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

[AtenRa]

This is an illogical way to measure a card's improvement in perf/w in so many ways.
1. New halo cards from Nvidia have no equal in performance, and therefore based on your logical have an infinite improvement in performance per watt.
2. The GTX 750 TI matched the GTX 480 in performance, but in no way was replacement to that card to anyone with a half brain.

Similar die sizes or same family hierarchy is the most logical way to measure perf/w improvements. The RX480 slotted the exact same way the 270x was with relation to price, die size, and performance amongst the rest of it's stack so therefore that is the most natural and logical comparison.

Technically we are always using perf/watt with iso perf OR iso power for consistency.

You can see this in every Lithographic process , they always compare at iso perf and iso power vs the previous process.

Now, if you want to use perf/watt for PR etc, you can use what ever context suit you best, like same die size, same prices etc. or when AMD used TDP to come to the 2.8x "perf/watt" of RX480. That was pure PR and nothing else, but actually RX 470 had at the time of release almost double the perf/watt (power ~120W for both) against R7 270X at 1080p gaming.

 

[DrMrLordX]

But you're a nobody! So you don't count!

Okay, I'll have to give you that one.

 

[Krteq]

PCGN - AMD Navi certification hints at multiple versions of the RX 5000 graphics cards

D18202
D18205
D18206
D18207
D18208

 

[Stuka87]

PCGN - AMD Navi certification hints at multiple versions of the RX 5000 graphics cards

D18202
D18205
D18206
D18207
D18208

If this is true (and it likely is), it confirms this will be a mostly full stack minus top end cards.

So could be something like (and this is a guess here):
5770
5750
5670
5650
5550

 

[Krteq]

So could be something like (and this is a guess here):
5770
5750
5670
5650
5550

I don't think so, that would be the same like Cypress based cards

You know - HD 5770 vs. RX 5770 etc.

 

[lifeblood]

If this is true (and it likely is), it confirms this will be a mostly full stack minus top end cards.

So could be something like (and this is a guess here):
5770
5750
5670
5650
5550

Based on driver leaks we know there are three dies, Navi 10, 14, and 20. Initially I thought

Navi 10 8 GB and Navi 10 4 GB
Navi 14 4GB and Navi 14 2 GB
Navi 20 8 or 16 GB

However, the article says 5 different versions of silicon which does not necessitate five dies, just that two of those dies are cut down versions of the three main dies. So I would say:

Navi 10 – 5770 & 5750
Navi 14 – 5670 & 5670
Navi 20 – 5870

The only problem is that Navi 20 isn’t until 2020 so it may be a little early for it to be certified. So maybe it’s a third Navi 14 like 5630 or 5550 as you suggest.

 

[Glo.]

Based on driver leaks we know there are three dies, Navi 10, 14, and 20. Initially I thought

Navi 10 8 GB and Navi 10 4 GB
Navi 14 4GB and Navi 14 2 GB
Navi 20 8 or 16 GB

However, the article says 5 different versions of silicon which does not necessitate five dies, just that two of those dies are cut down versions of the three main dies. So I would say:

Navi 10 – 5770 & 5750
Navi 14 – 5670 & 5670
Navi 20 – 5870

The only problem is that Navi 20 isn’t until 2020 so it may be a little early for it to be certified. So maybe it’s a third Navi 14 like 5630 or 5550 as you suggest.

Considering that GDDR6 are only 1 and 2 GB chips, unless Navi 10 is NOT 256 bit memory bus GPU, it will be impossible for AMD to offer 4 GB chip, for Navi 10. And it is actually 256 bit GDDR6 GPU.

Because of this also, expect 8 GB Navi 14 models, despite 128 bit memory bus.

 

[lifeblood]

Considering that GDDR6 are only 1 and 2 GB chips, unless Navi 10 is NOT 256 bit memory bus GPU, it will be impossible for AMD to offer 4 GB chip, for Navi 10. And it is actually 256 bit GDDR6 GPU.

Because of this also, expect 8 GB Navi 14 models, despite 128 bit memory bus.

So a Navi 14 with a 128 bit memory bus will can have no less than 8GB of memory? Or it can have either 4 or 8 GB?

I can see the lowest end Navi having no less than 4GB of memory given today's apps and all, but I can't see it having no less than 8GB. At that point AMD would probably just design it for GDDR5.

 

[Stuka87]

So a Navi 14 with a 128 bit memory bus will can have no less than 8GB of memory? Or it can have either 4 or 8 GB?

I can see the lowest end Navi having no less than 4GB of memory given today's apps and all, but I can't see it having no less than 8GB. At that point AMD would probably just design it for GDDR5.

