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

[CarstenS]

Indeed, some folks at AMD seem to really miss the old times. ATi-references are far more common than you'd think be necessary.

 

[DarthShader]

Just noticed something.

There is more "easter eggs". Check out the source of the page.

 

[Alexko]

There is more "easter eggs". Check out the source of the page.

header("X-LOL-Kepler: So Late");

All additional content on this page is the express property of Advanced Micro Devices. Also, everything depicted herein is completely fake.</div>

No kidding?

 

[no-X]

+"Rainbow Dash is best pony" - wtf?

 

[Lukfi]

Nah it's all GCN, that's pretty much confirmed.

Are you absolutely, positively sure about that? Where has this been "confirmed"?

 

[no-X]

During several AMD's events

 

[Kaotik]

Are you absolutely, positively sure about that? Where has this been "confirmed"?

I couldn't find the link, but I'm also 100% certain that it has been confirmed somewhere - might have been a slidedeck or interview, not sure.

 

[tunafish]

Are you absolutely, positively sure about that? Where has this been "confirmed"?

Most notably in the driver. The most recent drivers identify three "series" of cards that belong to the GCN family.

 

[CarstenS]

At least wrt to the mobile chips, AMD was quite open at CES this year that 7600 and below would utilize proven VLIW cores and 7700 and up would rely on GCN.

 

[Arty]

 

[snarfbot]

+"Rainbow Dash is best pony" - wtf?

http://mlp.wikia.com/wiki/Rainbow_Dash

also

http://games.adultswim.com/robot-unicorn-attack-twitchy-online-game.html

its a good game

 

[Lukfi]

@everyone: OK, I was just mildly confused by OBR posting the same fake specs twice. I mean, what kind of mentally slow person finds the same fake funny enough to post them for a second time?

 

[AlexV]

I'd appreciate it if the OBR stuff is kept far away from this forum, thank you. I don't know what internet blog drama is unfolding in the land of the Premysls, but I'm sure it's more than fine there and not here. Thank you.

 

[Man from Atlantis]

CPU-Z 0.5.9

Added support for AMD Radeon HD 7750 and HD 7770

Radeon HD 7770 Specifications Confirmed in GPU-Z Screenshot

640SPs, 16ROPs, 40TMUs, 128bit 1000/4500MHz

 

[mczak]

CPU-Z 0.5.9


Radeon HD 7770 Specifications Confirmed in GPU-Z Screenshot

640SPs, 16ROPs, 40TMUs, 128bit 1000/4500MHz

I don't think that really confirms anything (I don't doubt the values, however). Someone just entered the values as speculated into the database. Though it probably can read the clocks correctly, but I don't think it can read the CU count from these chips.

 

[CarstenS]

GPU-z actually reads CU/Shader counts from registers, not out of it's database. This way, there were faulty review samples of the Barts-GPU discovered that shipped with too many shaders enabled.

 

[LordEC911]

CPU-Z 0.5.9


Radeon HD 7770 Specifications Confirmed in GPU-Z Screenshot

640SPs, 16ROPs, 40TMUs, 128bit 1000/4500MHz

While 10CUs seems low, it seems much more likely than 20CUs.
I personally was hoping for either 12 or 16CUs.

 

[Man from Atlantis]

While 10CUs seems low, it seems much more likely than 20CUs.
I personally was hoping for either 12 or 16CUs.

I will be happier if this is 7750.. 1 GHz is kinda high for a PRO though..

 

[mczak]

GPU-z actually reads CU/Shader counts from registers, not out of it's database. This way, there were faulty review samples of the Barts-GPU discovered that shipped with too many shaders enabled.

I don't think that's always the case though, only if it knows how to read it. In particular it might not be true for new gpus which might return their basic configuration data differently than previous chips. There's also always the possibility of misinterpretation if the meaning of such data changes (for instance the chip might not return CU count but CU groups and hence if the group size suddenly changes and you don't take that into account you'll get wrong results obviously). And I'm quite sure you can't read TMU count for instance, you just have to know how many TMUs there are per CU.

 

[CarstenS]

Yeah, some things it derivates of others. Attached TMUs for a given SIMD/GCN/SM as you said are an example AFAIK. But wrt to SIMD/CU-counts, it calculates those from register readouts - as soon the cards are supported.