Monday, December 12th 2011

AMD Tahiti GPU Specifications Compiled

If the word on the optical fibers is true, we are less than a month away from the launch of AMD's next high-end graphics card family based on its next high-performance GPU, codenamed "Tahiti". According to 3DCenter, AMD will launch new graphics card models based on this GPU around January 10, 2012. It is expected that we'll learn a lot more about these GPUs, maybe even come across AIB-branded graphics cards, at the upcoming CES event.

3DCenter compiled specifications of "Tahiti", based on bits and pieces of information from various sources. The specs can be listed out as:
  • 4.50 billion transistors, die-area of 380 mm², built on TSMC 28 nm process
  • Advanced GCN 1D architecture
  • 2048 1D processing cores
  • 128 TMUs, 48 ROPs
  • 384-bit wide GDDR5 memory interface, memory clock slightly below 1 GHz, target bandwidth of 240~264 GB/s
In Gandhi's words, salt is as free as the air.Source: 3DCenter.org
Add your own comment

41 Comments on AMD Tahiti GPU Specifications Compiled

#1
Nesters
by: nt300
I read somewhere we should expect 40% better performance with these new cards.
Not at the same price, however.
Posted on Reply
#2
pantherx12
by: DrunkenMafia
I highly doubt it but will have to wait and see.. :wtf:
It's not that far fetched man look at 4870 vs 5870.

Die shrinks always have the biggest performance difference.

Well cept with Bulldozer, no one knows what happened there :laugh:
Posted on Reply
#3
HalfAHertz
Bigger die than a 6970 - looks like a power guzzler...
Posted on Reply
#4
devguy
by: HalfAHertz
Bigger die than a 6970 - looks like a power guzzler...
Aren't they both 380mm^2 in die size?
Posted on Reply
#5
Aetherius
new wave of GPU...tasty as always!! :laugh:
Posted on Reply
#6
bear jesus
I have relatively high hopes for the 7970 but really look forward to see what nvidia has and how both compare, hopefully nvidia allows 3 monitors on one card so i have 2 options for triple monitor with a single gpu.

by: HalfAHertz
Bigger die than a 6970 - looks like a power guzzler...
by: devguy
Aren't they both 380mm^2 in die size?
GPU-Z says 389mm^2 for the 6970.
Posted on Reply
#7
WarraWarra
Not sure but read somewhere the power would be fairly decent on these, hope it is so.
Specs looks promising.
Has anyone figured out what the trinity gpu's would end up as vs Tahiti ??
The 40% and 60% yeah right not likely more like 10% ~ 15% at most, AMD does not have 40%~60% in their bones or in their dreams, I wish they had.

Most users with cash then would aim price wise for a complete pc rebuild = lots of cash and so very stingy on spending anything up to tax year end.
As the market is now and major doubts about AMD cpu's and GPU's that might fail FX-8150 again, AMD would have a hard sell to get users to even look at AMD anything.

The question is does AMD want to go bankrupt in 2012 or would they be sane enough with their prices and performance.
Posted on Reply
#8
pantherx12
by: WarraWarra
Not sure but read somewhere the power would be fairly decent on these, hope it is so.
Specs looks promising.
Has anyone figured out what the trinity gpu's would end up as vs Tahiti ??
The 40% and 60% yeah right not likely more like 10% ~ 15% at most, AMD does not have 40%~60% in their bones or in their dreams, I wish they had.

Most users with cash then would aim price wise for a complete pc rebuild = lots of cash and so very stingy on spending anything up to tax year end.
As the market is now and major doubts about AMD cpu's and GPU's that might fail FX-8150 again, AMD would have a hard sell to get users to even look at AMD anything.

The question is does AMD want to go bankrupt in 2012 or would they be sane enough with their prices and performance.
Are you trolling? :laugh:

Does everyone forget this?




40%-60% would be a small improvement for a die shrink on a gpu.


Clear-sky just for one that's over 100% XD


This isn't like 5870-6970 where it's the same die size but tweaked, if they built a 6970 on the same process as the 7970 cards that are coming it would be around half the size of the current one!
Posted on Reply
#9
Benetanegia
by: pantherx12
Are you trolling? :laugh:

Does everyone forget this?
http://tpucdn.com/reviews/ATI/Radeon_HD_5870/images/crysis_1920_1200.gif

http://tpucdn.com/reviews/ATI/Radeon_HD_5870/images/fear_1920_1200.gif

40%-60% would be a small improvement for a die shrink on a gpu.

http://tpucdn.com/reviews/ATI/Radeon_HD_5870/images/clearsky_1920_1200.gif
Clear-sky just for one that's over 100% XD


This isn't like 5870-6970 where it's the same die size but tweaked, if they built a 6970 on the same process as the 7970 cards that are coming it would be around half the size of the current one!
Not the exact same situation though. The HD5870 had exactly twice as many SIMD units and ROPs as the HD4870/4890. The rumored specs for Tahiti presented in the OP are only a 33% increase in SIMD units. 5870 to 6970 was a 20% increase.

