

1. It started as a clean sheet but much of GCN needed to be pulled back in to give it that easy backward compatibility.

2. Interesting name, Radeon DNA. It’s like they're trying to say it’s new, but still possess a lot of what came before.

Conclusion:

When is a “new” architecture really new, and when is it the older architecture with some major/minor updates? Can’t blame AMD for this no matter what the truth is. It’s always a thin line trying to build something new and radical while still retaining the capability to run the old stuff efficiently.

 

[Glo.]

I was thinking about the “new” RDNA architecture.

Facts:


2. A rumor surfaced that the console folks (Sony/Microsoft) wanted the new GPU to be easily backwards compatible to make porting old games to it easy

Any GPU architecture that will use GCN ISA will be backwards compatible.

 

[Saylick]

I was thinking about the “new” RDNA architecture.

Facts:

1. AMD’s older roadmaps showed Navi as a new architecture. This later changed and now Arcturus or whatever the next card is called has the new architecture.

2. A rumor surfaced that the console folks (Sony/Microsoft) wanted the new GPU to be easily backwards compatible to make porting old games to it easy

3. AMD says Navi is a new architecture, call it RDNA

Interpretation:

1. It started as a clean sheet but much of GCN needed to be pulled back in to give it that easy backward compatibility.

2. Interesting name, Radeon DNA. It’s like they're trying to say it’s new, but still possess a lot of what came before.

Conclusion:

When is a “new” architecture really new, and when is it the older architecture with some major/minor updates? Can’t blame AMD for this no matter what the truth is. It’s always a thin line trying to build something new and radical while still retaining the capability to run the old stuff efficiently.

Let's use x86 as an example. We've seen so many architectures from Intel and AMD which run x86, which is an ISA. Would you not consider each generation of Intel/AMD products as a new architecture? Navi/RDNA is a new architecture vs. Vega in the same way as Haswell is a new architecture vs. Sandy Bridge. You take what was there previously and expand upon it. To follow in the same sentiments as Glo, GCN is analogous to an ISA. What came before GCN was Terrascale VLIW4 and before that was Terrascale VLIW5.

 

[lifeblood]

Let's use x86 as an example. We've seen so many architectures from Intel and AMD which run x86, which is an ISA. Would you not consider each generation of Intel/AMD products as a new architecture? Navi/RDNA is a new architecture vs. Vega in the same way as Haswell is a new architecture vs. Sandy Bridge. You take what was there previously and expand upon it. To follow in the same sentiments as Glo, GCN is analogous to an ISA. What came before GCN was Terrascale VLIW4 and before that was Terrascale VLIW5.

I completely agree with your point. "New" or "old" is a matter semantics. Haswell is an updated version of Sandy Bridge. But Pentium was just an updated version of 486, it was just WAY more modified. It was still x86, but it was a pretty different version of x86 compared to what had come before.

I don't know how different RDNA is from GCN. is it Haswell vs SB or was it Pentium vs 486? No idea. That's one of the questions I have and why I don't really offer a strong opinion on either.

 

[Stuka87]

Unfortunately I’m now much more skeptical after Computex.

Facts:

1. AMD shows the 5700 10% faster than RTX 2070 on a single game.

2. Some pictures have surfaced of unnamed but assumed to be Navi graphic cards Asrock displayed. They have three fans.

Second Post:
2. A rumor surfaced that the console folks (Sony/Microsoft) wanted the new GPU to be easily backwards compatible to make porting old games to it easy

They did show it in a game that nVidia just patched for and got way more FPS, and they still were ahead (the FPS of the RTX card match benchmarks of post-patch).

There are a ton of RTX 2070 cards with 3 fans. This is hardly unheard of.

Not sure you should put rumors under 'facts'. Yes its a fact there is a rumor, but the contents of the rumor is not proven fact. But MS/Sony will not have to port any games at all from PS4/XBone to run on their next consoles. As both old and new are running compatible ISA's.

 

[Saylick]

I completely agree with your point. "New" or "old" is a matter semantics. Haswell is an updated version of Sandy Bridge. But Pentium was just an updated version of 486, it was just WAY more modified. It was still x86, but it was a pretty different version of x86 compared to what had come before.

I don't know how different RDNA is from GCN. is it Haswell vs SB or was it Pentium vs 486? No idea. That's one of the questions I have and why I don't really offer a strong opinion on either.

I'd be willing to bet that RDNA is closer to SB vs. Nehalem than Zen vs. Bulldozer. SB shares a lot of similarities with Nehalem, especially from an organizational level, but also added a lot of new features like the micro-op cache, AVX, and even went so far as using a ring-based interconnect. Zen vs. Bulldozer changed much more, obviously. I do think that RDNA is a bigger change than Haswell vs. SB, though.

 

[lifeblood]

They did show it in a game that nVidia just patched for and got way more FPS, and they still were ahead (the FPS of the RTX card match benchmarks of post-patch).

There are a ton of RTX 2070 cards with 3 fans. This is hardly unheard of.

Not sure you should put rumors under 'facts'. Yes its a fact there is a rumor, but the contents of the rumor is not proven fact. But MS/Sony will not have to port any games at all from PS4/XBone to run on their next consoles. As both old and new are running compatible ISA's.

"There are a ton of RTX 2070 cards with 3 fans. This is hardly unheard of." Yes, but AMD also has a history of selling hot chips. those two items together point to something more.

