Real R520 Specs???

[Cota]

remember that neither nvidia nor ATI have reached the rumored clockspeeds of their recent releases.

I recall the 6800ultra was rumored to have 600/1.2, it never happened.

[SH64]

I as well think the mentioned specs could only deliver double the performance not triple!
also how a 24 pixel pipelines would have 32 textuering units ? .. maybe that guy need to elaborate this .
the core & mem clocks are too high as well & beyond belief .. well i know that ATi actually did 540Mhz in the X850XT-PE & 520MHz in the X800XT-PE/X850XT but they barly managed it so 700Mhz would be pushing it esp with a whole new manufacturing process .

[bkswaney]

Does TSMC offer low k on 90nm? can anyone confirm that?

No... not the last time I looked.
They were moving to 65.

[bkswaney]

I as well think the mentioned specs could only deliver double the performance not triple!
also how a 24 pixel pipelines would have 32 textuering units ? .. maybe that guy need to elaborate this .
the core & mem clocks are too high as well & beyond belief .. well i know that ATi actually did 540Mhz in the X850XT-PE & 520MHz in the X800XT-PE/X850XT but they barly managed it so 700Mhz would be pushing it esp with a whole new manufacturing process .


With -0 temps they might reach 700. :rofl:

[MUYA]

That Intrinsic or Trinsic thing probably won't help either attain such speeds for such a transistor beast.

[MUYA]

No... not the last time I looked.
They were moving to 65.
thought so!

With 110nm and low K and other techniques to cut current leaks may allow higher clocks? As Ati have been using TSMC's 110nm I think for some time now

There is a good article I think at Arstechnica which states why lower manufactruring processes introduce such leakage.

[Fotis]

AFAIK TSMC offers 90nm low-k only.No low-k for 110nm.

[MUYA]

OK so who is right? Bk or Fotis? TSMC offers 90nm with low k or not?? provide a link pls

[bkswaney]

OK so who is right? Bk or Fotis? TSMC offers 90nm with low k or not?? provide a link pls


http://www.tsmc.com/english/promotions/1122_04_brochure/b/b.htm

Here is some info.
Also I was reading somewhere that
that they were already making .65 and
having great luck with it.
This is all I could find though.

[Fotis]

beyond3d (http://www.beyond3d.com/forum/viewtopic.php?t=17171)

Quote:
" NVIDIA’s reluctance to use TSMC’s 130nm low-k may be causing them to be a little leery of TSMC’s 90nm node as it is currently only offered with low-k dielectric materials."

TSMC (http://www.tsmc.com/english/b_technology/b01_platform/b010101_90nm.htm)

Quote:
"The 90-nm process technology features:
Core supply voltage ranging from 1.0V to 1.2V
I/O and analog blocks ranging from 1.8V to 3.3V
Multiple threshold voltage (Vt) option for optimized transistor speed and power consumption trade-offs
Extremely tight process control for 50-nanometer gate length - the high speed process
Ni-salicide for better sheet resistance (Rs) in narrow line widths
Nine-layer copper interconnect, with an extra redistribution layer optional for flip-chip package
Low-k dielectrics with k less than 2.9 for the lowest RC delay and power consumption "

[Vagrant Zero]

remember that neither nvidia nor ATI have reached the rumored clockspeeds of their recent releases.

I recall the 6800ultra was rumored to have 600/1.2, it never happened.

475/1200.

Those were the rumours.

[bkswaney]

With no low-k.11 nvidia is just screwed in that department.

I would love to see a .13nm 6900U done in low-k.
If nvidia stays with .13 for a while on there top end cards
they could crank up the Mhz. ;)

But I bet they will go .11 and set them up for higher IPC.
Kinda like AMD has done. :)

[nutball]

:lol:

I love these Chinese-whispers things. Some geek dreams up a few numbers, and a few forum posts later they're the rumoured specs of the next-gen product, and people take them seriously

:ass: :smoking2:

[MUYA]

Who is taking them seriously here? I think most replies have shown the doubt in them "rumpured specs".

This is the Hardware Rumour Mills where everything has to be taken with a pinch of salt.

It's fun speculating!

[Fotis]

So anyone knows when ATI will launch R520? :afro2:
I have concrete evidence from the Inquirer that it will be in Cebit :D

[dan2097]

I as well think the mentioned specs could only deliver double the performance not triple!
also how a 24 pixel pipelines would have 32 textuering units ? .. maybe that guy need to elaborate this .
the core & mem clocks are too high as well & beyond belief .. well i know that ATi actually did 540Mhz in the X850XT-PE & 520MHz in the X800XT-PE/X850XT but they barly managed it so 700Mhz would be pushing it esp with a whole new manufacturing process .

From the 9800XT to X800XT PE they jumped from 412mhz to 520mhz going from 0.15 micron low k to 0.13 micron low k

From 540mhz (X850XT PE) adding on the same percentage gets you to 680mhz, 700mhz doesnt seem that unfeasible for the PE part.

