Let the 6970 Rumors Begin

Status
Not open for further replies.
vjcsmoke said:
This is false, I don't know where this rumor got started and why people keep spreading it...
Click to expand...
Sigh, read what I typed again. I said the 5970 has 2x5850 CLOCKED chips. I already knew it has 2 full 1600 SPU Cypress chips and I would expect anyone with any sort of interest for that card to know it as well. I'm not about to start spelling out every single thing for laymans to understand.

P.S: The Polish website just pulled the article, but I managed to save all the pictures. I'll put em up soon in this thread
 
Last edited:
Lorien said:
P.S: The Polish website just pulled the article, but I managed to save all the pictures. I'll put em up soon in this thread
Click to expand...

Haha, I'm trying to dig out the images out of my Firefox cache as we speak. I have so far most of the slides now.

EDIT: Tamlin saves the day!
 
Here we go:
fot010.jpg

fot011l.jpg

fot012.jpg

fot013.jpg

fot014c.jpg

fot015.jpg

fot016u.jpg

fot017.jpg

fot018.jpg

fot019.jpg

fot020r.jpg

fot021.jpg

fot022.jpg

fot023.jpg

fot024l.jpg
 
These slides were made a while back (when they thought they would release these on the 22nd of November) and the "TBD" for SIMD count, Stream Processors and TMUs is quite telling. They were having yield problems.
 
Silus said:
These slides were made a while back (when they thought they would release these on the 22nd of November) and the "TBD" for SIMD count, Stream Processors and TMUs is quite telling. They were having yield problems.
Click to expand...

Haha nice try. AMD suddenly had yield problems after having sold over 25 million 40nm chips. The lack of information was there to not let vital information leak and keep the competition guessing. They took that a step further and paraded a crippled sample around AIBs all for the sake of secrecy.They've been doing this since before the launch of the 4870, people should not be surprised of this anymore.
 
Lorien said:
Haha nice try. AMD suddenly had yield problems after having sold over 25 million 40nm chips. The lack of information was there to not let vital information leak and keep the competition guessing.
Click to expand...

What exactly does "yield problems" have to do with selling "25 million 40 nm chips" ? Not a very successful spin there :)
Also, Cayman should be a pretty big chip (even if not as big as GF100 or GF110) and it's not free of yield problems just because you think it is...

Hehe "competition guessing"...Yeah that must be it. It's funny that you seem to think that these leaks aren't known by NVIDIA for a while now. Much like AMD knows about some of NVIDIA's products, before we even saw any leaks.
 
Silus said:
These slides were made a while back (when they thought they would release these on the 22nd of November) and the "TBD" for SIMD count, Stream Processors and TMUs is quite telling. They were having yield problems.
Click to expand...

There is no Yield issue at all.. where you smoking from? can you be more specific than just saying "BASED on what I SEE"?
 
Silus said:
What exactly does "yield problems" have to do with selling "25 million 40 nm chips" ? Not a very successful spin there :)
Also, Cayman should be a pretty big chip (even if not as big as GF100 or GF110) and it's not free of yield problems just because you think it is...

Hehe "competition guessing"...Yeah that must be it. It's funny that you seem to think that these leaks aren't known by NVIDIA for a while now. Much like AMD knows about some of NVIDIA's products, before we even saw any leaks.
Click to expand...

Is that why they released GTX 580 early? To "surprise" AMD with their 20% sneak attack?
 
Silus said:
What exactly does "yield problems" have to do with selling "25 million 40 nm chips" ? Not a very successful spin there :)
Also, Cayman should be a pretty big chip (even if not as big as GF100 or GF110) and it's not free of yield problems just because you think it is...

Hehe "competition guessing"...Yeah that must be it. It's funny that you seem to think that these leaks aren't known by NVIDIA for a while now. Much like AMD knows about some of NVIDIA's products, before we even saw any leaks.
Click to expand...

Must be how Nvidia knew about RV770's 800 SPs and priced the GTX 280/260 accordingly :eek:
 
shansoft said:
There is no Yield issue at all.. where you smoking from? can you be more specific than just saying "BASED on what I SEE"?
Click to expand...

Oh I'm sorry, I should post based on what I don't see ? Or maybe based on what you say ? Would that be better ? Give me a break :)

Ok then, if there's no yield issue and you're not basing that "on what you see", go ahead and share what you say you know. As you criticized me for doing the same (on a speculation thread at that), you probably understand that just posting "there's no yield issue at all" isn't really enough :)
 
Facts:

Those slides say October. We know that sites were briefed on Cayman briefly in October during Barts' presentation (read the [H] report and you'll know this is true)

The new leaked slides were dated November 18, 2010 (the 6990 slides) and we know AMD had a presentation for Cayman for the end of November

Put two and two together... they didn't have the specs listed in October in order to hide em. Closer to release, the final specs were revealed

And FYI, those TBD specs do not say anything about yield. The card is so close to release, they can't lop things off within a month, even if its delayed a few weeks. That simply isn't possible, and believing so shows a serious lack of understanding in how a SKU is designed, built, and tested.

