AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

I had to diverge and google search some of the terms you used (round robin , So really what AMD did here is add to their scheduling capabilities to increase their ALU utilization rate , and If I understood your post correctly , the compiler has less to work for now (now that clauses are gone) .
Yes, that's correct. AMD specifically stated the fact, that a good compiler for GCN is much easier to write than for the VLIW architectures, as a distinct advantage of the new architecture in their presentation.

Edit:
Hiroshige Goto did a good job drawing a diagram visualizing the round robin scheduling. The picture only shows the instructions for the vector ALUs, putting all existing issue ports in makes it probably a bit convoluted when you also introduce a time axis like he did. The pictures from the AMD's architecture presentation show all ports, but not how it works with the round robin issue, so basically one would have to put both together to get the complete picture.

23.jpg


The issue basically "moves" each clock cycle to the next vec16 ALU (SIMD engine), after 4 cycles it starts over at the beginning. Each small circle in one of the rectangular blocks represents one element of a wavefront, the block itself an instruction for the whole wavefront. The elongated ellipses stretching over 4 cycles represents an instruction which takes longer (double precision or transcendental for instance, in this case 4 times as long, but can also be something else as AMD stated half rate DP will be possible in some models).
In a hypothetical reality , how much more performance does a GCN core (with 1532 ALUs @880 MHz) achieve over a Cayman core running at the same frequency and with the same number of ALUs ?
That is very dependent on the code you run.
Starting from a few corner cases, where it may be slower than Cayman (the minimum would be half speed, but extremely unlikely), I would guess it's maybe 20% faster for code showing good (not perfect, that would be a draw) utilization (and not much fine grained control flow) on the old VLIW architecture with the extreme being a speedup by a factor of 10 or so (again extremely unlikely).

I would guesstimate the average speedup of the shader performance (you can only say something about this isolated issue, not about the game performance) over a wide range of code (which is not memory access limited) is a speedup of the pure shader performance by maybe 30 to 40% compared to Cayman.

AMD restructured the memory hierarchy too, now there is a real read/write L1 like Fermi has (but the local memory is separated, AMD unified the general purpose L1 with the Texture L1 which is still separate in nvidia's designs). But I have no idea how this works out for the performance side of things.
 
Last edited by a moderator:
http://vr-zone.com/articles/amd-radeon-hd-7000-gpus-listed-in-leaked-driver/13766.html

Leaked driver, 8.91 aka probably 11.11:
AMD6778.1 = "AMD Radeon HD 7000 series"
AMD68FA.1 = "AMD Radeon HD 7300 Series"
AMD6850.1 = "AMD Radeon HD 7400 Series"
AMD6858.1 = "AMD Radeon HD 7400 Series"

AMD6859.1 = "AMD Radeon HD 7400 Series"
AMD677B.1 = "AMD Radeon HD 7400 Series"
AMD6772.1 = "AMD Radeon HD 7400A Series"
AMD675D.1 = "AMD Radeon HD 7500 Series"
AMD6742.1 = "AMD Radeon HD 7500/7600 Series"
AMD6841.7 = "AMD Radeon HD 7500M/7600M Series"
AMD675B.1 = "AMD Radeon HD 7600 Series"
AMD6751.1 = "AMD Radeon HD 7600A Series"
AMD6840.7 = "AMD Radeon HD 7600M Series"

AMD683B.1 = "CAPE VERDE PRO"
AMD683F.1 = "CAPE VERDE PRO"
AMD6839.1 = "CAPE VERDE XT"
AMD683D.1 = "CAPE VERDE XT"
AMD6838.1 = "CAPE VERDE XTX"
AMD682F.1 = "CHELSEA PRO"
AMD6824.1 = "CHELSEA XT"
AMD682D.1 = "CHELSEA XT"
AMD6831.1 = "GREAT WALL"
AMD6821.1 = "HEATHROW PRO"
AMD6827.1 = "HEATHROW PRO"
AMD6820.1 = "HEATHROW XT"
AMD6825.1 = "HEATHROW XT"
AMD6819.1 = "PITCAIRN PRO"
AMD6818.1 = "PITCAIRN XT"
AMD6799.1 = "SI NEW ZEALAND"
AMD679A.1 = "SI TAHITI PRO"
AMD6798.1 = "SI TAHITI XT"
AMD6830.1 = "SUMMER PALACE XT"
AMD9901.1 = "TRINITY DEVASTATOR DESKTOP"
AMD9904.1 = "TRINITY DEVASTATOR LITE DESKTOP"
AMD9903.2 = "TRINITY DEVASTATOR LITE MOBILE"
AMD9900.2 = "TRINITY DEVASTATOR MOBILE"
AMD9991.1 = "TRINITY SCRAPPER DESKTOP"
AMD9990.2 = "TRINITY SCRAPPER MOBILE"
AMD6800.1 = "WIMBLEDON XT"
 
They forgot the interesting stuff - the driver relations. That's a lot of families!
Code:
AMD68FA.1 = ati2mtag_Evergreen = "AMD Radeon HD 7300 Series"

AMD6742.1 = ati2mtag_Vancouver = "AMD Radeon HD 7500/7600 Series"
(vancouver is hd6400m, hd6900m)

