R700 is 55 nanometre (Fudzilla)

[Ubermann]

Our knowledgeable sources also confirmed that R700 is a multi core design and DAAMIT calls it Multi Core Unified Architecture. This sounds like a couple of cores stitched together in a high end chip and a single core for a small chip.

Read more at Fudzilla (http://www.fudzilla.com/index.php?option=com_content&task=view&id=1028&Itemid=1)

[xoqolatl]

55nm? I thought next transition would be 45 nm like in CPUs, but we have seen 80nm in GPUs and not in CPUs already... But why would AMD, having high-k 45nm technology, make 55nm GPUs?

[Ubermann]

Cuz they dont produce GPUs ?

[LOE]

R700 is 55 nm?

I think what they wanted to say is "will be"

Anyway... shrinking the R600 to 65nm and then jumping on new architecture and a new process doesn't sound like a good idea

they better make a 55nm R600 or a 65nm R700 before they do something that risky

[deathman20]

nVidia was talking about the 55nm. AMD/ATI was talking between the 45 and 55nm dies. Still way to far to know anything yet since its still at least 8 months off.

[Shintai]

Fudzilllaaa. I doubt he got the slighest clue.

Why do people even link to him anymore...

And I would bet its 65nm, unless the release is in the middle/end of 2008.

[largon]

55nm is a half node process from 65nm meaning it's technically more closer to 65nm than 45nm thus it's much easier to make happen.

ATi has a record of shrinking performance parts with ½node, followed by the mainstream chips with a complete shrink and only after that will next gen performance chips get a complete shrink.

[gOJDO]

Ubermann, here is graph updated:
http://img218.imageshack.us/img218/6426/bsmeterxj8.gif

[freecableguy]

AMD better release something good soon or there may not be a R700...they need to stop worrying about next-generation and get quad-core out the door before they are gone...

[STaRGaZeR]

Damn, those BS meters rlz :rofl:

[XS Janus]

Damn... who gave a star to this thread!? FuAd himself!?

[XeRo]

FuAd himself!?

Yes, I did. I am here at XS to spread FUD.

[Ubermann]

Fuad or not..its something to read.

[nn_step]

Ubermann, here is graph updated:
http://img218.imageshack.us/img218/6426/bsmeterxj8.gif

Oh ye of little faith :rolleyes:

[Ubermann]

It sounds a bit wicked that they already gonna leave "regular" gpus..
Theres problaly alot to improve in current one to make it fast enough.

BS ver.2 nice one!

[Noobie]

NVIDIA's G80 has been out for some time now. G90 is supposedly skipped (it's just names anyway), and the G100 (G90 if you wanna call it that) will come out Q1 2008. G100/G90 is a 65nm DX10.1 part. That sounds not so unreasonable to me. 65nm is the next logical step, NVIDIA has traded info with Intel recently (who already is at 65nm), ATI is supposedly on 65nm soon. DX10.1 isn't much of a big difference either; many features are already on the HD2900XT. So my guess is that we'll have a G100/G90 by Q1 2008 from NVIDIA.

Now ATI has a huge problem. They can not keep up with NVIDIA. Similar to Intel not being able to keep up with AMD for a while. What did Intel do? Put 2 core's on a CPU, and called it dual-core. They even admitted it was a rush job and not much of a real dual-core.

2 GPU core's on 1 card has been done for quite some time now. ATI's CF implementation is verrrrrrry good. If ATI fears NVIDIA they can just put 2 GPU cores on 1 card.

Intel has already shown that once you make the initial step of going from single-core to dual-core properly, quad-core or more isn't that hard.

Why is this so hard to believe then?

Here's something else to ponder:
ATI give's info to sites
Sites give info to us
ATI can't do what they promise
We blame sites, instead of ATI

[Haltech]

NVIDIA's G80 has been out for some time now. G90 is supposedly skipped (it's just names anyway), and the G100 (G90 if you wanna call it that) will come out Q1 2008. G100/G90 is a 65nm DX10.1 part. That sounds not so unreasonable to me. 65nm is the next logical step, NVIDIA has traded info with Intel recently (who already is at 65nm), ATI is supposedly on 65nm soon. DX10.1 isn't much of a big difference either; many features are already on the HD2900XT. So my guess is that we'll have a G100/G90 by Q1 2008 from NVIDIA.

Now ATI has a huge problem. They can not keep up with NVIDIA. Similar to Intel not being able to keep up with AMD for a while. What did Intel do? Put 2 core's on a CPU, and called it dual-core. They even admitted it was a rush job and not much of a real dual-core.

2 GPU core's on 1 card has been done for quite some time now. ATI's CF implementation is verrrrrrry good. If ATI fears NVIDIA they can just put 2 GPU cores on 1 card.

Intel has already shown that once you make the initial step of going from single-core to dual-core properly, quad-core or more isn't that hard.

Why is this so hard to believe then?

