When will we see AMDs next gen

S

Sunin

[H]ard|DCer of the Month - August 2008
Joined
Dec 27, 2005
Messages
3,421
Now that nvidia struck first, when will we see amd's response. I have been a fan of both camps as long as they are a good cost to performance. I have a 270x, 7850, 650ti currently running so looking to build high end rig at year end and am drooling at the next gen coming!
 
NVIDIA only launched their next gen so they can test the architecture. I don't think we will see the "real" next gen from either AMD or NVIDIA until the 20nm process is available. 28nm is tapped out.
 
blinx said:
Wait, the new AMD cards aren't considered next gen?
Click to expand...
They are a revision of the existing architecture. GCN 1.1 basically. I believe OP is referring to AMD's next architecture. I think we will probably see them stick with GCN derivatives for some time. Given how hot the Hawaii chips are running on 28nm process, they should be able to get some good gains just by a die shrink.
 
Rizen said:
They are a revision of the existing architecture. GCN 1.1 basically. I believe OP is referring to AMD's next architecture. I think we will probably see them stick with GCN derivatives for some time. Given how hot the Hawaii chips are running on 28nm process, they should be able to get some good gains just by a die shrink.
Click to expand...

28nm -> 20nm will only offer a 20-25% savings in power consumption while potentially increasing transistor density 2x and decreasing die size ~40%.
So they need to have similar efficiency gains as GM107 if they want to build anything larger than 350mm2.
 
Last edited:
LordEC911 said:
28nm -> 20nm will only offer a 20-25% save in power consumption while potentially increasing transistor density 2x and decreasing die size.~40%.
Click to expand...

Saving only 20-25% in power consumption!!! that is huge There will also be the increased speed of the chip. AMD was hoping that the 20nm process would be ready for the current crop but the process was delayed. That would have given them a killer card and they knew it. With everything else going on though, they had to release the R9 early. Who knew the mining craze would work out the way it has? When they come out with chips on 20nm you might want to pre-order

That is the basis of my dilemma, get a crossfire setup now or wait.
 
hcforde said:
Only 20-25% in power consumption!!! that is huge.
Click to expand...

Not really. Take the current chip, reduce its power (let's say 200w) and cut it down to 160w. Now take the current die size, reduce that to 60%.

You'll see the net heat output per square mm of die area has actually increased drastically,

Not the only thing in the equation, but no, a 20% efficiency gain over a die shrink isn't always an ace in the sleeve.
 
Pirate Islands Q4 2014, yes?

AKA Mining Islands. AKA Price-Gouge Islands.
idk I'm not good with puns.
 
Arcygenical said:
Not really. Take the current chip, reduce its power (let's say 200w) and cut it down to 160w. Now take the current die size, reduce that to 60%.

You'll see the net heat output per square mm of die area has actually increased drastically,

Not the only thing in the equation, but no, a 20% efficiency gain over a die shrink isn't always an ace in the sleeve.
Click to expand...

I understand that, I was commenting on the fact that the statement seems to minimize the savings by saying "ONLY". However,you also have to take into consideration the efficiency of the circuits when you have a die shrink. that is where a decrease in the heat comes from. The longer a circuit is the more heat that is produced.
 
hcforde said:
I understand that, I was commenting on the fact that the statement seems to minimize the savings by saying "ONLY". However,you also have to take into consideration the efficiency of the circuits when you have a die shrink. that is where a decrease in the heat comes from. The longer a circuit is the more heat that is produced.
Click to expand...

The "only" being included because other similar node shrinks offered 30-35% maybe a little more in power savings. Remember that these numbers come directly from the foundry so while I'm sure they aren't stretched too far they are likely of the "best case" variety.
http://www.tsmc.com/english/dedicatedFoundry/technology/20nm.htm

Correct, that 20-25% is what is saved from a die shrink from 28nm to 20nm.


Let's take Hawaii, 438mm2 @ 28nm ~300w TDP (typically gaming is less but using round numbers is easier)

Hypothetical numbers- ( no way accurate )
We will say they want to increase die size about 30% and they see about the same increase in TDP @ 28nm.
So on 28nm we are looking at 570mm2 and a TDP of 390w.
Let's shrink this down to 20nm.
Not everything shrinks linearly but let's say 40% savings in die size and go with the 25% savings in TDP.
570mm2 @ 28nm becomes 342mm2 @ 20nm and 390w @ 28nm comes back down to ~293w @ 20nm.

So they are going to be completely limited by TDP on a chip that is about the size of Tahiti.
That is why you have to fine tune your designs to the specific process, make architectural changes and tweaks to milk whatever efficiency you can because solely relying on new process nodes to advance your performance and power efficiency hasn't been working the last 3-4 generations.
 
Last edited:
Thats not entirely true.

the R9-290/290x, and the 260/260x are new silcon, the others are pitcairn/tahiti rebrands

the new silicon has trueaudio, XDMA, and frame pacing on board.
 
