Dual-GPU AMD Radeon HD 8990 specifications leak online

Shawn Knight

Shawn Knight

Posts: 15,292   +192
Staff member

Spanish website BitDreams.se (Google translated) has posted what they claim is information on AMD’s next generation Radeon HD 8000 Series GPUs. Of particular interest to hardcore gamers is the indication that AMD will be launching a dual-GPU card sometime in the second quarter of 2013.

The purported leaked roadmap reveals that two 8800 Series cards are scheduled to ship this quarter. If that is to be believed, AMD is quickly running out of time to make that happen. But perhaps more interesting is specifications of the Radeon HD 8990 (Venus XTX), which will essentially be two 8970 GPUs combined on a single PCB.

This monster card is listed as having 5.1 billion transistors (between the two GPUs), 5,210 stream processors and 160 texture units per core. The 8990 is also listed as having 48 ROPs per core and a pair of 384-bit buses. Memory is expected to be anywhere in the range of 6GB to 12GB and should be clocked from the factory at 1,250MHz. Each GPU will ship running at 950MHz, according to the roadmap.

It’s worth mentioning that the card will carry a maximum TDP of 375 watts and that each core and memory set will be clocked slightly lower than a standalone 8970 GPU. For comparison, a single 8970 will run at 1050MHz core / 1500MHz memory. This means the card won’t be quite as powerful as two individual 8970s but should take up less space on your motherboard and in your case.

The card will support DirectX 11, Shader Model 5.0 and OpenGL 4.2 just like the rest of the 8000 Series lineup. Expect to pay upwards of $1,000 (estimate) to outfit your gaming rig with this dual-GPU beast when (and if) it ships next year.

Permalink to story.

https://www.techspot.com/news/51013-dual-gpu-amd-radeon-hd-8990-specifications-leak-online.html

 
All those specs are fake/made up. Whoever put them together must have been bored.

1) If HD8970 uses 260W of TDP and AMD couldn't put 2x HD7970Ghz cards into 1 GPU (heck not even 925mhz HD7990s), then no way will they be able to do a 375W HD8990.

2) HD8970
- Memory bandwidth is wrong: 6000mhz @ 384-bit = 288GB/sec. They have it at 322GB/sec.
- Double precision is wrong: DP is 1/4 of SP in GCN. If you have SP of 5.38Tflops, DP has to be 1.34. They have it as 1.6 Tflops. Impossible. If they somehow go to 1/3rd of SP, the number is 1.79. No matter how you slice it, theirs is made up.

3) HD8950
- Memory bandwidth is wrong: 5500mhz @ 384-bit = 264 GB/sec. They have 300GB/sec.
- TMUs are wrong: GCN has a 1 compute unit to 4 TMU ratio. To get 2304 SPs, you need 36 Compute units or 144 TMUs. They have it as 140 TMUs.

4) Odd power consumption. There is a 50W difference between HD8970 and 8950 despite very small differences in specs. Yet, the same 50W difference exists between HD8870 and HD8950 despite the latter having 50% more ROPs and huge advantages in memory bandwidth and shaders. This logically doesn't make any sense. Also, you'd end up with a huge gap in performance between 48 ROP and 32 ROP parts.

5) How can they only increase transistors from 4.3B to 5.1B and yet squeeze in 50% more ROPs, 25% more shaders and 25% more TMUs and TDP only goes up 10W? I'll believe it if they raised shaders and TMUs but ROPs stay at 32 or at most 40. If anything, I can see more transistors being used to improve Rasterization and Geometry engines in the architecture. 48 ROPs is probably not possible until 20nm.

Overall, these specs are sloppy, have mathematical mistakes, don't adhere to GCN ratios of how things work. It's pure speculation as so many mistakes rule them out as credible.
 
  • Like
Reactions: fimbles, St1ckM4n and misor
Blue Falcon said:
All those specs are fake/made up. Whoever put them together must have been bored.

1) If HD8970 uses 260W of TDP and AMD couldn't put 2x HD7970Ghz cards into 1 GPU (heck not even 925mhz HD7990s), then no way will they be able to do a 375W HD8990.

2) HD8970
- Memory bandwidth is wrong: 6000mhz @ 384-bit = 288GB/sec. They have it at 322GB/sec.
- Double precision is wrong: DP is 1/4 of SP in GCN. If you have SP of 5.38Tflops, DP has to be 1.34. They have it as 1.6 Tflops. Impossible. If they somehow go to 1/3rd of SP, the number is 1.79. No matter how you slice it, theirs is made up.

3) HD8950
- Memory bandwidth is wrong: 5500mhz @ 384-bit = 264 GB/sec. They have 300GB/sec.
- TMUs are wrong: GCN has a 1 compute unit to 4 TMU ratio. To get 2304 SPs, you need 36 Compute units or 144 TMUs. They have it as 140 TMUs.

4) Odd power consumption. There is a 50W difference between HD8970 and 8950 despite very small differences in specs. Yet, the same 50W difference exists between HD8870 and HD8950 despite the latter having 50% more ROPs and huge advantages in memory bandwidth and shaders. This logically doesn't make any sense. Also, you'd end up with a huge gap in performance between 48 ROP and 32 ROP parts.

