Radeon HD 7000 Revealed: AMD to Mix GCN with VLIW4 & VLIW5 Architectures

[Olivon]

12 module, 384-bit is more logical ^^
Good eyes Jawed and boxleitnerb

[spursindonesia]

Hmm, 384 bit Tahiti, this certainly won't be any smaller chip than Cypress, though perhaps not up to Cayman's size.

[Craftyman.]

I remember how people were laughing at his posts about bulldozer until bulldozer launch day...

if all you have is hate for a brand, you're gonna be right some of the time. You don't really think that just because he was right about bulldozer means that he'll be right about Southern Islands do you?

ive been looking through this thread and im disappointed in the hate and expectations.

this gen is a 1st gen on a new arch so the drivers will suck, its on a new process affter skipping a gen and having very little test hardware if at all, and it was supposed to be out last year showing up the 5870 and since there were none of thees out they did not get the time to make a hardware changes based on performance needs. so its going to be alot like the 4890 v 5870, with that fight u had a fully working driver with a high clocked reworked core to maximize overclocking vs a heavy change in the GPU on a new process. this will be the same and at launch it wont do much better as the hd6 is the perfected hd5.

so as i see it, if amd gets better than the hd6 at launch it will be a success since if they can keep xfire scaling up, it will let amd keep the performance crown

Wasn't the 8800GTX on a new architecture? That thing turned out to be amazing. Of course, that wasn't AMD...

Anyway I'm just throwing out ideas here.

[tajoh111]

if all you have is hate for a brand, you're gonna be right some of the time. You don't really think that just because he was right about bulldozer means that he'll be right about Southern Islands do you?



Wasn't the 8800GTX on a new architecture? That thing turned out to be amazing. Of course, that wasn't AMD...

Anyway I'm just throwing out ideas here.

The reason the gtx 580 was amazing at the time was there was also such a huge increase in the amount of shaders compared to the old architecture.

From 24 and 8 pixel and vertex shaders respectively to 128 unified at much higher clocks too. These were real shaders too. The r600 was so disappointing because it relied on only 64, 5 way shaders and relied on a new AA scheme which relied on more shaders to do the work.

[DilTech]

The reason the gtx 580 was amazing at the time was there was also such a huge increase in the amount of shaders compared to the old architecture.

From 24 and 8 pixel and vertex shaders respectively to 128 unified at much higher clocks too. These were real shaders too. The r600 was so disappointing because it relied on only 64, 5 way shaders and relied on a new AA scheme which relied on more shaders to do the work.

The 8800GTX was also in the hands of more than a few companies before the 7800GTX 256 hit the market. I still remember the arguments on this very site because I had already had hands on time with the thing when sites like xbitlabs were claiming it'd be a hybrid chip. Granted, the original G80 chip wasn't the 384-bit monster that released to the public (it was 256bit, actually closer to matching the 8800GT than the 8800GTX), but NVidia spent a LOT more time perfecting that card than anyone on here would believe. After all, it was actually in the design phase before we even saw the NV40 (G80 was titled the NV50 at that time, IIRC) and had received mention a few times during that time frame. NVidia spent YEARS on that design, and it paid off big time.

NVidia actually puts a lot more time into the design and testing phase than most people on here would believe could be even remotely profitable (a part of the reason their cards are a bit higher in the price range), just to make sure they never have another FX mistake. Late is one thing, but late AND extremely under-performing isn't going to happen again.

AMD/ATi have had issues when it comes to full scale brand new designs the past few attempts, and they usually fix it the next time around. HD 2900 was fixed with the HD 4800 (the 3870 was a step in the right direction, but it still wasn't without issue). The X1800 was fixed with the X1900, although the X1800 performance wasn't "bad" per-say either.

