R650, 65 nanometre comes in Q3

A new hope at 800MHz+

Quote:

---

ATI is working hard to bring a new high end card and this one is scheduled for Q3. This is basically the R600 done right and is nothing much more than R600 shrinked to 65 nanometre.

ATI obviously mastered 65 nanometre and is ready for mainstream and entry level DX 10 at 65 nanometre production. The next step is to release the real high end card clocked at more than 800 MHz core and 1024 MB GDDR4 memory.

The next high end card is R650 and it should be out until September in the worst case scenario. Let's hope that things will go right this time. Some of our sources indicated way higher clocks than 800 MHz, closer to 1000 MHz.

---

Source

Now how can i go out and buy the R600 after reading this. Looks like my current card will have to do for some months more.

finally a card that can do 1000 mhz stock cooling.

65nm gpu ...wasn't expecting that this soon.

oooo nice...but wasn't this supposed to be out January of this year? And second what happens when the G90 comes out? :stick:

But this is from FUD. FUD hates ATi with a passion.

I really feel sorry for AMD/ATI. They just need to get REAL PRODUCT out in a timely manner. It feels like Intel and nVidia are eating their lunch in the meantime.

If this is true the only real advantage of 65nm is manufacturing cost, not necessarily better cooling or OC. Unless they revamp the arch some how to accommodate better cooling or OC'ing. If it's true they are doing this (taking with a grain of salt) making this nothing more then a die shrink, there is no tangible advantage for the consumer.

Quote:

---

Originally Posted by Eastcoasthandle

If this is true the only real advantage of 65nm is manufacturing cost, not necessarily better cooling or OC. Unless they revamp the arch some how to accommodate better cooling or OC'ing. If it's true they are doing this (taking with a grain of salt) making this nothing more then a die shrink, there is no tangible advantage for the consumer.

---

ehh

lower chip complexity/size = lower likelyhood of errors in manufacturing.

everytime there is a dieshrink, somebody wins and usually the consumer is one of the winners(think trickledown effect)

Kind funny XT2900XT just good in synthetic benchmarks like 3D06. I hope they will make tests on Vista rather than WinXP, as for 65nm maybe 2900XTX will flow like water. Well I think the only card to get benefit from 65nm process will be 2900XTX.

q3 of what year?

Quote:

---

Originally Posted by mion

q3 of what year?

---

I think thats not yet decided :rolleyes: :rolleyes:

Quote:

---

Originally Posted by ANP !!!

I think thats not yet decided :rolleyes: :rolleyes:

---

Q3 2010.:woot:

Quote:

---

Originally Posted by Eastcoasthandle

If this is true the only real advantage of 65nm is manufacturing cost, not necessarily better cooling or OC. Unless they revamp the arch some how to accommodate better cooling or OC'ing. If it's true they are doing this (taking with a grain of salt) making this nothing more then a die shrink, there is no tangible advantage for the consumer.

---

Smaller process size has several benefits, the most popular and common is more clock speed. Then of course you have trade offs. For example do you use the extra transistor budget to get more performance out or do you get better yields

Quote:

---

Originally Posted by nn_step

Smaller process size has several benefits, the most popular and common is more clock speed. Then of course you have trade offs. For example do you use the extra transistor budget to get more performance out or do you get better yields

---

Or more importantly:

LOWER POWER CONSUMPTION @ similar clock speeds.

Of course obviously the smaller die size usually translates to a higher clockspeed at similar voltage because there is less heat, but I think AMD will welcome the former.

It looks like the current 80nm R600 will overclock very well on the stock cooler and stock volts, and is underclocked to stay in the 2x6-pin power envelope of <225W. It seems to me overclocking is disabled with 2x6-pin because of this, but yet with >225w it soars way above the set ~750mhz default clock with the stock cooler, and even higher given proper cooling; but that's a given considering it has the leg-room to use up to 300; although granted we don't know how much is actually used with a high overclock and proper cooling. The point is, what's holding it back is the power usage AND heat.

The shrink to 65nm screams to me that power-suckage will drop dramatically, perhaps to around 175W or so, that this will no longer be an issue compared to current parts. Heat will obviously as well be less of a factor. It may clock better overall, but I imagine we'll see a default clock jump to below what the current 80nm chips are capable of just to keep, as you mentioned, yields high, but more importantly power consumption low(er). One can assume the default vGPU will be lower, with a nice trade-off between clocks and power usage. The situation may have turned out different if the R600 wasn't inherently a power hog, but it is, so this to me seems the only way AMD can go.

