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

I agree, I think the 7970 launch was perfect. Reviews plus availability 2 weeks after. As long as the promised date isn't too far in the future and they actually hit the date then it's better than launch day availability IMO. Give us some time to digest and discuss reviews before dropping cash.
 
Such is the annoyingness of people complaining about "paper launches". Now because of them we dont get the 7950 info for nearly another month, for no good reason. Just so that people wont complain about a "paper launch".

Card would have available early February either way, only difference is now we wont know more about it for longer.

Longtime paper launch supporter here.
I hardly see the term "paper launch" being toss around other industries. Seems like it's only derogatory for the videocard fan wars. I mean, who gives a rats-ass as long as it's a reasonable time period between launch and consumer availability.
 
Only possible reasons I could think of to delay 7950 is either the yields are unexpectedly good, thus no confidence in stock availability of 7950 without purposely castrating working 7970 models. Or expected demand is far higher than supply of 7950's on the 9th.

Or...the 7950 is BIOS unlockable like the 6950 and they want to get as many sales of 7970 before releasing a part that may cannibalize some 7970 sales.

Regards,
SB
 
The negativity surrounding paper launches is the result of those times in the past when IHVs "launched" parts with expectations of immediate availability and failed to deliver. No fault but their own.
 
dd6h.jpg


24bw.jpg

Gigabyte GV-R797OC-3GD
1GHz/6GHz
 
FYI - redundancy means that there are extra elements built in to attain a full configuration in more cases, but those redundant elements can never be used in addition. I think your question would be whether this is harvested.

And? Is the HD7970 a harvested part from the full Tahiti ASIC?
 
And? Is the HD7970 a harvested part from the full Tahiti ASIC?

Loaded question for an employee! Maybe he shouldn't answer it. Could have answered in that post you quoted instead of just correcting me on my misuse of "redundancy" haha!

I'll just say what I want the answer to be: yes! :LOL:
So original NI was supposed to be 2560sp 4D array 640 ALU @ 32nm or 28nm. 2:1 increase in ALU count over Evergreen. Now for the sake of die size while having to remain @ 40nm, 480 ALU ...
For good or bad, I think it's 2560 shaders, 40 CUs. IMO, Transistor count is the tell-all. 4.313 Billion trans seems like a lot for 2048. [fantasy] Then there's this paper suggesting a 2304 part would be 36 CU's. Think about the marketing group's strategy versus NV's GK, what if AMD needs a "4890" or "580" out of the Tahiti die to compete with Kepler? What if yields were shoddy on the new node at TSMC? Then harvested parts down from 40 to 32 compute units to supply working quantities to meet historical market demand for number of flagship cards purchased at launch? Maybe the foundry could yield a supply of 2000 40 CU dies or contrarily 200,000 32 CU dies? New process lithography, redundancy, TSMC's 28nm. [/fantasy] Nvidia's gtx480 and 460 (SIMD designed) new 40nm launches did not boast usage of the full dies either. Have we even seen a Cypress die or Cayman die confirming no redundancy? If this really is the case, then i dont think we should hate on AMD, I'm still getting a 7970. :p
Consider these numbers (with Cayman a stepping-stone):

rv770....: 1 mil tran, 40 TMU, 10 arrays, 160 alu blocks, 55nm - 800 VLIW
cypress: 2.15 mil tran, 80 TMU, 20 arrays, 320 blocks, 40nm - 1600 VLIW
tahiti......: 4.30 mil tran 160 TMU, 40 arrays, 640 blocks, 28nm - 2560 GCN

700x299imagesstoriesnew.jpg


However, this is just my guess. Maybe GCN arch changes a lot about how you can guestimate die size with 1 Shader SIMD versus VLIW-4/5 ALU group. I doubt it. :D Pardon the brevity.
 
For good or bad, I think it's 2560 shaders, 40 CUs.
It's 32. 40 would be just too much. 36 doesn't make a lot of sense (from a floor planning point of view.) Transistor counts are not that useful. Smaller technologies have higher resistance wires and require more buffers to drive signal. The additional compute flexibility must have some overhead. We'll know soon enough. Anybody has a spare sample and some sulfur?
 
For a while, a particularity of the Radeons that really bothered me was their minimal framerates in games (something that is far too rarely considered in reviews). It seems that Scott's review has finally put those concerns to rest.

Not sure if the VLIW architecture itself or the drivers were to blame but, even with the less mature GCN drivers, AMD has made huge strides there.
 
Not sure what review you just read, but the 7970 is on par with or WORSE than a GTX580 in Skyrim, Batman, Crysis2. Its perf isnt even that much better in BF3. It's all about the percieved micostutter (which happens on single GPUs too).

That review highlights what I've been saying since the 7970 was announced. The vanilla card just isn't a decent enough step forward over a 580. It's obvious from the cooler set ups that AMD have been seriously cautious with the speeds of these chips. I see great potential and will happily buy a 1400-1500mhz factory OC'd version, but at 925mhz it in no way interests me.
 
Not sure what review you just read, but the 7970 is on par with or WORSE than a GTX580 in Skyrim, Batman, Crysis2.

Are you going to cherry pick 3 more reviews at a low resolution that supports your delusions again?

You might want to try sites like hardocp (although you'll have to settle for a heavily overclocked 580 getting beat here), hardwarecanucks or techpower for evidence that the 7970 is ahead.
 
Interesting tidbit from Scott Wasson's review of the HD 7970

However, AMD's David Nalasco tells us Tahiti handles special functions like transcendentals in the vec16 units, at a very nice rate of four ops per clock cycle.

Does it mean that the shader units work together within the Vec16 units on Transcendentals, or is he referring to throughput?
 
Yes, my dellusions, based on facts and graphs by a 3rd party at resolutions the majority of people will be using. How dare I be so bold!

I don't care how high a single avg fps number is, its the slight stuttering thats most annoying:
crysis2-beyond.gif

Worse than a 580GTX.

skyrim-99th.gif

Worse than a 580GTX.

batman-99th.gif

On a par with 580GTX.

Yes, it has higher framerates (20%-30%), but as you can see from the Techreport review, hardly any time does this take a game from unplayable to playable.

My other issues with the card are it being both too conservative, and pushed slightly too hard. In this I'm talking about the fan. On full load, the fan is more than mildly annoying, but it doesn't need to be. The card is actually quite cool really. The fan should be 20% slower for 10% more heat. It's like they set the fan up for having the cards run at 1100mhz, then forgot to reset them when they dumbed the card down.

There is a VERY good card waiting to get out of the 7970. The default one is a bit schizophrenic. I just want the best card for my money. That card is not the default one.
 
It's cute how according to you when the nVidia part wins by a tiny margin it's "better", but when the AMD chip comes out ahead by a tiny margin it's "on par".
 
Back
Top