5) How can they only increase transistors from 4.3B to 5.1B and yet squeeze in 50% more ROPs, 25% more shaders and 25% more TMUs and TDP only goes up 10W? I'll believe it if they raised shaders and TMUs but ROPs stay at 32 or at most 40. If anything, I can see more transistors being used to improve Rasterization and Geometry engines in the architecture. 48 ROPs is probably not possible until 20nm.

Overall, these specs are sloppy, have mathematical mistakes, don't adhere to GCN ratios of how things work. It's pure speculation as so many mistakes rule them out as credible.
Click to expand...

I still want one.
 
Blue Falcon said:
All those specs are fake/made up. Whoever put them together must have been bored.
Click to expand...
Yup. They are basically extrapolations made on some crystal ball gazing by the guys at 3DCenter over five months ago. Guesstimates that every man and his dog has since turned into "fact".
IMO, there wont be any legit information until pre-production samples start doing the rounds for testing and validation, and we're probably still a couple of months away from that.
 
Remember that little switch that had a yellow label saying not to use it on the 6990? That little switch allowed the card to suck back over 100 watts of extra power bringing their 375 watt tdp to almost 500 watts. With that said I don't see it being impossible for AMD to pull it off again with this card, GCN 2.0 should already feature revisions to reduce power consumption and improvements almost everywhere else. Still, false information aside, a card of this class is almost unnecessary, it fills such a small niche in the gaming world. Everyone ogles over it, few buy it, and out of those who do buy it, fewer put it to proper use.
 
I can't help but think the delay is more bad than good. If AMD had product ready to go, we would have seen Q1 2013 launch. They either ran into technical issues, the company is strapped for resources to do a proper launch due to re-organization at the moment/focus on other areas, or they are waiting for NV to launch first so that they revert to price/performance strategy of the HD4000-6000 series. I can't think of many valid reasons to delay a product into Q2 2013, which misses the absolutely phenomenal Q1 when it comes to launch of new PC games (look this up!).

The other possibility is that they want to wait another quarter to do a later stepping re-spin, allowing more time for 28nm node to mature to either lower voltages or achieve higher GPU clocks at the same voltages. As HD7970 paper launched around Dec 22, 2011, I really don't have a good explanation why a 28nm refresh that's rumored to be only 15-30% faster is going to take 1.5 years to get out. Q2 2013 is still far out and at that point I am not even sure if it's worth getting next generation GPUs or just wait a bit longer for 20nm Maxwell/HD9000 in 2014. Fingers crossed that those aren't delayed to 2015...
 
Or they have full bins and an inventory of stock with card sales slowing, and want to unload some/most of it without dropping prices for the new line. A plan that is afforded them with no threat of team green preempting them.
Just an idea.
 
red1776 said:
Or they have full bins and an inventory of stock with card sales slowing, and want to unload some/most of it without dropping prices for the new line. A plan that is afforded them with no threat of team green preempting them.
Just an idea.
Click to expand...
Bingo.
AMD is (and always has been) about return on investment. Yields on Tahiti/Barts/Cape Verde are obviously good- the salvage-of-a-salvage Tahiti LE is appearing late and in small quantities for instance, so why not continue to stock the channel with them...the lithography masks and production line are already in place, so it makes sense to maximize your profit line when the opportunity arises.
Personally I don't think Sea Islands are fully baked yet anyhow. AMD releases aren't strictly yearly...
HD 4800 (June 2008) -> HD 5800 (Sept. 2009) = 16 months
HD 5800 (Sept 2009) -> HD 6900 (Dec. 2010) = 16 months
HD 6900 (Dec. 2010) -> HD 7900 (Jan. 2012) = 14 months
14-16 months on takes us to Feb-Apl 2013...and April is Q2.

...and given the scuttlebutt about a larger GPU die... fewer viable die candidates per wafer, and possibly lower initial yields in comparison to the current generation you'd think that AMD (and likely Nvidia) would attempt to launch at their leisure rather than forced to accelerate their timetable. I'm thinking that AMD and Nvidia aren''t about to cut each others throats for some short lifespan PR any more than they indulge in pricing wars.
 
The 7990 was released by third parties...http://www.newegg.com/Product/Product.aspx?Item=N82E16814131483
 
"Does it come with a nuclear reactor built-in?"

..yes, but the nuclear fuel sold separately, please buy it from another vendor :D
 
@BlueFalcon... I can see that you have no clue what you are talking about.
You can not compare two separate cards and expect the single dual GPU to use the same amount of resources as two 8970s...

1) Having both GPUs located on same board (Underclocked) decreases the amount of power needed to power them up, so 375W is not that unbelievable.

2) Same about memory bandwidth. The pipeline design can be tweaked and significantly altered to accommodate additional bandwidth for the dual GPUs.

3, 4 & 5... dude... seriously read up on design... you are pulling conclusions out of your arse.
 
I know this is a very old post but the HD 7990 and HD 7970x2 both exist still,the HD 7990 is made by POWERCOLOR and the HD 7970x2 is made by HIS! Both are on Newegg.com!
 
You must log in or register to reply here.

Similar threads

Jimmy2x
Asus promo offers UK customers cashback on Radeon 7000 GPUs
Replies
20
Views
410
fadingfool
fadingfool
Daniel Sims
AMD patent potentially describes future GPU chiplet design
Replies
21
Views
450
HardReset
H
Z
Coder crafts smooth 3D graphics engine from scratch, runs it on DIY GPU
Replies
3
Views
115
loki1944
L

Latest posts

Back