Hopefully the 225W wall (I'm thinking R650 will def be 2x6-pin or 1x8-pin) doesn't hurt their ability to clock the 65nm part higher at stock, or from overclockers with xtreme cooling from doing their thing past the default like the current (is 9 days still the future?) cards seem to be with the up to 300w allowance.

Most likely not before Q1 08...they won't need the R650 till then.

Perkam

Quote:

---

Originally Posted by perkam

Most likely not before Q1 08...they won't need the R650 till then.

Perkam

---

If the R650 is all they have for 1st/2nd quarter of 08(and they sit on it until then), then they're in deeper trouble than they were the past 7'ish months on the high end, however, AMD stated we'll see the refresh within a few months of the R600 in the press conference...

I've said it before and I'll say it again, there's still more big fish in the sea this year from both camps. Both are merely baiting the hook with the first run of the year, being the 2900XT and 8800Ultra... The amount of fish seen depends on each camps moves.

The battle is likely to run all the way until Q1/Q2 08.

H108 brings R700, and supposedly either 55nm or 45nm, depending on which way the wind is blowing.

R650 is a Q4 product at worst, I think late Q3 sounds right from what we've heard lately, although I expected RV670 before it, also in late-ish Q3.

True enough, R650 better have the clocks to compete with surely a 512-bit and/or highclocked and/or G80-type beast with more scalars.

Exactly like the X1800/X1900 "strategy".

Poor AMD, getting spit-roasted from both sides by Intel and nV atm :(

Quote:

---

Originally Posted by turtle

H108 brings R700, and supposedly either 55nm or 45nm, depending on which way the wind is blowing.

R650 is a Q4 product at worst, I think late Q3 sounds right from what we've heard lately, although I expected RV670 before it, also in late-ish Q3.

True enough, R650 better have the clocks to compete with surely a 512-bit and/or highclocked and/or G80-type beast with more scalars.

---

Sounds similar to the word going around here as well. The rest of the year should be rather interesting.

So what it will be facing..
Not much info about its competition?
Its been half year since 8800..

Interesting if true... But if it's slated for Q3 that means they officially state the R600 sucks donkeyballs and those who actually buy a R600 card will see their investment go down the drain in 6 months...

I'm sorry but I'm not going to buy a $500 GPU and see it devaluate by half in 6 months..

Quote:

---

Originally Posted by NightCrawler™

Interesting if true... But if it's slated for Q3 that means they officially state the R600 sucks donkeyballs and those who actually buy a R600 card will see their investment go down the drain in 6 months...

I'm sorry but I'm not going to buy a $500 GPU and see it devaluate by half in 6 months..

---

if nVidia had competition, the price of their cards would have gone down to around 50% of release value. Stop complaining. To have a GPU last 6 months at high price isnt bad for the high-end

80nm 900mhz seems avialble with stockcooling, 1ghz then its watercooling time.

For extreme clockers 80nm using cold and ln2 stuff the heat will not be an issue at all ;)

a FUDzilla is not a source of news. It is a source of speculations and most of them are BS.

Quote:

---

Originally Posted by gOJDO

a FUDzilla is not a source of news. It is a source of speculations and most of them are BS.

---

Well said my friend :clap:

Quote:

---

Originally Posted by NightCrawler™

Interesting if true... But if it's slated for Q3 that means they officially state the R600 sucks donkeyballs and those who actually buy a R600 card will see their investment go down the drain in 6 months...

I'm sorry but I'm not going to buy a $500 GPU and see it devaluate by half in 6 months..

---

I'm sorry, that's been the case as far as i can remember. That's why I personally never ever buy a brand new gpu.

Ie: I got X800Xt for about $65USD(this was when X1800XLs were getting launched), x1800XT for $170USD(when X1900s were just popping out)

Must also consider in my country is way more expensive than USA...

Quote:

---

Originally Posted by XeRo

But this is from FUD. FUD hates ATi with a passion.

---

QFT :toast:

AFAIK they didn't actually hate them, they used to be fanboys, but since their arguments with them they now do spread utter FUD.

Nothing bad in that they live up to the point WHY they opened the site and named it..

