TR Forums

from pcper. non story? does this mean anything?

The AMD mainstream GPU lineup still includes GCN 1.0 cards like Pitcairn and even the GCN 1.1 Bonaire cards are the better part of 3 years old now. To say they will release 2 new GPUs means that the disappointing Tonga and perhaps even the power hungry Hawaii will likely still be part of their product stack in a year's time.

Tonga does not disappoint. It provides better gaming performance than similarly-priced GPUs from the competition.

makinbaconpancakes wrote:

from pcper. non story? does this mean anything?

All we can do is guess at the moment-

Fury is "NEW" lets shrink it to 16nm call it top end.
New GPU lets guess and call 490/490X
Tonga isn't that old lets shrink to 16nm call 480/480X
New GPU lets guess and call 470/470X

Anything below that is not really significant and can be rebrands of any type-

silly question, but are the future chips going to be merely die shrinks, or are they coupling that with a new architecture? im assuming both.

edit: i think that's where "hopefully" comes into play in terms of the die shrink (from article).

makinbaconpancakes wrote:

silly question, but are the future chips going to be merely die shrinks, or are they coupling that with a new architecture? im assuming both.

edit: i think that's where "hopefully" comes into play in terms of the die shrink (from article).

AFAIK, from what has been made public, the FinFET chips will be mainly based in the GCN architecture with minor modifications. The most dramatic change will be the ISA that will be completely new. So, more GCN (the same) but used in a new and maybe more efficient way. That means a big driver rewrite.

From the software point of view they will be quite different beasts.

I was under the impression that Greenland was a brand new GCN-based architecture, not just a shrink. Not sure about Baffin and Ellesmere though.

GCN next (Graphics Core Next next ) is changing a lot but keeping ISA/etc all the same.

At least that's what the most reliable fact finder I know has found out (no, it's not Charlie ). Of course this person also claims they have a part that is about 32CU coming along with 2 larger parts. Which contradicts the only 2 AMD GPUs coming here.

Given AMD's GPU department has super-secrecy I'm not going to fully discount said person claiming 3 GPUs.

A 24CU part replacing Pitcairne should be pretty close to 100mm^2 and probably pretty good yields. Wonder if it is possible to "launch" it in 2015 with no real volume until 2Q2016. lol.

NoOne ButMe wrote:

GCN next (Graphics Core Next next ) is changing a lot but keeping ISA/etc all the same.

Really?

http://amd-dev.wpengine.netdna-cdn.com/ ... ecture.pdf

Page xv wrote:

 Removed Instructions
– V_MAC_LEGACY_F32
– V_CMPS* - now supported by V_CMP with the “clamp” bit set to 1.
– V_MULLIT_F32.
– V_{MIN, MAX, RCP, RSQ}_F32.
– V_{LOG, RCP, RSQ}_CLAMP_F32.
– V_{RCP, RSQ}_CLAMP_F64.
– V_MUL_LO_I32 (it’s functionally identical to V_MUL_LO_U32).
– All non-reverse shift instructions.
– LDS and Memory atomics: MIN, MAX and CMPSWAP on F32 and F64 data

NoOne ButMe wrote:

At least that's what the most reliable fact finder I know has found out (no, it's not Charlie ).

My source at AMD says different! I cited "him" above, so who is yours?

NoOne ButMe wrote:

Of course this person also claims they have a part that is about 32CU coming along with 2 larger parts. Which contradicts the only 2 AMD GPUs coming here.

OHHH! So it's a secret source. Riiiiiiighttt!

NoOne ButMe wrote:

Given AMD's GPU department has super-secrecy I'm not going to fully discount said person claiming 3 GPUs.

Me neither. I have an additional source, also supra-mecha-mega-chrono-trigger-crypto-bitcoin-secret, who claims 16.3333 (repeating of course) GPUs.

NoOne ButMe wrote:

A 24CU part replacing Pitcairne should be pretty close to 100mm^2 and probably pretty good yields.

