AMD: R8xx Speculation

How soon will Nvidia respond with GT300 to upcoming ATI-RV870 lineup GPUs

  • Within 1 or 2 weeks

    Votes: 1 0.6%
  • Within a month

    Votes: 5 3.2%
  • Within couple months

    Votes: 28 18.1%
  • Very late this year

    Votes: 52 33.5%
  • Not until next year

    Votes: 69 44.5%

  • Total voters
    155
  • Poll closed .
It's still the tree names bugging me for some reason, why trees?

It keeps me awake late at night, honest. :oops:

Guilty conscience ;) ... about the wasteland they left in the third world with this new products quick release scheme.

Do you really think they think green as they logo shows. Bingo :thumbup: they do .... they think about green bucks they rip off from a happy customers :LOL: :rofl:
 
It's still the tree names bugging me for some reason, why trees?

It keeps me awake late at night, honest. :oops:

Because being all ecoactivist and green and whatnot is good marketing nowadays, and what could be more green than evergreen trees? (Family name Evergreen, all chips named after really evergreen trees)
 
Because being all ecoactivist and green and whatnot is good marketing nowadays, and what could be more green than evergreen trees? (Family name Evergreen, all chips named after really evergreen trees)

IIRC, AMD got some complaints for naming its new mobile platform Congo; that evokes genocide more strongly than it does the river which is what AMD had intended. These new names are a far cry from the early days of the industry. After 3DFX used the code name Napalm and hearing retail names like "Apocalypse 5D," I wholly expected a card to debut as the "Genocide 128" or be code named something like "My Lai FX," "Darfur 3D," or "Zyklon-B 256."
 
It's still the tree names bugging me for some reason, why trees?

It keeps me awake late at night, honest. :oops:

It's probably just random. Geeks tend to quickly run out of meaningful code names, so you end up with groups of things named in a random fashion.

In a place I worked, servers were named after planets, training room 1 machines were named after cartoon characters, training room 2 machines were named after spaceships from movies, etc. Every time someone commissioned a group of machines for something, someone chose a random group they could easily pick names out of.

It simply wasn't worth the time or effort to come up with something incredibly clever and meaningful, so it was random groups/names just pulled out of the air by whoever was doing the work at the time.

Whoever it was at ATI that chose to use Canadian trees is probably laughing at the attempts to divine some kind of meaning out of something he probably spent ten seconds choosing because he happened to be looking out the window at a Juniper tree at the moment he wrote the memo with the new product code names.
 
It's still the tree names bugging me for some reason, why trees?

It keeps me awake late at night, honest. :oops:

Hemlock is a poisonous plant. They used it to kill people sentenced to death, in ancient Greece! Socrates died from that.

So hemlock could suggest that this is what ATI is going to give Nvidia to drink! lol

It could be Terry Makedon's idea too, since he has Greek roots and all. I imagine him while writing drivers for the new babies and after seeing what they can do, mumbling to himseld "that's what i am goind to feed you Nvidia, hemlock"! :devilish:

Other than that, eco friendly has become trendy and it is one of the very few trends that are actually any good. Trees/plants are eco friendly...!
 
Other than that, eco friendly has become trendy and it is one of the very few trends that are actually any good. Trees/plants are eco friendly...!

That's just Eco-Marketing. every biologist knows that a Tree produces as much CO2 when rotting away as it can ever "consume" whilst alive.

Only the Ocean has a positive CO2 balance, stupid tree huggers.
 
That's just Eco-Marketing. every biologist knows that a Tree produces as much CO2 when rotting away as it can ever "consume" whilst alive.

Only the Ocean has a positive CO2 balance, stupid tree huggers.

Actually, it depends on what tree are we talking about. Some have positive, others not.
 
It's still the tree names bugging me for some reason, why trees?

It keeps me awake late at night, honest. :oops:

No one has thought that these cards will be super fast in 3DMark2001 Nature test? Especially while rendering trees and grass? :D

On a serious note: are we getting any pre-launch event in London Dave?
 
On a serious note: are we getting any pre-launch event in London Dave?

Shh! that's secret! (and that's why Theo's story was BS!)

If he got anything close to invited, he would know he didn't need to be in SF on sept 10th.
 
As this should be "more than" 2 Teraflops then it could be that Cypress XT is something like 2000 SP @ 600-650 MHz (about 2.5 Teraflops) or 1600 SP @ 750 MHz (2.4 Teraflops). I would not be surprised if the first solution will be right, as the lower clock could also mean that faster versions could come later should the GT300 and derivatives be a lot faster at launch than Cypress.

Some of us hoped for 1920 shader @900MHz+ and around 3.5-4TFlops, but all we'll get for now is 1200SP @900MHz as rumor goes all way back to June.
And they don't need to do that kind of developement anymore, with MHz pumping. Since dx10 all they need to do is have a working product at a proper moment and they'll easily add more SPs later for 32nm TSMC node (or GF) to keep chip cost low as much as they can. So 1800-2400 SPs we could easily expect later for 60% smaller node and a lot of performance jump. Would it be RV890 or RV970 it's just to be seen later. Err, yep it would be a bush again.
 
Oh, so 3 independent display controllers? Would be very nice.
Every radeon (except the very first one, r100, so only since rv100) had two display controllers. (And it's a very similar story on the green side, every card since GeForce 2 MX / GeForce 4 (not GeForce 3) had two display controllers.)
That's 8 years now we've been stuck with 2 display controllers (ok there was an even longer time graphic cards only had 1...)

Ever since GF4MX/GF4 ;) since GF2MX had problems even with TV-out that needed extra TVool just like GF3s. And afair GF4 had only one tmdswhile very since rv100 Radeons rv150-r200-r250 had two tmds so you could have two LCDs on them even so nobody put two DVIs on these boards but DSUB HD-15 and DVI-I or 2x HD-15
 
keritto: It was a bit different:

R100: Single display controller, built-in TMDS/DAC, external TV-out (Rage Theater) required. Independent outputs not possible.