0.09 micron low k theoretically is a bigger jump than 0.15 to 0.13 micron but thats probably more in adding more transistors than in ramping speeds I think

[MUYA]

That is not the point, Low K helps but doesn't eradicate the problem of current leakage all together. Intel have had a hard time, so have AMD. It's a given at lower processes the leakage problem will hamper you as you try to clock it insanely. I think 700+mhz is acheivable through special cooling like phase change. I seriously doubt that on 350+ million transistor die which will be very power hungry ..partly due to the leakage. Just a thought...not qualified in anyway of course :D

[jbirney]

Isn't fun to talk about rummored specs :)

I agree some of those are do-able...while others are a bit more far fetched.

[MUYA]

Agreed, 6 quad GPU, 8 VS units, SM3.0 etc are all viable...maybe even new AA unit etc. 96 ALUs...four per each plausible, but the 700mhz+ and 32 texture units (doesn't make mathematical sense) or 3X performance over 850XT are doubtfull, erm PPP and teselator or soemthing? I dunno about that at all..literally.

[RpPRO]

You throw in a thread like this from time to time and - surprise - everyone in interested. Users visit the thread more often and make more traffic and it's good for hardware sites to pay the bills.

Sweet dreams. :)

[SH64]

From the 9800XT to X800XT PE they jumped from 412mhz to 520mhz going from 0.15 micron low k to 0.13 micron low k

From 540mhz (X850XT PE) adding on the same percentage gets you to 680mhz, 700mhz doesnt seem that unfeasible for the PE part.

0.09 micron low k theoretically is a bigger jump than 0.15 to 0.13 micron but thats probably more in adding more transistors than in ramping speeds I think

Remember the X850XT-PE is a refresh product (with a whole new cooling solution) while the R520 will represent a new gener . so we are actually talking about jumping from 520Mhz to 700Mhz which is too much especially when you are experiancing a whole new manufacturing prcoess . if i would speculate for a proper high end core speed i'd say no more than 600Mhz . a refresh would go higher but that will take some time .

[holmes]

If ATI tries to go for anything near 700MHz and go to .09 I think they'll have a much worse time trying to make their cards available than the X800XT-PE. Somethings gotta give. Specs like that are too risky at the moment. It could be done, but it would cost them more money to expend the effort than they would get a return on. My guess is that if they go for a 24 pipe card, the core will clock at no more than 500MHz. Besides, I thought the whole point of increasing parallelism (e.g. SLI) was to address the shortcomings of what is feasible at the moment...

[JoKeRr]

If ATI tries to go for anything near 700MHz and go to .09 I think they'll have a much worse time trying to make their cards available than the X800XT-PE. Somethings gotta give. Specs like that are too risky at the moment. It could be done, but it would cost them more money to expend the effort than they would get a return on. My guess is that if they go for a 24 pipe card, the core will clock at no more than 500MHz. Besides, I thought the whole point of increasing parallelism (e.g. SLI) was to address the shortcomings of what is feasible at the moment...

That 700 mhz core should be an expectation, but often expectations doesn't meet the actual product. 700mhz will require a fair amount of voltage--> major exponential leak of current-->what, Tri-dust buster V8 + 150W of psu only for graphic card?? Yield will go nuts, so is the price and profit will plumb like mashed potato.

[dan2097]

If ATI tries to go for anything near 700MHz and go to .09 I think they'll have a much worse time trying to make their cards available than the X800XT-PE.

Personally I think if they went for 700mhz on 0.13 micron low k with a 250+ million transistor core that they would have even bigger problems then ;)

Remember the X850XT-PE is a refresh product (with a whole new cooling solution) while the R520 will represent a new gener . so we are actually talking about jumping from 520Mhz to 700Mhz which is too much especially when you are experiancing a whole new manufacturing prcoess . if i would speculate for a proper high end core speed i'd say no more than 600Mhz . a refresh would go higher but that will take some time .

ATI and Nvidia both managed to hit 500mhz on 0.13 micron. Well okay Nvidia didnt with a conventional cooler, but they managed 450-475mhz later with a more usual cooler.

ATIs 0.15 micron card, the 9800 pro which admitedly was more complex clocked at 380mhz, the geforce 4 ti which is alot less complex than the gffx only reached 300mhz.

Those are huge percentage increases from a relatively small process size reduction.

I mean did people expect the fx 5800 ultra to debut at say 375-400mhz on 0.13 micron?

[holmes]

Obviously you can hit higher clock speeds by going to a smaller transistor, that's not the point. Going to a new process on your top end card is rarely a smart move. The possible benefits are greater, but the risk is more so. Kinda like gambling... As was pointed out earlier, leakage gets exponentially harder to deal with at .09 and beyond.... If they do manage to hit specs similar to these I'd be surprised if they can get yields high enough to match the X800XT-PEs availability, which was basically non-existant for 6 months+...