AMD GCN 2.0 "Tenerife" leaked Slide from S/A

[tviceman]

I think it's fake. Even if you toss out the spelling error, just has a fake feeling to it.

Yeah I just noticed the spelling error. "enableing" Hahaha. Also, the slide is saying this card will have less outputs than the hd7970 has. I'm going with fake now.

 

[tviceman]

Not seeing any spelling errors.

Look in the fine print. "enableing"

 

[Homeles]

Look in the fine print. "enableing"

Ah. Well that's very "AMD" of them. They done it before.

 

[BallaTheFeared]

You need to go look up that word and learn its definition if you you're going to use it correctly.

I used it correctly. You need to stop talking to me, or has that fact slipped your mind already like the past six years have?

 

[aaksheytalwar]

It is impossible that a 8970 is just 10-15% faster than a 7970. Even an Oced 7970 would beat a stock 8970 in that case. IMO we should expect 8970 to be about 40-50% faster at stock

 

[Homeles]

It is impossible that a 8970 is just 10-15% faster than a 7970. Even an Oced 7970 would beat a stock 8970 in that case. IMO we should expect 8970 to be about 40-50% faster at stock

For an architecture tweak on the same manufacturing process? You're going to get let down pretty hard. An overclocked 5870 could beat a stock 6970... this is a similar situation.

 

[tviceman]

"As of March 2012... has been calculated at over 4.5..."

Hahahahaha, read that entire sentence. That sounds like a 14 year old kid wrote it, not a PR department of a major chip company. This is clearly the work of a bored individual trying to get attention.

 

[Homeles]

"As of March 2012... has been calculated at over 4.5..."

Hahahahaha, read that entire sentence. That sounds like a 14 year old kid wrote it, not a PR department of a major chip company. This is clearly the work of a bored individual trying to get attention.

Or just AMD's world-renowned marketing team at it again.

 

[LOL_Wut_Axel]

Did I miss the release and review of the GK110 coz there's an awful lot of assumptions being made about it. Or we just taking NVIDIA Italy PR as gospel now?

History: read it, learn it, love it.

AMD knows Tahiti XT will be fine against GK104, but against GK110 it will simply be outclassed.

And it's a very good assumption to make that GK104 won't be much faster than Tahiti XT based on the fact that by the end of this year it'll slot in as a Performance card in NVIDIA's lineup. GK104 CAN be faster than Tahiti XT, just not anything over 20% faster. To me 10-15% faster seems more likely IF it ends up being faster. Remember that the engineering goal every time is to get 50-65% higher performance with a new node + process comparing the highest-end GPU replacements. As in, GK104 needs to be 50-65% faster than GF114 and GK110 will need to be 50-65% faster than GF110.

There's some talk about PhysX and what not giving GK104 a big boost, but PhysX is irrelevant for 95% of PC gamers so I don't care for it in the overall scheme.

 

[MrK6]

GK110 beating Tahiti is a pretty logical assumption if you ask me. Gigantic die of doom squashing a cute little 352mm² die... seems pretty reasonable.

Given Nvidia's consistent failures with their large GPUs, I'd imagine they'd start toning back on the die size. GT200 was a disaster, GF100 was a disaster, and GK100 got scrapped, with GK110/112 (or whatever you want to call it) coming far later. GF110 was just fixing their broken GF100, although kudos to Nvidia for releasing it in a timely manner.

It'll be interesting to see if GK110 is an actual tweak to the GK100 Kepler line - the cards we should be seeing in around a month.

Logic would imply that an expectation would be based on fact(s) or concrete evidence. So far we all we have are rumors and hearsay, and to bet anything on it is simply faith-based. For all we know, the GK110 may never exist. To claim something is certain based on no proof is relying on faith, which is something fanatics do, not logical people.


Anyway, back to reporting people who directly disobey moderator orders 😉.

 

[LOL_Wut_Axel]

It is impossible that a 8970 is just 10-15% faster than a 7970. Even an Oced 7970 would beat a stock 8970 in that case. IMO we should expect 8970 to be about 40-50% faster at stock

You never looked at the HD 6970 vs the HD 5870, did you?

The HD 6970 is 16% faster than the HD 5870. Going from 5000 series to 6000 series was a refresh that brought a newer, more refined architecture. The exact same thing is in play going from the 7000 to 8000 series.

The HD 8970 will be anywhere from 10-20% faster than the HD 7970, and the card will probably be aimed against the GTX 770.

History repeats itself time after time after time.

 

[skipsneeky2]

Now, assuming this the HD 8970 will be anywhere from 10-20% faster than the HD 7970. If that's the case, and it probably will be, the top-tier GK110 card will easily have higher performance (by 10-15%). We'll be at the same situation as the HD 6970 vs. GTX 580 by the end of the year, where GK110 will be 10-15% faster but consume much more power. AMD will end up competing with the GTX 770 at the top. Mark my words.

Will this reflex the price of the 7970 you believe ?

10-20% isn't much faster so either this card retails for like $650+ or it takes the 7970 price and the 7970 sits back seat around $350-$400 and the 7950 sits around $300 maybe?

 

[formulav8]

That's just 4 more CUs, what are the chances that they're there in Tahiti and fused off waiting for a more mature 28nm process?