R200: Dual DC, built-in TMDS/DAC, external secondary DAC or TMDS required for 2 independent outputs (both analog or both digital). I'm not sure about TV-out, diagram of R200 shows internal controller, but almost every R8500 had RageTheater... maybe the diagram isn't correct, or the internal TV-out controller is somewhat broken. I think Herucles R8500 LE had Rage Theater chip, while external DAC was missing and if I remember, the board wasn't capable to deliver independent TV-output (at least when using ananlog D-SUB as primary output).

RV200/RV250 and SM2.0 Radeons: Dual DC, built-in dual DAC, single TMDS, external TMDS required for 2 independent digital outputs. RV100 was likely similar.

Dual (built-in) TMDS was present on R5xx and above.
 
Some of us hoped for 1920 shader @900MHz+ and around 3.5-4TFlops, but all we'll get for now is 1200SP @900MHz as rumor goes all way back to June.
And they don't need to do that kind of developement anymore, with MHz pumping. Since dx10 all they need to do is have a working product at a proper moment and they'll easily add more SPs later for 32nm TSMC node (or GF) to keep chip cost low as much as they can. So 1800-2400 SPs we could easily expect later for 60% smaller node and a lot of performance jump. Would it be RV890 or RV970 it's just to be seen later. Err, yep it would be a bush again.

There's nothing wrong with having very high expectations (irrelevant of IHV), but the reasoning above shows in essense that any IHV cannot just throw around endlessly with extreme SP counts and/or huge TFLOP rates for any given space of time because there are physical limits too.

Meaningless sterile FLOPs aside I'd be personally very surprised if Cypress wouldn't be up to 2x RV770 in terms of performance; first above all because those darned FLOPs (thank God) do not define a GPUs final performance.
 
Some of us hoped for 1920 shader @900MHz+ and around 3.5-4TFlops, but all we'll get for now is 1200SP @900MHz as rumor goes all way back to June.
And they don't need to do that kind of developement anymore, with MHz pumping. Since dx10 all they need to do is have a working product at a proper moment and they'll easily add more SPs later for 32nm TSMC node (or GF) to keep chip cost low as much as they can. So 1800-2400 SPs we could easily expect later for 60% smaller node and a lot of performance jump. Would it be RV890 or RV970 it's just to be seen later. Err, yep it would be a bush again.

I think we will see more than 1280 SP, because I don't think that ATI will go 900+ Mhz on the first parts. Process is still relatively new and high yields are needed that are unlikely to happen on an high clock part. If we look at the 40 nm parts until now, we see that RV740 and the announced Geforce shrinks have not much higher clocks compared to previous 55nm parts, if there is an improvement at all.
So I don't think we will see at launch something with clocks higher than RV790, i.e. (and I mean the base clock, not the OC parts). If we see the history R600 (80nm) was 743 Mhz, RV670 (55nm first iteration) was 750 Mhz and Rv770 (55nm second iteration but larger chip) was a 750 Mhz part, too. Cypress is rumored to be larger than Rv770, too, even considering the DX11 compliance. So I will bet on a 1600+ SP part.
 
I think we will see more than 1280 SP, because I don't think that ATI will go 900+ Mhz on the first parts. Process is still relatively new and high yields are needed that are unlikely to happen on an high clock part. If we look at the 40 nm parts until now, we see that RV740 and the announced Geforce shrinks have not much higher clocks compared to previous 55nm parts, if there is an improvement at all.
So I don't think we will see at launch something with clocks higher than RV790, i.e. (and I mean the base clock, not the OC parts). If we see the history R600 (80nm) was 743 Mhz, RV670 (55nm first iteration) was 750 Mhz and Rv770 (55nm second iteration but larger chip) was a 750 Mhz part, too. Cypress is rumored to be larger than Rv770, too, even considering the DX11 compliance. So I will bet on a 1600+ SP part.
Yes, I too for this exact reason expect Cypress' clocks to be in the 600-700mhz range.
 
I think we will see more than 1280 SP, because I don't think that ATI will go 900+ Mhz on the first parts. Process is still relatively new and high yields are needed that are unlikely to happen on an high clock part. If we look at the 40 nm parts until now, we see that RV740 and the announced Geforce shrinks have not much higher clocks compared to previous 55nm parts, if there is an improvement at all.
So I don't think we will see at launch something with clocks higher than RV790, i.e. (and I mean the base clock, not the OC parts). If we see the history R600 (80nm) was 743 Mhz, RV670 (55nm first iteration) was 750 Mhz and Rv770 (55nm second iteration but larger chip) was a 750 Mhz part, too. Cypress is rumored to be larger than Rv770, too, even considering the DX11 compliance. So I will bet on a 1600+ SP part.

I personally don't bet on anything just yet. Besides where's the major difference in theory in terms of yields if you have let's say a part with 1280@900MHz vs. a part with 1600@700MHz? What tells me that lesser complexity cannot be theoretically invested in higher frequencies or vice versa?

As long as we don't know how much additional die area the X11 requirements account for, how many TMUs, ROPs etc. the guessing game can go on forever.

Let's just take the TMU part of the speculative math; you get roughly similar texel fillrates whether you have 80@700MHz or 64@900MHz. First scenario though could mean that the TMUs there capture ~20% more die area. Now wouldn't those TMUs also need some additional logic for some of the X11 requirements?

Frankly I don't know how you guys make those equasions like it's a simple 1+1=2, but for me personally for the time being it's adequate to know that Cypress has a maximum theoretical arithmetic efficiency of over 2TFLOPs.
 
Back
Top