That said, I'm actually seriously hoping AMD show up with a serious competitor. I'd prefer we don't see stagnation in the gpu market like we have in the CPU market. These 30-40% jumps just don't catch me like we use to see. Remember the jump from the 9800xt to the x800xt? What about the 7900GTX to the 8800GTX? Speed jumps have dwindled, and games taking advantage have dwindled along with it. On a positive note though, you can buy a solid card and actually have it perform admirably for a longer time-frame now, which is definitely a positive note.

[cal_guy]

The reason the gtx 580 was amazing at the time was there was also such a huge increase in the amount of shaders compared to the old architecture.

From 24 and 8 pixel and vertex shaders respectively to 128 unified at much higher clocks too. These were real shaders too. The r600 was so disappointing because it relied on only 64, 5 way shaders and relied on a new AA scheme which relied on more shaders to do the work.

Those pixel shaders had two sets VEC3+1 Scalar per pixel shader (however texture unit address calculation were done by the pixel shader units) while the vertex Shaders had 1 VEC4+1 unit per shader. So it was 192 + 40 versus 128.

[LordEC911]

The reason the gtx 580 was amazing at the time was there was also such a huge increase in the amount of shaders compared to the old architecture.

From 24 and 8 pixel and vertex shaders respectively to 128 unified at much higher clocks too. These were real shaders too. The r600 was so disappointing because it relied on only 64, 5 way shaders and relied on a new AA scheme which relied on more shaders to do the work.

Not really... what you should have said is G80 could handle 128 threads per clock while R600 could only do 64 threads per clock.
http://www.anandtech.com/show/2231/4

or if you want Derek's comparison...
http://www.anandtech.com/show/2231/6

[HelixPC]

I hope these are a big step above the 6000 series. I'm still going to wait for nVidia's response before i buy a new card. AMD's drivers just arent that good.

[HelixPC]

Ive run gtx 580 SLI on the corsair AX 750, with a 2600K @ 4.8ghz and GTX 580s overclocked. No problems at all

[AliG]

I hope these are a big step above the 6000 series. I'm still going to wait for nVidia's response before i buy a new card. AMD's drivers just arent that good.

Yeah I agree, ATI might have a better hardware model at the moment, but I'm fed up with the instability of their drivers. If the Nvidia product is at a similar performance level and runs at reasonable temps/power consumption I'm going to go with them next round simply because I trust their software guys a lot more

[tajoh111]

Not really... what you should have said is G80 could handle 128 threads per clock while R600 could only do 64 threads per clock.
http://www.anandtech.com/show/2231/4

or if you want Derek's comparison...
http://www.anandtech.com/show/2231/6

The Jist of what I wanted to say was, first generation, r600. It was shaders starved. That why it did so bad at the time with AA, since it used shader power for it.

Another thing I wanted to clarify if any people didn't know is I was talking about the gtx 8800. I read the anandtech article a long time ago. The thing with AMD 5 way shaders is in a best case scenario they have the ability to process 5 threads or 320. But in a worst case it is 64. Increase the clock or adding more shaders, removed this bottleneck and AMD had fantastic gains. Which is why AMD had better results with the 4870 than NV did with the gtx 280.

[Kain665]

The 8800GTX was also in the hands of more than a few companies before the 7800GTX 256 hit the market. I still remember the arguments on this very site because I had already had hands on time with the thing when sites like xbitlabs were claiming it'd be a hybrid chip. Granted, the original G80 chip wasn't the 384-bit monster that released to the public (it was 256bit, actually closer to matching the 8800GT than the 8800GTX), but NVidia spent a LOT more time perfecting that card than anyone on here would believe. After all, it was actually in the design phase before we even saw the NV40 (G80 was titled the NV50 at that time, IIRC) and had received mention a few times during that time frame. NVidia spent YEARS on that design, and it paid off big time.

NVidia actually puts a lot more time into the design and testing phase than most people on here would believe could be even remotely profitable (a part of the reason their cards are a bit higher in the price range), just to make sure they never have another FX mistake. Late is one thing, but late AND extremely under-performing isn't going to happen again.