FUD :D

I expect it to be the XTX in 1GB GDDR4 @ 2200-2400 at 65nm with a few more subtle additions, like higher clock shaders and more TMU/ROPs. :woot:

Quote:

---

Originally Posted by KTE

QFT :toast:

AFAIK they didn't actually hate them, they used to be fanboys, but since their arguments with them they now do spread utter FUD.

Nothing bad in that they live up to the point WHY they opened the site and named it..

FUD :D

I expect it to be the XTX in 1GB GDDR4 @ 2200-2400 at 65nm with a few more subtle additions, like higher clock shaders and more TMU/ROPs. :woot:

---

FUD chases the money, that's the sole purpose of his articles...

Quote:

---

Originally Posted by GMX

FUD chases the money, that's the sole purpose of his articles...

---

We think alike. ;)

Actually we don't. It's just a fact. :D

Quote:

---

Originally Posted by Eastcoasthandle

If this is true the only real advantage of 65nm is manufacturing cost, not necessarily better cooling or OC.

---

Sorry, but what are you talking about?
It even said that it will have higher clocks. Probably it will also use less power.

Hmm, Q3 they say.... Expect it in Q1 2008.

Quote:

---

Originally Posted by |JACK-

That's life no?
remember the X1800? that was a pure rip off :slapass:

---

Depends..... my 7800GTX had a long life span.... I skipped the X1800 cause I knew the X1900 was around the corner...
Sometimes the internet news helps you decide... even if it is from Fudzilla...

If you take the G80 and how long it's on the market and will be until the next one comes out it's more and longer value for the money...
Hell you don't see nVidia making news before they released the G80 that the G90 was going to make it's appearance 6 months later.. bla bla bla...
This is just stupid marketing.....

sounds nice, but it will probably have to compete with nvidias next gen architecture which might be a problem

g90 will be to the g80 what g70 was to nv40, essentially the same architecture, just a bigger and wider implementation with a few new tricks thrown in. I'd expect a die shrink g81 first though, unless a 65nm g80 design is what the g96 is going to be.

Quote:

---

Originally Posted by flopper

80nm 900mhz seems avialble with stockcooling, 1ghz then its watercooling time.

For extreme clockers 80nm using cold and ln2 stuff the heat will not be an issue at all ;)

---

Heat wouldn't be an issue but available power might, there's only so much they can get from PCIe & 2x3 + 2x4 pin connectors.

Quote:

---

Originally Posted by saaya

sounds nice, but it will probably have to compete with nvidias next gen architecture which might be a problem

---

I would say it will most definitely be a problem. It seems the Radeon cycle is now perma-behind Nvidia. If they launch a next-gen part at the end of the year to compete with the R600 refresh, no doubt a refresh will come shortly after R700 (at 55, if not 45nm), and the cycle continues.

Quote:

---

Originally Posted by OWA

g90 will be to the g80 what g70 was to nv40, essentially the same architecture, just a bigger and wider implementation with a few new tricks thrown in. I'd expect a die shrink g81 first though, unless a 65nm g80 design is what the g96 is going to be.

---

I bet Nvidia skips 80nm and goes straight to 65nm. They been ordering test waffers for some time (late last year iirc), and there was some reports out of TSMC that nvidia wanted assurance they could handle a top-end switch this early to 65nm. That being said, who knows if there will be a G80 straight-shrink (G81) before they release a part with added shaders etc (G90). That still seems up in the air, as far as public domain information goes. If I were a betting man, I would think they would soup G80 up while shrinking it, essentially becoming G90. A 512-bit bus, and perhaps 160 dual-issue scalars + MUL (granted, if rumours are true, G80 may already contain 160 scalars). That seems the most probable to me, as it would tie in nicely with G80's architecture, essentially increasing it by 20-25&#37; architecturally speaking, plus whatever die size/clock advantages the shrink would yield.

Quote:

---

Originally Posted by Death Reborn

Heat wouldn't be an issue but available power might, there's only so much they can get from PCIe & 2x3 + 2x4 pin connectors.

---

I would agree with that statement if you said 2x 2x3 connectors. Remember they HAVE to make R600 work with 225w, and that seems to cripple it. With the up to 300W available from the 2x3 + 2x4 connectors (plus pci-e slot), it appears to perform just fine...but I don't think AMD wanted to take the step to REQUIRE the 8-pin connector. Hopefully they can get clocks to a reasonable spot within that power envelope (225W) for the refresh. I reckon they can, but ponder if it will be enough for whatever nvidia has cooking. I am thinking the negative.