So please, try harder
 
Vexerz said:
Is that why they released GTX 580 early? To "surprise" AMD with their 20% sneak attack?
Click to expand...

Was it early ? In the other threads in here, the GTX 580 was a year late or something. So now, it's early ? I can't keep up with the spin around here :)
 
Silus said:
Was it early ? In the other threads in here, the GTX 580 was a year late or something. So now, it's early ? I can't keep up with the spin around here :)
Click to expand...


http://hardforum.com/showpost.php?p=1036432681&postcount=201
AMD either has manufacturing problems or was caught off guard with the GTX 580, which they didn't think would provide that sort of performance increase over the GTX 480 (you know the one you call "pathetic") and is preparing new BIOS for the cards to make sure it is faster (it will not spank anything) or at least be close in performance, which requires further testing.
Click to expand...
The underlined part kills me. Thanks for the laugh. Everyone and their pet dogs new what Fermi done right would perform at.
 
ZerazaX said:
Facts:

Those slides say October. We know that sites were briefed on Cayman briefly in October during Barts' presentation (read the [H] report and you'll know this is true)

The new leaked slides were dated November 18, 2010 (the 6990 slides) and we know AMD had a presentation for Cayman for the end of November

Put two and two together... they didn't have the specs listed in October in order to hide em. Closer to release, the final specs were revealed

And FYI, those TBD specs do not say anything about yield. The card is so close to release, they can't lop things off within a month, even if its delayed a few weeks. That simply isn't possible, and believing so shows a serious lack of understanding in how a SKU is designed, built, and tested.

So please, try harder
Click to expand...

So you think that if they had yields problems, they would mention it in some marketing slide ? And if it's not, it's because there isn't any ? That IS funny :)

Well, let's make one thing clear about yields then, since you threw the "you don't know anything about what you are talking about" card.

Having yields problems, doesn't mean the chip isn't manufactured already. It just means that in its current state, they can't put it up to what they designed it for, which is their target from the start. This means that if they want to meet a certain release date, they have to make some compromises in the chip itself, like disabling some units. If they have more serious problems, they might go for a respin, but that doesn't necessarily improve yields.

Having said that, these problems didn't occur in a month of the release date. They happened much before. Some people (you included) seem to think that when we get the leaks or rumors of delays, it's because that happened just a few days ago prior to the rumor, which is ludicrous.

Still I'm curious, what is exactly the reason you believe to be behind Cayman's delay ?
 
LOL at Silus being burned by FERMI release. :p

He's seeing bad yeilds around every corner after the woodscrew fiasco. :D

On topic, it seems like there are major changes in Cayman vs. Barts or Cypress:
Google translate from the review that got pulled (I still had it open)
The most important difference between the architecture, which can be immediately seen when comparing the block diagram for the systems and Barts Cayman is the existence of dual graphics engine (Graphics Engine) in the first. Thus was not only doubled the number of teselatorów, rasteryzatorów, hierarchical modules responsible for the elimination of invisible to the observer on the 3D scene objects (Hierarchical Z-buffer), or modules Geometry and Vertex Assembler Assembler, but most of all greatly increased the efficiency of the vertex during the operation.
Click to expand...
fot013.jpg
 
Last edited:
vjcsmoke said:
A 5850 only has 1440 SPs.
It looks to me like the 6970 is going to have 1920 SP and 2GB DDR5.

Going by SPs alone, and assuming perfect scaling, that would mean that a 6970 would be approximately 2.2x as powerful as a 5850.
Click to expand...

How does 1920/1440 = 2.2 ?

My calculator says that is 1.3.
 
Not sure where 2.2x more powerful comes from, but I don't see 1.3 either.

1920/4=480, 1440/5=288, 480/288=1.67, and with 90% efficiency= 1.5x the shader power.

Everyone is using this based on utilizing 80% 5d shaders.
 
I'm hoping for a 6950 that OC's to 6970 speeds, similar to what the 6850/6870 are like. ;)
 
Snowdog said:
How does 1920/1440 = 2.2 ?

My calculator says that is 1.3.
Click to expand...

flexcore said:
Not sure where 2.2x more powerful comes from, but I don't see 1.3 either.

1920/4=480, 1440/5=288, 480/288=1.67, and with 90% efficiency= 1.5x the shader power.

Everyone is using this based on utilizing 80% 5d shaders.
Click to expand...

It's either a innocent mistake or that over-hyping happening again. It's like in Braveheart when William Wallace greets the scotsmen on the battlefield and says "Patriots of Scotland, I am William Wallace" and the guy says "William Wallace is 7 feet tall" then Mel says "I know and if he were here he'd consume the english, with fireballs from his eyes and lightning bolts from his arse, I am William Wallace!"

It's like when the 6970 comes out and trades blows with GTX 580 people will say wait it's supposed to be much more powerful? What happened. I'm rooting for the card but again I can't help but snicker at the inflated expectations every time a new leak comes out.
 
I'll be surprised if it beats the GTX 580, simply because that's not ATI's track record.
Heck, I'd be impressed even if they trade blows. Really comes down to the price, I suppose.
 