The Mac said:
Thats not entirely true.

the R9-290/290x, and the 260/260x are new silcon, the others are pitcairn/tahiti rebrands

the new silicon has trueaudio, XDMA, and frame pacing on board.
Click to expand...

And don't forget driver issues. I’ve got crackling/static in many games and I can’t use the video ports on the second card.
 
thats a crossfire XDMA issue. only effects 290/290x/260/260x on certain chipsets.

They are working on it supposedly
 
next year at the earliest would be my guess i remember reading somewhere that was it for new vid cards for the year...
 
Id be surprised if its that soon, took 2 years between tahiti and hawaii.
 
Technically GCN 1.1 is a "next generation" from GCN 1.0. GCN 2.0 will be a "next generation" from GCN 1.1.

Technically Maxwell is a "next generation" from Kepler.

Maxwell is an evolution of Kepler. Just like GCN 1.1 is an evolution from GCN 1.0

Hope that clears it up.
 
Brent_Justice said:
Technically GCN 1.1 is a "next generation" from GCN 1.0. GCN 2.0 will be a "next generation" from GCN 1.1.
Click to expand...

The changes between GCN 1.0 and 1.1 do not equate to a completely new generation. its more like mid-generation tweak.

Technically Maxwell is a "next generation" from Kepler. Maxwell is an evolution of Kepler. Just like GCN 1.1 is an evolution from GCN 1.0

Hope that clears it up.
Click to expand...

The equivalent of Maxwell -> Kepler transition is the next gen AMD architecture on TSMC 20nm which is likely to feature a enhanced GCN (maybe GCN 2.0 as you call it) and very likely HBM aka High bandwidth memory.

http://electroiq.com/blog/2013/12/amd-and-hynix-announce-joint-development-of-hbm-memory-stacks/

http://wccftech.com/amd-working-hynix-development-highbandwidth-3d-stacked-memory/

Just as AMD was one generation ahead of Nvidia with GDDR5 with the HD 4870, they will once again be a generation ahead of Nvidia with HBM. Volta in 2016 is the first Nvidia GPU with HBM.
 
LordEC911 said:
The "only" being included because other similar node shrinks offered 30-35% maybe a little more in power savings. Remember that these numbers come directly from the foundry so while I'm sure they aren't stretched too far they are likely of the "best case" variety.
http://www.tsmc.com/english/dedicatedFoundry/technology/20nm.htm

Correct, that 20-25% is what is saved from a die shrink from 28nm to 20nm.


Let's take Hawaii, 438mm2 @ 28nm ~300w TDP (typically gaming is less but using round numbers is easier)

Hypothetical numbers- ( no way accurate )
We will say they want to increase die size about 30% and they see about the same increase in TDP @ 28nm.
So on 28nm we are looking at 570mm2 and a TDP of 390w.
Let's shrink this down to 20nm.
Not everything shrinks linearly but let's say 40% savings in die size and go with the 25% savings in TDP.
570mm2 @ 28nm becomes 342mm2 @ 20nm and 390w @ 28nm comes back down to ~293w @ 20nm.

So they are going to be completely limited by TDP on a chip that is about the size of Tahiti.
That is why you have to fine tune your designs to the specific process, make architectural changes and tweaks to milk whatever efficiency you can because solely relying on new process nodes to advance your performance and power efficiency hasn't been working the last 3-4 generations.
Click to expand...

I understand what you are saying. But also it is my understanding that the current R9 290/290X die was suppose to be on 20nm but the 20nm process got delayed. The time was already scheduled for production so they had to make the necessary changes to the design to put it on a 28nm chip. If that is so, what we may get next will either be a new offering this fall (on a 20nm die) of what it was suppose to be in the first place, or a tweaked offering sometime later. I personally love to see them come out with an offering with more DP ports like the Eyefinity6 or the MATRIX cards.
 
hcforde said:
I understand what you are saying. But also it is my understanding that the current R9 290/290X die was suppose to be on 20nm but the 20nm process got delayed. The time was already scheduled for production so they had to make the necessary changes to the design to put it on a 28nm chip. If that is so, what we may get next will either be a new offering this fall (on a 20nm die) of what it was suppose to be in the first place, or a tweaked offering sometime later. I personally love to see them come out with an offering with more DP ports like the Eyefinity6 or the MATRIX cards.
Click to expand...

After Ibiza, that is extremely unlikely.
TSMC started building phase 5 in April 2012... Pretty sure TSMC customers got advance notice that 20nm wasn't likely to be ready for mass production in mid 2013.
 
LordEC911 said:
After Ibiza, that is extremely unlikely.
TSMC started building phase 5 in April 2012... Pretty sure TSMC customers got advance notice that 20nm wasn't likely to be ready for mass production in mid 2013.
Click to expand...

Yes, but that does not mean the design for 20nm was not well underway.
 
The Mac said:
Id be surprised if its that soon, took 2 years between tahiti and hawaii.
Click to expand...

However, they were working on the console chips at the time and all the effort was going into the designs for Sony, Microsoft and Nintendo boxes.
 
You must log in or register to reply here.
Back
Top