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

[exquisitechar]

199-250$, IMO.

199$ would be perfect, but, you know, 7 nm process...

Yeah, the full fat chip probably won't be just $200 at launch. That would be amazing, though.

 

[Glo.]

Yeah, the full fat chip probably won't be just $200 at launch. That would be amazing, though.

You know, 160 mm2 die, with 4 GDDR6 chips. AMD may decide to once again undercut Nvidia and completely outplay them. Because no way can Nvidia offer anything similar, if Navi 14 would be 199$, considering the closest thing in performance would be GTX 1660 Ti. A GPU that costs 280$.

 

[fleshconsumed]

Well, I'm guessing shifting price brackets upward is finally complete then. I hope that at least NAVI will force some price competition into the market.

Although if I was bolder I shouldn't have sold my AMD stock two years ago, it would have paid for quite a few videocards.

 

[exquisitechar]

You know, 160 mm2 die, with 4 GDDR6 chips. AMD may decide to once again undercut Nvidia and completely outplay them. Because no way can Nvidia offer anything similar, if Navi 14 would be 199$, considering the closest thing in performance would be GTX 1660 Ti. A GPU that costs 280$.

Actually, yeah, it's possible.

 

[crisium]

Oh if that 24 CU / 1536 shaders is true that's very nice. 1660 series competitor let's go.

I hope they have versions with more than 4GB. As I understand it, it's possible to get a 128-bit or 256-bit GDDR6 cards to have 6GB (or 12GB) of VRAM at full speed with no shenanigans, we just haven't seen it. 6GB on the 1660 series is a good fit, and I'd like AMD to match. Or do 8GB and one up them, though that's extra $ on that 2GB.

 

[mohit9206]

Yea it cannot be 4gb. Days of selling 4gb card for 200 is gone.

 

[SPBHM]

couldn't they go for 256bit GDDR5 vs 192bit GDDR5/6 on the Nvidia stuff?

I'm guessing it would be cheaper, and they could have 8GB vs 6GB!?

 

[itsmydamnation]

Based off AMD data 7nm is 2x the price of 16nm per mm sq.

So whats cheaper 256 mid/upper GDDR5 with larger % of GPU being I/O or 128bit GDDR6 with a smaller % of GPU being I/O?

 

[Shivansps]

We are still missing Navi 12... Navi 14 is the chespest Navi, if it really going for $200 minimum we are in trouble. Also i expected Navi 12 to be something like 20 to 30 CU, not the 14.

 

[maddie]

H

We are still missing Navi 12... Navi 14 is the chespest Navi, if it really going for $200 minimum we are in trouble. Also i expected Navi 12 to be something like 20 to 30 CU, not the 14.

How do you reason this?

Are you thinking along the lines of Navi 10 >12>14? If so, why? It could be Navi 10>14>12.

Almost certainly they'll want a PCIe-less design, which will be cheaper than this.

 

[mohit9206]

I think the smallest Navi will be DOA because rx570 would make infinitely more sense any day.

 

[maddie]

I think the smallest Navi will be DOA because rx570 would make infinitely more sense any day.

From AMD's point of view?

 

[mohit9206]

From AMD's point of view?

Yes i think. Because last few years AMD has proved they don't know how to make an entry level card. So their previous gen cards at entry level prices always make more sense.
(specifically referring to pcie only cards like 260x,460,560 etc).

 

[Shivansps]

H

How do you reason this?

Are you thinking along the lines of Navi 10 >12>14? If so, why? It could be Navi 10>14>12.

Almost certainly they'll want a PCIe-less design, which will be cheaper than this.

I know it was kind of a mess, but based on they older naming, Navi 12 should be smaller than Navi 10 and Navi 14 smaller than Navi 12.

Polaris 10(RX580/RX570) > Polaris 11(RX560) > Polaris 12 (RX550/540)

But anything is possible.

 

[GodisanAtheist]

Has there been any word or reasonable speculation on the potential configuration of a big Navi 5800 series card?

Are we looking at 60CUs or 80CUs? I hope against hope that AMD has designed Navi to scale like NV GPUs where each GPC is basically a fully contained GPU and they just slap more of them together to make a more performant part.

 

[Glo.]

Oh if that 24 CU / 1536 shaders is true that's very nice. 1660 series competitor let's go.

I hope they have versions with more than 4GB. As I understand it, it's possible to get a 128-bit or 256-bit GDDR6 cards to have 6GB (or 12GB) of VRAM at full speed with no shenanigans, we just haven't seen it. 6GB on the 1660 series is a good fit, and I'd like AMD to match. Or do 8GB and one up them, though that's extra $ on that 2GB.

You cannot have 6 GB on 128 bit or 256 Bit memory bus.

Yea it cannot be 4gb. Days of selling 4gb card for 200 is gone.

It can be 4 GB. Considering the engineering sample is 4 GB chip.

 

[Paul98]

With where the 5700 xt is right now, it's looking good moving forward. Proper 3rd party cooler, and settings should be able to get a good bit higher clocks than reference.

Hopefully a driver update can increase performance on some of the weaker performing titles

 

[Shivansps]

The other thing that could be is that Navi 12 is a APU IGP.

But im expecting to be two more Navis dies for GPUs.