Over unity, I'm sure.

NoOne ButMe wrote:

Wonder if it is possible to "launch" it in 2015 with no real volume until 2Q2016

Why?

NoOne ButMe wrote:

lol.

That might be the only thing I agree with you about.

Hmmm I still have my fingers crossed for a true 7850/7870 replacement, with 2GB or 4GB options, latest GCN revision, 1 six/eight pin connector, 3 display ports @ v.1.3, 1 hdmi port @ ver.2.0, 14nm/16nm finfet.

anotherengineer wrote:

Hmmm I still have my fingers crossed for a true 7850/7870 replacement, with 2GB or 4GB options, latest GCN revision, 1 six/eight pin connector, 3 display ports @ v.1.3, 1 hdmi port @ ver.2.0, 14nm/16nm finfet.

Apart from the output versions and the fabrication node, that's the R9 Nano

JustAnEngineer wrote:

Tonga does not disappoint. It provides better gaming performance than similarly-priced GPUs from the competition.

At the time Tonga was released as the R9 285 it provided roughly similar performance to the old Tahiti based R9 280 and could not outperform the R9 280X. It had similar power consumption to the R9 280/7950 Boost and came equipped with 2GB of memory, 33% less than the card that preceded it. Sure you got similar performance from lesser hardware but that hardware cost just as much and used just as much power as the card it was replacing. Aside from Freesync support, which was at the time vaporware and is still only trickling into market, Tonga offered no tangible benefit to the consumer at the time of release which is a disappointment for a brand new GPU.
Some reading material on the topic; (pricing) http://www.anandtech.com/show/8460/amd- ... 285-review (power consumption) http://www.guru3d.com/articles-pages/am ... iew,5.html

Let's fast forward to today and you will see the 2GB and 4GB versions of the R9 380 and GTX 960 trading blows within single digit percentage points apart from one another in terms of both frame rates in games and prices on the street. Meanwhile you can save yourself 50+ watts of power going green with no tradeoff.
Again an example: http://www.hardocp.com/article/2015/08/ ... _review/12

To sum up R9 285 provided little incentive to purchase at launch and the R9 380 continues that legacy today. I know I am pretty disappointed, not sure why you aren't...

Edit: Added some examples to back up my opinions.

Slightly off-topic, I do agree with MDBT that Tonga is disappointing.

After three years of Tahiti, Tonga came along. It was about the same die size, about the same power draw, about the same performance and about the same cost as existing Tahiti products. I liken the Tahiti>Tonga jump to the Sandy>Ivy jump. Everything was fractionally better but by such a small margin that nobody really cared. The disappointing part was that after three years of waiting, a factory-overclocked 7950 from was basically matching it's new replacement in all but some new features which still, to this day, have barely been used by any games.

A squad mate in War Thunder ordered a r9 380 and I told him you ordered a cut down Tonga based card and I steered him torwards the GTX 960 4gb card from evga. It was like 10-15$ more but I also informed him he can upgrade to a 970 without loosing any money from buying and trying the 960 4gb Since EVGA has the 90 day step up program. He needed a way to get another 110-120$ past his wife for the 970 and the stepup program EVGA uses is the perfect tool for that. I already told him going from his HD7750 to a gtx 960 will be like night and day but stepping up to the gtx 970 around 60 days later will be like stepping up a few nights and days and should last him a good 4 years. He has a corsair 750 watt psu in a HP tower with a i5 2400 so he will be fine up to a 2560-1440 panel with the 970 where the 960 will not do nearly as well as the 970 if he finally goes 1440p over 1080p.

I explained 1440p to him is like going from SD to HD, it does not seem like a lot but it is a much better resolution allowing you to use less AA in game since the pixels are that much smaller. Also I told him a 4k panel under 37" is overkill IMHO.
So once he gets his 960 from evga I will make sure he gets a 970 before his 90 days are up.