Thanks for the slides. Really fascinating. 6970 is probably what I need but the 6990 is what I want.
 
UtopiA said:
I'll be surprised if it beats the GTX 580, simply because that's not ATI's track record.
Heck, I'd be impressed even if they trade blows. Really comes down to the price, I suppose.
Click to expand...

Me too I'd be impressed. Especially if it comes at a aggressive price point it'd be a winner. I dont think they need to beat nvidia in performance just in price and deliver a fantastic card. They seem to be on track to do that so I'm excited for the card. The wait is killing me though. I've already got the money for my new build. Budget $3600. It'll be a fuckin monster. It'll include 3 27 inch monitors or so. 120hz, 2 6970s or GTX 580s 12gb gddr 3 push pull water cooling. I'll be waiting for sandy bridge to see if I can pick a SB proc that will get me to 5.5Ghz on water or more. This all has the thumbs up from the wife. Black friday and Cyber monday I'll order the ram, 3 monitors, Monitor arm for easy mounting, water cooling kit, Solid state primary hdd. The rest after Xmas time. I'm long overdue for an upgrade so as you can tell I'm excited.
 
Last edited:
UtopiA said:
I'll be surprised if it beats the GTX 580, simply because that's not ATI's track record.
Heck, I'd be impressed even if they trade blows. Really comes down to the price, I suppose.
Click to expand...

It'd be hilarious if it beats the 580 and is cheaper in the process. I recall how people predicted the 68xx series was going to be "slow;" and we saw the difference between that and the 57xx series.

Again, the next few weeks will be exciting.

maybe a 300-325 6950 is bette
Click to expand...

Hopefully it's $300, that'd be awesome.
 
I'm wondering if there will be a way to disable the "power containment" feature. Or are we just ignoring that AMD has now opened the door to throttling in all application (and not just OCTT/Furmark like Nvidia)? Also, so I'm up to speed, what's the plan for how are we going to spin that into a good thing?
 
Forceman said:
I'm wondering if there will be a way to disable the "power containment" feature. Or are we just ignoring that AMD has now opened the door to throttling in all application (and not just OCTT/Furmark like Nvidia)? Also, so I'm up to speed, what's the plan for how are we going to spin that into a good thing?
Click to expand...

Correct me if I'm wrong but didn't Wizard crack this and is offering a tool in GPUZ to bypass the throttling.

Edit
Actually yes he did here it is http://techiser.com/disable-geforce-gtx-580-power-throttling-gpuz-129758.html
 
Forceman said:
That's for the 580 - I was talking about the 6970. Hopefully the quick success cracking the 580 would lead to quick work cracking the AMD version, but it sounds like the AMD version may be more low-level than Nvidia's implementation.
Click to expand...

Oh sorry dude, I misunderstood u.
 
Well since the slide says that power containment can be configured with AMD Overdrive, I imagine that either means it can be disabled or that the power limit can be raised by overclockers. Maybe it won't be as much of an inconvenience as you think. I guess we wait until reviews...
 
wouldnt the two polygons per clock make it automatically 2x as fast as the 5870?
 
2x faster where tessellation is a bottleneck for the HD 5870, looks like 1920 shader units at ~10% increased clocks. Around 33% faster just by raw numbers than the HD5870.
 
razor1 said:
2x faster where tessellation is a bottleneck for the HD 5870, looks like 1920 shader units at ~10% increased clocks. Around 33% faster just by raw numbers than the HD5870.
Click to expand...

1920 / 4 = 480(6970)4D
1600 / 5 = 320(5870)5D

so you have a 50% increase in raw shader power, Factor in the clock change, then you have a bit more, now you have 150% shader power of a 5870+ clock speed difference.

That's not counting the extra TMUs and extra memory bandwidth, improved tessellation engines, and this new thing they are calling "dual graphics engines".

so who knows, it might end up being some monster card after all.


For those complaining about the new power management implemented in the 6900 series, if you missed the last bullet on the slides it said.

"User controllable via AMD Overdrive Utility"
 
Digital Viper-X- said:
1920 / 4 = 480(6970)4D
1600 / 5 = 320(5870)5D

so you have a 50% increase in raw shader power, Factor in the clock change, then you have a bit more, now you have 150% shader power of a 5870+ clock speed difference.

That's not counting the extra TMUs and extra memory bandwidth, improved tessellation engines, and this new thing they are calling "dual graphics engines".

so who knows, it might end up being some monster card after all.


For those complaining about the new power management implemented in the 6900 series, if you missed the last bullet on the slides it said.

"User controllable via AMD Overdrive Utility"
Click to expand...


That isn't 50% increase in shader performance, you have 50% increase in the number vector units, but only 20% increase in shader units. So effectively increasing shader power by 20%. We have seen from theoretical shader tests that the 5d unit is pretty effecient over all and is very compiler bound with its performance. Same should be true with the 4d unit.

The HD5870 doesn't seem to be TMU bound in todays games so we probably won't see any major increase in performance from the extra TMU's, tessellation there should be increased performance there, and at high AA we should see the extra bandwidth come into play.
 
Last edited:
Status
Not open for further replies.
Back
Top