AMD675D.1 = ati2mtag_NI = "AMD Radeon HD 7500 Series"
AMD6772.1 = ati2mtag_NI = "AMD Radeon HD 7400A Series"
AMD6778.1 = ati2mtag_NI = "AMD Radeon HD 7000 series"
AMD677B.1 = ati2mtag_NI = "AMD Radeon HD 7400 Series"
AMD675B.1 = ati2mtag_NI = "AMD Radeon HD 7600 Series"
AMD6751.1 = ati2mtag_NI = "AMD Radeon HD 7600A Series"

AMD6850.1 = ati2mtag_R475 = "AMD Radeon HD 7400 Series"
AMD6858.1 = ati2mtag_R475 = "AMD Radeon HD 7400 Series"
AMD6859.1 = ati2mtag_R475 = "AMD Radeon HD 7400 Series"

AMD6840.7 = ati2mtag_R476 = "AMD Radeon HD 7600M Series"
AMD6841.7 = ati2mtag_R476 = "AMD Radeon HD 7500M/7600M Series"

AMD6799.1 = ati2mtag_R575 = "SI NEW ZEALAND"
AMD679A.1 = ati2mtag_R575 = "SI TAHITI PRO"
AMD6798.1 = ati2mtag_R575 = "SI TAHITI XT"
AMD683B.1 = ati2mtag_R575A = CAPE VERDE PRO
AMD6819.1 = ati2mtag_R575 = "PITCAIRN PRO"
(etc rest of cap verde, pitcairn)

AMD682F.1 = ati2mtag_R576 = "CHELSEA PRO"
AMD6824.1 = ati2mtag_R576A = "CHELSEA XT"
AMD682D.1 = ati2mtag_R576 = "CHELSEA XT"
AMD6821.1 = ati2mtag_R576A = "HEATHROW PRO"
AMD6827.1 = ati2mtag_R576 = "HEATHROW PRO"
AMD6820.1 = ati2mtag_R576A = "HEATHROW XT"
AMD6825.1 = ati2mtag_R576A = "HEATHROW XT"
AMD6800.1 = ati2mtag_R576 = "WIMBLEDON XT" 
AMD6831.1 = ati2mtag_R576 = "GREAT WALL"
AMD6830.1 = ati2mtag_R576A = "SUMMER PALACE XT"

AMD9901.1 = ati2mtag_RS711_Desktop = TRINITY DEVASTATOR DESKTOP
(etc rest of trinity)

Also found a couple of new workstation cards:
AMD6780.1 = ati2mtag_R575GL = "ATI FirePro V (FireGL V) Graphics Adapter"

7300 probably renamed 5400.
7400 and 7500M/7600M are probably GCN (except that some R475 parts are called "ATI Mobility Radeon HD 6000 series", but no specific)
7400-7600 desktop looks like VLIW5 (cayman is using ati2mtag_NICayman) - renames or shrinks. But do we have both GCN and VLIW5 7400 series?
All the remaining series should be GCN, except for Trinity about which we can't say.
 
Last edited by a moderator:
"TRINITY DEVASTATOR DESKTOP", I sort of like it.

Thanks Gipsel, I'm still trying to digest all you posted. Lots of good info if I can ever get my head around it, thanks again. :)
 
Oh wow that was impressive.

..
...
....
.....
......
.......
........
.........
..........
...........
............
.............
..............
...............
................
.................
..................
...................
....................................................

[boratvoice]NOOOOT[/boratvoice]
 
Oh wow that was impressive.

..
...
....
.....
......
.......
........
.........
..........
...........
............
.............
..............
...............
................
.................
..................
...................
....................................................

[boratvoice]NOOOOT[/boratvoice]

Far more impressive than having your CEO make a (bigger)clown of himself showing a fake board with woodscrews :LOL:
 
No desktop Radeon this year

According to what we know so far, partners got the boards but launch is still scheduled for January. When asked about the January launch, partners are still saying that Tahiti is the GPU that is going to come first while the rest of the lineup will launch after that, including the dual-GPU New Zealand, and mid-range and low-end Pitcairn and Cape Verde GPUs.

Of course, AMD has enough breathing room as currently Nvidia's desktop GPU should come in March, at least according to what we are hearing from partners, and we are talking about best case scenario and earliest date, so AMD has a good three months head-start on Nvidia.
 
Apparently from ChipHell (sadly don't have link) "But I have to point out: HD7770 SKU's is 2G, memory. This weekend can provide pictures."
 
Apparently from ChipHell (sadly don't have link) "But I have to point out: HD7770 SKU's is 2G, memory. This weekend can provide pictures."

That's not too surprising. I assume 2GB should be the new 1GB next gen.
 
Yes, that's correct. AMD specifically stated the fact, that a good compiler for GCN is much easier to write than for the VLIW architectures, as a distinct advantage of the new architecture in their presentation.
Thanks you very much .sorry or the late feedback (busy medical year) ..

I would guesstimate the average speedup of the shader performance (you can only say something about this isolated issue, not about the game performance) over a wide range of code (which is not memory access limited) is a speedup of the pure shader performance by maybe 30 to 40% compared to Cayman.
Then I guess the emphasis now should be on clocks , texture units/rops and front end efficiency , I just hope it wouldn't repeat the HD 6970 scenario where it didn't differentiate itself from HD 5870 in many old games . However I think that would be highly unlikely given the new fab process , which should allow for an HD 6990 like performance .
 
If it's a far-future design, it probably wouldn't have a 7**** designation.

An interposer could allow for a wider/faster bus, and stacked memory could alleviate capacity concerns for VRAM.
The proximity to the hot GPU could have adverse effects on the reliability and power consumption of the stacked memory, though.
 
Last edited by a moderator:
Back
Top