For 256bit, the lowest amount would be 8GB, for 128bit the lowest would be 4GB. This is based on each memory chip having a dedicated 32bit bus, which was the norm with GDDR5.

Going by nVidia's designs, this seems to hold true GDDR6.

 

[Glo.]

So a Navi 14 with a 128 bit memory bus will can have no less than 8GB of memory? Or it can have either 4 or 8 GB?

I can see the lowest end Navi having no less than 4GB of memory given today's apps and all, but I can't see it having no less than 8GB. At that point AMD would probably just design it for GDDR5.

GDDR6 comes in 3 sizes. 1 GB, 2 GB, and for professional market - 4 GB, per chip. If you have 256 Bit memory bus, you will get AT LEAST 8 GB memory buffer. But you still can make 16 GB GPU, with 256 bit bus.

Same thing for 128 bit GPU. 4 GB's is minimum, and 8 GB's is possible.

For GDDR6 1 GB is the new 512 MB which was for GDDR5, and 2 GB chips are the new 1 GB(GDDR5). So don't be surprised if we will see Navi GPUs with 16 GBs for 256 bit GPU, and 8 GB for 128 bit GPU.

 

[Glo.]

Considering that Nvidia rumors started to break out, about them lowering RTX GPUs prices by 100$, I will make my prediction, which will be directly related to previous one:

Navi 10 GPUs:
1) RTX 2070+10-15% performance, for 399$.
2) 90-95% of RTX 2070 performance for 299$.
Navi 14:
1) Between GTX 1660 and GTX 1660 Ti performance, for 199$
2) RX 590 performance for 150$.

Now lets look at a rumored RTX Super lineup:

RTX 2080 Super - 3072 CUDA Cores, 16 Gbps GDDR6, 256 Bit memory bus.
RTX 2070 Super - 2560 CUDA cores, 14 Gbps GDDR6, 256 Bit memory bus.
RTX 2060 Super - 2176 CUDA cores, 14 Gbps GDDR6, 256 Bit memory bus.

Launching in the middle of July. Interesting, isn't it? .

 

[Stuka87]

Now lets look at a rumored RTX Super lineup:

RTX 2080 Super - 3072 CUDA Cores, 16 Gbps GDDR6, 256 Bit memory bus.
RTX 2070 Super - 2560 CUDA cores, 14 Gbps GDDR6, 256 Bit memory bus.
RTX 2060 Super - 2176 CUDA cores, 14 Gbps GDDR6, 256 Bit memory bus.

Launching in the middle of July. Interesting, isn't it? .

I had heard the whole 'Super' thing was just price cuts?

 

[Glo.]

I had heard the whole 'Super' thing was just price cuts?

It involves also price cuts, for non-Super Models(presumably).

 

[Mopetar]

I had heard the whole 'Super' thing was just price cuts?

That's what I'd heard as well, but it's possible that such dies could exist. It doesn't quite fit with what NVidia has done in the past (not that this is any guarantee of what they'll do in the future since they hadn't done anything like the 1660 naming either) but they've got some room in the stack for new chips. An eventual 2070 Ti feels more likely since TU104 currently doesn't have a lower bin, and I suspect that's where a 2,560 CUDA core part comes in. I suspect that if they needed to there are some TU102 dies that can't quite make the cut as a 2080 Ti but could make a new product category if necessary.

NVidia even reported the sales on these new cards were weaker than anticipated in financial earnings calls, so what they really need to do is reduce the prices. Adding more cores for the same price does improve the value, but I think consumers are largely balking at having to spend increasingly more money for what used to be low or mid-tier product categories.

 

[exquisitechar]

I doubt that Nvidia would do Super+price cuts of the base models if Navi was something very underwhelming like the alleged "slightly faster than a 2070 at $499, cut down version at $399". I don't think they would even need to do much to respond to that, aside from maybe releasing an RTX 2070ti.

If the price cuts are as rumored, we will basically be getting back to sanity in the GPU market, but I'll wait and see. In any case, I can't wait for AMD's E3 keynote.

 

[maddie]

I doubt that Nvidia would do Super+price cuts of the base models if Navi was something very underwhelming like the alleged "slightly faster than a 2070 at $499, cut down version at $399". I don't think they would even need to do much to respond to that, aside from maybe releasing an RTX 2070ti.

If the price cuts are as rumored, we will basically be getting back to sanity in the GPU market, but I'll wait and see. In any case, I can't wait for AMD's E3 keynote.

Correct. Nvidia never willingly tries to undersell AMD.