 

[JDG1980]

You know, 160 mm2 die, with 4 GDDR6 chips. AMD may decide to once again undercut Nvidia and completely outplay them. Because no way can Nvidia offer anything similar, if Navi 14 would be 199$, considering the closest thing in performance would be GTX 1660 Ti. A GPU that costs 280$.

I'm not sure that a 128-bit memory bus would be a viable choice for a GTX 1660 Ti competitor. The TU116 cards (1660/1660 Ti) have a 192-bit bus. Now, it appears that AMD has managed to catch up to Nvidia in memory bandwidth utilization when previously they lagged behind (both Navi 10 and TU106 have the same memory bandwidth and roughly the same performance) but I don't think they've managed to dramatically surpass Nvidia on this front.

 

[JDG1980]

Has there been any word or reasonable speculation on the potential configuration of a big Navi 5800 series card?

Are we looking at 60CUs or 80CUs? I hope against hope that AMD has designed Navi to scale like NV GPUs where each GPC is basically a fully contained GPU and they just slap more of them together to make a more performant part.

If there was a big Navi chip with 80 CUs, it might actually be able to take the performance crown. That would probably require a 512-bit memory bus, but if they do grab the crown they could charge well over $1,000.

If AMD could afford it they would probably want to do a flagship chip with 80 CUs/512-bit bus and a lesser enthusiast chip with 60 CUs/384-bit bus. The latter would clearly beat TU104 but would fall short of TU102.

 

[JDG1980]

Based off AMD data 7nm is 2x the price of 16nm per mm sq.

So whats cheaper 256 mid/upper GDDR5 with larger % of GPU being I/O or 128bit GDDR6 with a smaller % of GPU being I/O?

I/O doesn't scale well, so the 7nm GPU is going to have a larger percentage of the die devoted to it than on 14nm. That's why AMD went with 14nm for Zen 2's I/O die.

 

[Mopetar]

I'm not sure that a 128-bit memory bus would be a viable choice for a GTX 1660 Ti competitor. The TU116 cards (1660/1660 Ti) have a 192-bit bus. Now, it appears that AMD has managed to catch up to Nvidia in memory bandwidth utilization when previously they lagged behind (both Navi 10 and TU106 have the same memory bandwidth and roughly the same performance) but I don't think they've managed to dramatically surpass Nvidia on this front.

It’s hard to say much about memory bandwidth utilization if neither is likely to be memory bound.

Perhaps AMD has made some additional improvements but with Polaris it was clear that NVidia had much better compression which made for more effective use of their bandwidth.

If I recall Vega made some improvement in that area, but I don’t recall if they talked much about anything with Navi. Of course I could have just missed it.

 

[Glo.]

I'm not sure that a 128-bit memory bus would be a viable choice for a GTX 1660 Ti competitor. The TU116 cards (1660/1660 Ti) have a 192-bit bus. Now, it appears that AMD has managed to catch up to Nvidia in memory bandwidth utilization when previously they lagged behind (both Navi 10 and TU106 have the same memory bandwidth and roughly the same performance) but I don't think they've managed to dramatically surpass Nvidia on this front.

Two Words. Memory bandwidth.

128 bit 14 Gbps GDDR6 memory has 224 GB/s bandwidth. 32 GB/s more, than 192 bit, 8 Gbps GDDR5.

AMD can even go for 16 Gbps GDDR6 chips for small Navi,and be extremely close to 192 bit GDDR6 12 Gbps from GTX 1660 Ti(288 GB/s).

Navi, because it has so much cache on die, is not so much prone to memory bandwidth as previous AMD architectures, were.

There is however a theory that small Navi 14 has... 256 Bit GDDR5 memory. 8 Gbps would give it 256 GB/s bandwidth, and 9 Gbps would give it 288 GB/s, exactly the same as 12 Gbps GDDR6 gives GTX 1660 Ti. Two problems: power draw of GDDR5, which is higher than GDDR6, for single chip. 256 GB/s GDDR5 memory subsystem was taking 37-40W of power on Radeon RX 480. Second one is the cost. GDDR5 costing 8$ per chip, is making 64$ for just the chips, alone. 4 GDDR6 chips, costing 10$ are 40$ total. For this solution to make sense, GDDR5 would have to cost 5$ per chip. Which actually might not be realistic expectation...

IMO, small Navi will have 128 Bit memory bus with 224 GB/s. I wish however it has 192 bit bus, and 6 GB's...

 

[crisium]

You cannot have 6 GB on 128 bit or 256 Bit memory bus.

Hmm, older article but:

http://monitorinsider.com/GDDR6.html

"Up to now, the storage capacity of all memory chips has been a nice, clean power fo two, if you exclude error detection or correction bits.

GDDR6 breaks with that tradition and offers in-between options. The standard allows a capacity of 8 to 32 Gbit, but 12 Gb and 24 Gb are possible as well. This will probably make GPU makers happy since it will increase the ability to segment the market based on the amount of memory.

Today, a GPU with a 256-bit bus can only cleanly support 4GB, 8GB or 16GB. With GDDR6, they will also be able to support 12GB, while still maintaining a full balanced load with identical sized memories connected to each controller."

6GB should also be possible then using the right memory chip size. Of course if the memory manufacturers never bothered to make these then it's all moot.

 

[Paul98]

Yeah an 80CU would really be impressive, even a 60CU version may be able to compete with a 2080 ti depending on implementation.