There's the scalar-like SIMD versus VLIW thing, but until we see hard numbers, there's not much point believing that will suppose a massive improvement. Remember that AMD/Ati has been saying to everyone+dog, for more than 5 years that VLIW was much more efficient for graphics. Of course now they are/will be saying that non-VLIW SIMD arrays are 10x better. In both cases it's PR and marketing, and VLIW was not as good as they said for 5+ years, nor is as "bad" as they tell you now and GCN is not as good (in comparison) as they want make you believe.

The point is, if we are to believe the specs from OP, you'd be much better off saving yourself from dissapointment by staying skeptical until reviews come in. I'm not sayng I agree with Warrawarra, but just don't expect 2048 SPs to be more than 30% faster than 1536 SPs just yet, until there's hard numbers at least.
Posted on Reply
#10
AsRock
TPU addict
by: Benetanegia
Not the exact same situation though. The HD5870 had exactly twice as many SIMD units and ROPs as the HD4870/4890. The rumored specs for Tahiti presented in the OP are only a 33% increase in SIMD units. 5870 to 6970 was a 20% increase.

There's the scalar-like SIMD versus VLIW thing, but until we see hard numbers, there's not much point believing that will suppose a massive improvement. Remember that AMD/Ati has been saying to everyone+dog, for more than 5 years that VLIW was much more efficient for graphics. Of course now they are/will be saying that non-VLIW SIMD arrays are 10x better. In both cases it's PR and marketing, and VLIW was not as good as they said for 5+ years, nor is as "bad" as they tell you now and GCN is not as good (in comparison) as they want make you believe.

The point is, if we are to believe the specs from OP, you'd be much better off saving yourself from dissapointment by staying skeptical until reviews come in. I'm not sayng I agree with Warrawarra, but just don't expect 2048 SPs to be more than 30% faster than 1536 SPs just yet, until there's hard numbers at least.
I am skeptical but hopefully as it does seem that made some daring changes. Maybe they will add VLIW back to the chip later on. Lets face it there trying to make a 2 in 1 here.

Very interested to see how thing turn out, would be nice to get some news on the other side too.
Posted on Reply
#11
HalfAHertz
by: bear jesus
I have relatively high hopes for the 7970 but really look forward to see what nvidia has and how both compare, hopefully nvidia allows 3 monitors on one card so i have 2 options for triple monitor with a single gpu.





GPU-Z says 389mm^2 for the 6970.
you are both right - my bad :(
Posted on Reply
#12
pantherx12
by: Benetanegia
Not the exact same situation though. The HD5870 had exactly twice as many SIMD units and ROPs as the HD4870/4890. The rumored specs for Tahiti presented in the OP are only a 33% increase in SIMD units. 5870 to 6970 was a 20% increase.

There's the scalar-like SIMD versus VLIW thing, but until we see hard numbers, there's not much point believing that will suppose a massive improvement. Remember that AMD/Ati has been saying to everyone+dog, for more than 5 years that VLIW was much more efficient for graphics. Of course now they are/will be saying that non-VLIW SIMD arrays are 10x better. In both cases it's PR and marketing, and VLIW was not as good as they said for 5+ years, nor is as "bad" as they tell you now and GCN is not as good (in comparison) as they want make you believe.

The point is, if we are to believe the specs from OP, you'd be much better off saving yourself from dissapointment by staying skeptical until reviews come in. I'm not sayng I agree with Warrawarra, but just don't expect 2048 SPs to be more than 30% faster than 1536 SPs just yet, until there's hard numbers at least.
Don't worry I'm not holding my breath about it, I'm just saying it's not impossible.

:toast:
Posted on Reply
#13
Fishymachine
Yeah but Tahiti has almost twice the transistors, and we're talking about a different architecture, so maybe those 2048CU behave closer to 3000 VLW4
Posted on Reply
#14
RejZoR
They could as well behave as 1500 compute shaders...
Posted on Reply
#15
Benetanegia
by: Fishymachine
Yeah but Tahiti has almost twice the transistors, and we're talking about a different architecture, so maybe those 2048CU behave closer to 3000 VLW4
Double the transistors means very little in reality in this case and 4.4 is not really 2x 2.7 is it? To the point, AMD has been postponing many GPGPU features that all require many more transistors and they are introducing them all at once now.

-Fat 1D shaders require much more transistors than VLIW.

-It will also feature a true scalar core per CU, which requires lots of transistors and will have little to no effect on gaming performance.

-Tahiti will support half rate double precision, that requires a massive ammout of transistors, compared to 1/4 rate. Hell AMD didn't even include any kind of 64 bit support on smaller chips, i.e. Barts to save transistors.

- Larger and more "CPU-like" caches and interconnects require much more transistors.

And the list goes on and on and on. You could take Cayman, convert it to this GPGPU oriented architecture and end up with 3.5 million transistors and 0 gain on gaming performance.

Tahiti will be much much faster on compute, but gaming performance is not so clear. Gaming performance is no longer the only focus of GPU architecture designs, whether we like it or not.
Posted on Reply
#16
erocker
It's a little picture but you get the idea... This would be the HD 7970




Not to scale 6970 for comparison
Posted on Reply
Add your own comment