Here's something else to ponder:
ATI give's info to sites
Sites give info to us
ATI can't do what they promise
We blame sites, instead of ATI

Insiders from ATI give sites information.. NDA information at that, so... how can you really blame ATI?

ATI was left better by themselves instead of AMD buying them, esp in the middle of GPU war and product release.

AMD was stupid for making the huge purchase of a graphics card company that didnt even have 100% working silicone, while in a middle of a CPU war with Intel.

This is almost as good as the presidential canidates now.:ROTF:

[Verisimilitude]

But more importantly, if AMD fails we all lose big time.

[7he ]-[0rr0r]

I am sorry to say i feel that chart is pretty accurate after the r600 being hyped up so much and being that late then still failing to dethrone nvidia.
I really feel that i should have seen that coming: that the increasing heat and power consumption and electrial issues were a sign that they were having great difficulties getting the performance to where they wanted it.
It seems to me the old amd would have set more appropriate expectations in this situation I really hope things turn around for them now that intel is not so lethargic.

[savantu]

But more importantly, if AMD fails we all lose big time.

Or maybe that's what we need to fire up the whole industry.

Without AMD , we can have Microsoft/IBM/Nvidia doing CPUs.Imagine the wars...:devil:

[thephenom]

It make sense, considering how many transistors we find in the G80 and R600, it's probably hard to get a VERY good yield comparing to what you get out of a G84 or RV630.

So the next logical step to get yields up and still improve performance would be stick 2 of cores together and call it a high end. If this is a viable solution to solve yield problems, perhaps we'll finally see the end of a $500 high-end card. :D However, the biggest problem we'll see here would be CrossFire and SLI. If one high-end card already has 2 cores (in CF or SLI on their own), by adding a second card, you'll have 4 GPUs in 2 Cards. The last time I checked, Quad GPUs scaling was still horrendous (ie 7950), not sure if this is still the case with Vista, but I can imagine it to be the same case since 4-way AFR is far from being efficient. I guess you can Scissors the screen for 1 card (2GPU) and than AFR it, but you still won't see much scaling since AFR is the only mode in CF and SLI that really scales well.

[nn_step]

Or maybe that's what we need to fire up the whole industry.

Without AMD , we can have Microsoft/IBM/Nvidia doing CPUs.Imagine the wars...:devil:

as much as I would love to see PowerPC finally replace x86. Imagine what it would do to the desktop market. There would be absolute chaos in the desktop market and the people who would have the edge would be those with the best virtualization performance.

[madcho]

half node use too much power, it's the ATI's way but it's not the good one :/

do it in 65nm and go down in 45nm after.

[Bonkers]

I'm thinking r700 will be a fuad-core design..

[largon]

half node use too much power, it's the ATI's way but it's not the good one :/

do it in 65nm and go down in 45nm after.:eh:
55nm being a "half node" is nothing but an optical shrink of the process that allows lower power consumption without most of the problems usually related to making a chip compatible with a shrink like being forced to retrofit the design for the smaller process.

55nm is better than 65nm.

[perkam]

Ubermann, here is graph updated:
http://img218.imageshack.us/img218/6426/bsmeterxj8.gifSome of us are saving up for that green bs :stick:

Perkam

[[XC] gomeler]

I don't know of this is possible but could ATi perhaps create a single simpler GPU with performance equivalent to ~8600GT and then do something like AMD/Intel's dual-core dual-die designs? Design the entire architecture so that it can scale to multiple dies on a single package. Don't know if this is a viable solution or if there is an obvious problem, but what I see is a huge simplificiation of their design process, focus on 1 core, scale number of cores to scale performance. Could even make it possible to have a "quad-core" GPU that'll dynamically shutdown cores when they aren't needed. Makes no sense for my GPU to be sucking tons of watts when it's just displaying a windows GUI.

[Verisimilitude]

Just found this as a HardOCP link... wow is all I have to say

http://weblog.infoworld.com/yager/archives/2007/05/intel_fud_versu.html

Title is Intel FUD versus AMD fact

[thephenom]

gomeler;2196429']Makes no sense for my GPU to be sucking tons of watts when it's just displaying a windows GUI.
Does SLI and CF power down one of the cards when it's idling? If it doesn't, I don't expect it to be a feature shutdown core. However, with 2 cores on 1 board, now they can probably lower the 2D clock even more than say it's just 1 core on 1 PCB.

I really don't mind this concept as long as the cost savings are past on to the consumers like us. :)

[ownage]

GUIDE TO STUP THE FUD!!
Dont let you be brainwashed by the FUD!!