 

[jpiniero]

Now lets look at a rumored RTX Super lineup:

RTX 2080 Super - 3072 CUDA Cores, 16 Gbps GDDR6, 256 Bit memory bus.
RTX 2070 Super - 2560 CUDA cores, 14 Gbps GDDR6, 256 Bit memory bus.
RTX 2060 Super - 2176 CUDA cores, 14 Gbps GDDR6, 256 Bit memory bus.

Launching in the middle of July. Interesting, isn't it? .

I doubt that. Isn't the 2080 and 2070 as it is a full chip? That would mean you would need to go to a bigger chip for both. I thought it was just going to be clocks and faster memory. And the existing models get a price cut as they are being phased out.

 

[Glo.]

I doubt that. Isn't the 2080 and 2070 as it is a full chip? That would mean you would need to go to a bigger chip for both. I thought it was just going to be clocks and faster memory. And the existing models get a price cut as they are being phased out.

...

GT104 Full chip IS 3072 CUDA core.
As for RTX 2070 Super, yes, it requires using the GT104, cut down Chip to get 2560 CUDA cores. And that is exactly what we will get.

 

[Glo.]

I doubt that Nvidia would do Super+price cuts of the base models if Navi was something very underwhelming like the alleged "slightly faster than a 2070 at $499, cut down version at $399". I don't think they would even need to do much to respond to that, aside from maybe releasing an RTX 2070ti.

If the price cuts are as rumored, we will basically be getting back to sanity in the GPU market, but I'll wait and see. In any case, I can't wait for AMD's E3 keynote.

This is what I predicted as for Navi Lineup, which you can find in this very thread.

Navi 10 GPUs:
1) RTX 2070+10-15% performance, for 399$(Full die)
2) 90-95% of RTX 2070 performance for 299$(cut-down version)
Navi 14:
1) Between GTX 1660 and GTX 1660 Ti performance, for 199$(full die)
2) RX 590 performance for 150$(cut-down version).

Compare now this to the rumored RTX Super lineup, and rumors about price cuts, for standard RTX GPUs(RTX 2070 selling for 399$, RTX 2060 selling for 249$).

Also there are rumors, that Super GPUs will be cheaper to buy, than RTX series were launched. So by this, we can say, that RTX 2060 Super will cost 299$, with 90-95% of RTX 2070 performance, and RTX 2070 Super will launch at 399$ for RTX 2070+10-15% performance.

There is no coincidence here, guys.

 

[Dribble]

Also there are rumors, that Super GPUs will be cheaper to buy, than RTX series were launched. So by this, we can say, that RTX 2060 Super will cost 299$, with 90-95% of RTX 2070 performance, and RTX 2070 Super will launch at 399$ for RTX 2070+10-15% performance.

The rumour is the Supers go in at current RTX prices and the non supers get a price drop- so I'd expect current RTX 2060 to be $299 and super to be $349

 

[tviceman]

I doubt that. Isn't the 2080 and 2070 as it is a full chip? That would mean you would need to go to a bigger chip for both. I thought it was just going to be clocks and faster memory. And the existing models get a price cut as they are being phased out.

The 2080 is cutdown. I'm not sure about the 2070.

 

[Glo.]

The rumour is the Supers go in at current RTX prices and the non supers get a price drop- so I'd expect current RTX 2060 to be $299 and super to be $349

If the Supers will go for standard RTX price points, they will be on a disadvantage over AMD .

 

[Thala]

If the Supers will go for standard RTX price points, they will be on a disadvantage over AMD .

Lol no, AMD Navi GPUs do not even have Raytracing acceleration. And sorry, i am not going to play likes of Metro featuring raytraced global Illumination without this feature or even worse - without having the option to enable raytracing at all!
So Navi must be significantly cheaper than Turing.

 

[Mopetar]

Lol no, AMD Navi GPUs do not even have Raytracing acceleration. And sorry, i am not going to play likes of Metro featuring raytraced global Illumination without this feature or even worse - without having the option to enable raytracing at all!
So Navi must be significantly cheaper than Turing.

Unless there's a version of Navi that matches up against the 2080 Ti it's a moot discussion anyway. If they're only competing against the 2070 and below, you're not going to purchase either of those cards if you want a good RT experience either.

 

[exquisitechar]

Lol no, AMD Navi GPUs do not even have Raytracing acceleration. And sorry, i am not going to play likes of Metro featuring raytraced global Illumination without this feature or even worse - without having the option to enable raytracing at all!
So Navi must be significantly cheaper than Turing.

I believe Navi will support DXR. The performance is a whole other question.