AMD/ATi have had issues when it comes to full scale brand new designs the past few attempts, and they usually fix it the next time around. HD 2900 was fixed with the HD 4800 (the 3870 was a step in the right direction, but it still wasn't without issue). The X1800 was fixed with the X1900, although the X1800 performance wasn't "bad" per-say either.

That said, I'm actually seriously hoping AMD show up with a serious competitor. I'd prefer we don't see stagnation in the gpu market like we have in the CPU market. These 30-40% jumps just don't catch me like we use to see. Remember the jump from the 9800xt to the x800xt? What about the 7900GTX to the 8800GTX? Speed jumps have dwindled, and games taking advantage have dwindled along with it. On a positive note though, you can buy a solid card and actually have it perform admirably for a longer time-frame now, which is definitely a positive note.

NV40 was 6800, NV50 was changed to G70/71 pretty early in development IIRC. Not sure what G80 was codenamed.

[DilTech]

NV40 was 6800, NV50 was changed to G70/71 pretty early in development IIRC. Not sure what G80 was codenamed.

NV47 was the G70/G71. NV50 was the G80. Trust me on that one.

[spursindonesia]

Yeah, G70/71 is much closer to NV 40 than to G80, mArch wise speaking.

[Elmy]

Supposedly less than a month away and no leaked benchmarks yet .....

[SKYMTL]

Supposedly less than a month away and no leaked benchmarks yet .....

What difference would benchmarks done with early non-release drivers make at this juncture?

Wait for launch so you get the full picture right away instead of mostly pointless selective benchmarks.

[kaktus1907]

5+1 PWM, 2x6 pin, dual bios switch, looks smaller than Cayman

http://chinese.vr-zone.com/index.php...hiti-12092011/

[Mechanical Man]

5+1 PWM, 2x6 pin, dual bios switch, looks smaller than Cayman

http://chinese.vr-zone.com/index.php...hiti-12092011/

Closer one looks 8-pin to me.

[3lfk1ng]

I see dual 6's

[mrcmtl]

Closer one looks 8-pin to me.

Just count the number of silver connectors coming out from the back.

[highoctane]

There are better shots floating around, it's dual 8 pin connectors.

[3lfk1ng]

Nope, we are all wrong.

There are two different variants of the car floating around as confirmed by VR-Zone.

There is this screen shot (Tweaktown)

It also includes two 8-pin PCI Express power connectors, which should give it up to 375W of power to suck down. The card, however, is powered by the SIG-approved 6-pin plus 8-pin config, which uses up to 300W of power.

Then there is this one (vr-zone)

There is a crucial difference from the previous leak - this sample consists of 2 x 6-pin PCI-e connectors, as opposed to the 8-pin + 6-pin configuration previously captured. This could suggest that this is either an advanced production ready sample, or Tahiti Pro, or HD 7950, which should feature a lower TDP than Tahiti XT / HD 7970. This card thus has a TDP somewhere between 150W and 225W.

Read more: http://vr-zone.com/articles/more-amd...#ixzz1g4OyQPE1

[bingo13]

Here we go with the wild, over inflated hopes and dreams of some.....again.

I am waiting on reputable sites like DigiTimes to report benchmark numbers before setting my expectations. Oh wait, I guess I could just benchmark the card sitting here and see if those hopes and dreams will be dashed or not.

[Lightman]

I am waiting on reputable sites like DigiTimes to report benchmark numbers before setting my expectations. Oh wait, I guess I could just benchmark the card sitting here and see if those hopes and dreams will be dashed or not.

I suggest you do that! Plus get new GPUClock Tool from AMD and check how well/bad your sample is clocking! New 28nm process so a lot to be learnt again

[LordEC911]

Crossfire config is two 7970 ES.
Retail card should be ~210w TDP, 6+1 phase and 8+6pin.

VR-Zone's is a MSI 7950.
384bit and will come in both 1.5gb and 3gb flavors.
TDP is supposedly around 170-180w, 5+1phase and two 6pins.