Start notepad and open C:\windows\system32\drivers\etc\HOSTS
http://img511.imageshack.us/img511/9986/stopthefudix8.th.jpg (http://img511.imageshack.us/my.php?image=stopthefudix8.jpg)
type 127.0.0.1 fudzilla.com

[billdavis]

ubermann. dude knowing that it is mostly dudes on this fourm and you put a picture of your self for us dudes to look at is weird. it is great and all if you love your self but id rather not look at a dude if you know what i mean

[Shintai]

gomeler;2196429']I don't know of this is possible but could ATi perhaps create a single simpler GPU with performance equivalent to ~8600GT and then do something like AMD/Intel's dual-core dual-die designs? Design the entire architecture so that it can scale to multiple dies on a single package. Don't know if this is a viable solution or if there is an obvious problem, but what I see is a huge simplificiation of their design process, focus on 1 core, scale number of cores to scale performance. Could even make it possible to have a "quad-core" GPU that'll dynamically shutdown cores when they aren't needed. Makes no sense for my GPU to be sucking tons of watts when it's just displaying a windows GUI.

The problem is compability etc. Currently games need to be made for SLI for it to work. I dont know about CF. So there is a possible application problem atleast. Then the next problem is speed and BW between the cores and the PCIe sharing.

Also quadcore GPU, well..we already got that. 1pipe=1core roughly. So your 700M R600 would just end up as something like 4x200-225million due to overhead for interconnection and god knows what. So you end up with a higher cost setup with equal or lower performance. Another option is 2 700M chips+overhead operating at 50% speed to save power (I guess 2 of those could be within 125W combined), but you would have to pay for 2 chips...

[largon]

Shintai, thephenom,
What does Crossfire have to do with a multicore GPU? However, with 2 cores on 1 board, now they can probably lower the 2D clock even more than say it's just 1 core on 1 PCB.Dual GPU (2+ die packages) makes no sense whatsoever compared to a multicore solution. Dual GPU only has skyrocketing PCB complexity, performance losses and inefficiency to offer, clustercore OTOH would be the solution to the problems that are inevitable with conventional singlecore GPUs.

I think something like what [XC] gomeler described will be reality within a year.

[Ubermann]

I think the original source for this news is www.mydrivers.com

ubermann. dude knowing that it is mostly dudes on this fourm and you put a picture of your self for us dudes to look at is weird. it is great and all if you love your self but id rather not look at a dude if you know what i mean

Sorry but please stop sending me all those kinky messages. I wont do it for a million bux.

[turbox997]

As mentioned before Fudzilla, the guy(so called journalist) pulled it out of his butt to get a reaction.

[gOJDO]

Some of us are saving up for that green bs :stick:

Perkam

La Rumba De Barcelona (http://www.youtube.com/watch?v=KPdrRvrkTSk)

[thephenom]

Shintai, thephenom,
What does Crossfire have to do with a multicore GPU? Dual GPU (2+ die packages) makes no sense whatsoever compared to a multicore solution. Dual GPU only has skyrocketing PCB complexity, performance losses and inefficiency to offer, clustercore OTOH would be the solution to the problems that are inevitable with conventional singlecore GPUs.

I think something like what [XC] gomeler described will be reality within a year.
For simplicity of design for NV and ATI, it makes more economical sense for them to begin with 2 physical cores, ie what you see on a 7950 and the dual 1950 pro that Sapphire and Asus sampled. In the long run, you might see ATI/NV do a more traditional dual-core design on 1-die, but if R700 is going to be "dual-core" and is due within a yr, then it's MOST likely to be multiple physical cores presense on the pcb.

[Noobie]

ubermann. dude knowing that it is mostly dudes on this fourm and you put a picture of your self for us dudes to look at is weird. it is great and all if you love your self but id rather not look at a dude if you know what i mean
What's the story of that avatar? A painting of some lady knighting some dude? What's the painting's name & whose the painter?

For simplicity of design for NV and ATI, it makes more economical sense for them to begin with 2 physical cores, ie what you see on a 7950 and the dual 1950 pro that Sapphire and Asus sampled. In the long run, you might see ATI/NV do a more traditional dual-core design on 1-die, but if R700 is going to be "dual-core" and is due within a yr, then it's MOST likely to be multiple physical cores presense on the pcb.
Yup, it's only a temp meausure for ATI to skip a design cycle, same as what Intel did.

If they are going to do a 2-GPU-on-1-die chip, it would mean that a 1-GPU is a large bottleneck compared to the mem-speed. I don't see the mem-speed increasing drastically anytime soon, so I guess it won't happen anytime soon.

[thephenom]

What's the story of that avatar? A painting of some lady knighting some dude? What's the painting's name & whose the painter?


Yup, it's only a temp meausure for ATI to skip a design cycle, same as what Intel did.

If they are going to do a 2-GPU-on-1-die chip, it would mean that a 1-GPU is a large bottleneck compared to the mem-speed. I don't see the mem-speed increasing drastically anytime soon, so I guess it won't happen anytime soon.
R600 itself might not be that much of a competitor, but it's mem bandwidth available is nothing to skip over, and that's with GDDR3. GDDR4 will probably be much more affordable and faster when R700-series takes off, and we might also see GDDR5 in action since Samsung already has it.