"Yes its a fact there is a rumor,..." Which is what I meant. I'm just tying disparate rumors together.

Games run much closer to the metal on consoles than on Windows. Optimizations are made for the specific hardware targeted. Two GPU’s can support the same ISA but in different ways. The more a developer optimizes for one GPU the more they have to modify to port to the other for optimal performance. Again I say consoles, not PC's. Their is far more abstraction between hardware and app on PC's.

And you have the right to say I'm completely wrong as I am not a developer. The most I do is a bit of scripting, but based on a few developers I’ve chatted with or read about, there is a certain amount of work that must be done when moving apps to different consoles.

 

[caswow]

Unfortunately I’m now much more skeptical after Computex.

Facts:

1. AMD shows the 5700 10% faster than RTX 2070 on a single game.

2. Some pictures have surfaced of unnamed but assumed to be Navi graphic cards Asrock displayed. They have three fans.

Interpretation:

1. AMD would not show Navi vs the competition on any game, but on the game showing Navi in the best light possible. They have done this frequently in the past, it can be safely assumed they’re doing it again.

2. You know why you put three fans on a video card? Because you need them. AMD has historically overclocked their GPU’s to squeeze the max performance out of them. They have done this frequently in the past, it can be safely assumed they’re doing it again.

Conclusion:

Navi/5700 will perform on average 5% - 10% below the RTX 2070 and use lots of power to do it. We already knew it will not be feature equivalent (no RTS). The only question left in my mind (barring new evidence or convincing argument) is price. Since the RTX 2070 is (about) $500 I estimate Navi at $450.


Edited to correct 2070 current price

how about asus rog 1060 with 3 fans. i dont think it was a powerhog

 

[lifeblood]

how about asus rog 1060 with 3 fans. i dont think it was a powerhog

Yes, but the asus rog 1060 with 3 fans is a custom card. The FE version of RTX 2060 was a 2 fan setup (was their even a "reference" 2060 released? I can't remember). I am assuming the Asrock cards seen a Computex are semi custom (i.e. custom coolers). They did seem a way too pretty to be reference cooled cards.

And again, yes, some companies do sell cards with more cooling then they need, but add "concept card has three fans" plus "AMD historically sells hot GPU's" and you start seeing something more.

 

[Kenmitch]

Unfortunately I’m now much more skeptical after Computex.

Facts:

1. AMD shows the 5700 10% faster than RTX 2070 on a single game.

2. Some pictures have surfaced of unnamed but assumed to be Navi graphic cards Asrock displayed. They have three fans.

Interpretation:

1. AMD would not show Navi vs the competition on any game, but on the game showing Navi in the best light possible. They have done this frequently in the past, it can be safely assumed they’re doing it again.

2. You know why you put three fans on a video card? Because you need them. AMD has historically overclocked their GPU’s to squeeze the max performance out of them. They have done this frequently in the past, it can be safely assumed they’re doing it again.

Conclusion:

Navi/5700 will perform on average 5% - 10% below the RTX 2070 and use lots of power to do it. We already knew it will not be feature equivalent (no RTS). The only question left in my mind (barring new evidence or convincing argument) is price. Since the RTX 2070 is (about) $500 I estimate Navi at $450.


Edited to correct 2070 current price

Maybe it's just to supply enough airflow while keeping the thermals and sound levels in check. I wouldn't jump to any conclusions as there really isn't enough data available at this time to determine one way or another.

 

[railven]

Games run much closer to the metal on consoles than on Windows. Optimizations are made for the specific hardware targeted. Two GPU’s can support the same ISA but in different ways. The more a developer optimizes for one GPU the more they have to modify to port to the other for optimal performance. Again I say consoles, not PC's. Their is far more abstraction between hardware and app on PC's.

I'd look at how consoles are emulated on PC. Usually through brunt force considering how weak the hardware of the console being emulated usually is. Or even how changes in a game code affected performance greatly for Mantle users almost breaking it.

On the one hand, I can see the console side not wanting to have these kind of issues in porting games. On the other hand, I have seen enough "Remasters" to not instantly discount the console side to jump on a "4K Remaster" non-sense just to resell the same games yet again, but of course now in 4K!!!

 

[guskline]

railven, just sent you a PM about your custom loop.

 

[Stuka87]

"There are a ton of RTX 2070 cards with 3 fans. This is hardly unheard of." Yes, but AMD also has a history of selling hot chips. those two items together point to something more.

"Yes its a fact there is a rumor,..." Which is what I meant. I'm just tying disparate rumors together.

Games run much closer to the metal on consoles than on Windows. Optimizations are made for the specific hardware targeted. Two GPU’s can support the same ISA but in different ways. The more a developer optimizes for one GPU the more they have to modify to port to the other for optimal performance. Again I say consoles, not PC's. Their is far more abstraction between hardware and app on PC's.

And you have the right to say I'm completely wrong as I am not a developer. The most I do is a bit of scripting, but based on a few developers I’ve chatted with or read about, there is a certain amount of work that must be done when moving apps to different consoles.

I do not have any experience with PS4 development, but at least for XBoxOne MS stressed being able to have the same game run on XBoxOne and Windows 10, and that DX12 ran natively on XBox One. Being able to utilize Windows dev tools for XBox One made developing for it a whole lot easier than previous XBox consoles.