Even if a fake slide, I was thinking from the get go that the GCN gpu just might have parts disabled. So I won't be surprised if thats the case. They wanted to be first to market. And they did. First 1ghz gpu as well, and so on. Waiting for improved yields and so on. But only amd knows for sure.

So anyways, They can already easily get 30% performance just using clockspeeds. So when the real, full kep comes out, AMD will have something waiting for them, and I don't think clockspeed is the only thing.

One wacky thing though is that many games are really getting cpu limited. So it will end up being at some point where you will no matter what but have to overkill the graphics settings to really show any major differences. 79xx already easily handles every single game out there and at very nice eye candy settings...

 

[LOL_Wut_Axel]

Will this reflex the price of the 7970 you believe ?

10-20% isn't much faster so either this card retails for like $650+ or it takes the 7970 price and the 7970 sits back seat around $350-$400 and the 7950 sits around $300 maybe?

Remember that GK104, which will launch soon, will force prices down. If it ends up being the same speed as Tahiti XT, expect prices for the 7970 to only go down to $500 or so because NVIDIA won't launch a card that's faster than the GTX 680 for more than half a year. Now, if it ends up being 10-15% faster than the 7970, expect the 7970 to go down to $400-450.

As far as the other cards go, I expect we'll see more competition there. I wouldn't be surprised if lower-tier GK104 cards means we get to see the HD 7950 at $350-400, the HD 7870 at $250-280, and the HD 7850 at $200-230. If there's a price spread like the one above ($100 between the 7950 and 7870), AMD could can launch a 1.5GB HD 7950 at $300-330 and/or an HD 7890 at $270-300. Everything depends on how much NVIDIA forces prices down, which they probably will. That's good, because I can get an AMD card much cheaper than I normally would have. 😛

 

[Homeles]

Even if a fake slide, I was thinking from the get go that the GCN gpu just might have parts disabled. So I won't be surprised if thats the case. They wanted to be first to market. And they did. First 1ghz gpu as well, and so on. Waiting for improved yields and so on. But only amd knows for sure.

The rest of the world would know as well, if they looked at the die shot linked on the first page.

 

[gladiatorua]

If you take into account that GCN is new, the 28nm is immature and 79xx have quite large overclocking headroom you can easily expect 8970 to be 30% faster than 7970. It would be stupid not to refresh very soon after BigK.

 

[MegaWorks]

Kepler killer?

 

[Nintendesert]

Kepler killer?

How can you kill that which does not exist? :whiste:






...yet. 😛

 

[VulgarDisplay]

For an architecture tweak on the same manufacturing process? You're going to get let down pretty hard. An overclocked 5870 could beat a stock 6970... this is a similar situation.

Like the other poster said, a 7970 can be 25% faster than a stock 7970 so any amd refresh being 40-50% faster is not an unreal expectation.

 

[MrK6]

Like the other poster said, a 7970 can be 25% faster than a stock 7970 so any amd refresh being 40-50% faster is not an unreal expectation.

It'd be pretty difficult to squeeze out that kind of performance consistently unless AMD has a breakthrough in architecture design. The slide itself says it has 1.2+ times more compute power than a 7970, so that's where my expectation lies. I'd imagine 2304 shaders at ~1GHz would do the trick.

 

[tangrisser]

Good thing I didn't buy a 7970. Can't wait for 8xxx series!

 

[3DVagabond]

It'd be pretty difficult to squeeze out that kind of performance consistently unless AMD has a breakthrough in architecture design. The slide itself says it has 1.2+ times more compute power than a 7970, so that's where my expectation lies. I'd imagine 2304 shaders at ~1GHz would do the trick.

Just so I understand, are you thinking that the improved performance comes from purely the SP increase and increased clocks? I read that as the arch has been improved so it gives 20% improvement.

Good thing I didn't buy a 7970. Can't wait for 8xxx series!

Q3 2012 to Q1 2013 means a 6 month to a year wait. Unless you are sitting on acceptable performance now, that's a pretty long wait.

 

[OCGuy]

Chiphell with more fake BS. AMD could do a 7970 refresh by just cherry-picking the top chips and setting the clocks higher.

 

[LOL_Wut_Axel]

If you take into account that GCN is new, the 28nm is immature and 79xx have quite large overclocking headroom you can easily expect 8970 to be 30% faster than 7970. It would be stupid not to refresh very soon after BigK.

Even with a more mature process you still have the fact that they're adding more compute units to the GPU, thereby making the die size bigger and raising power consumption. Also, no matter what you do, you still can't eliminate the fact that power consumption increases exponentially as you raise clock speeds, especially in graphics cards. Yeah, the 7970 has big overclocking headroom, but when you overclock to high amounts power consumption surpasses that of the GTX 580 by a noticeable amount.

AMD does not want to go over the 250W TDP limit/barrier they've set themselves.

In other news, there's some rumors going on about the HD 7990 being canceled. IMO, that makes no sense: the HD 4870X2 consumed much more power than the HD 5970 and HD 6990 IIRC.

 

[Homeles]

Like the other poster said, a 7970 can be 25% faster than a stock 7970 so any amd refresh being 40-50% faster is not an unreal expectation.

I go based off previous evidence. Previous evidence shows that refreshes on the same manufacturing process don't get anywhere near 40-50%.

5870 -> 6970 = 10-15%
480 -> 580 = 10-15%