[Arty]

Originally Posted by Florin

256-bit memory in that price class. The X1600 may be up for an uphill battle just like the X1300 now looks to be.

6800GS looks more like a response to GTO/GTO2 rather than X1600 considering its not even out yet ..

[trinibwoy]

Originally Posted by Dave Baumann

As for "8", if its the configuration that I would expect then it would be handling 24 fragements per clock.

I'm a little confused by this language. When we're talking about fragments per clock, if RV530 is 24/clk, I assume that makes R580 48/clk. Now is there any comparison to G70 architecture here? If there is, does that make G70 24/clk or 48/clk?

Or in other words, is G70 24-1-2-2 ?

[Dave Baumann]

G70 is 24 fragements per clock (In those terms its 16-1.5-1.5-2).

[Arty]

Originally Posted by trinibwoy

I'm a little confused by this language. When we're talking about fragments per clock, if RV530 is 24/clk, I assume that makes R580 48/clk.

24 frag/clk if RV560's config is 8-1-3-2/1.

[trinibwoy]

Ah ok thx, just wanted to clear up (in my own mind) that the x3 multiplier isn't simply referring to the number of ALU's working on each pixel in flight.

So R580 can have 48 different pixels in flight, one ALU each. G70 is 24 in flight, 2 ALU's each?

[Dave Baumann]

To elaborate on that at little - a fragment is the OpenGL term for a unit of work in the pixel shader - its termed as a fragment work is often carried out that is less that an pixel (i.e. ifyou are rendering to a texture the original "fragment" processed ould be greater or smaller than a "pixel", if you are supersampling each pixel is actually a sub-pixel of a resultant pixel, etc.)

A fragment shader for G70 is represented as:


A fragment shader for R5xx is represented as (the black area, less the texture elements):


Each of these are handling a single fragment, but they each have different capabilities and will spit the fragment out in different numbers of cycles dependant on the composition and how its requirements meets the capabilities of the fragment pipeline.

[Matasar]

The cards seem to be Limited Edition..

[_xxx_]

Originally Posted by Matasar

The cards seem to be Limited Edition..

Because of the red PCB? Is the "regular edition" blue?

[Matasar]

No Leadteks 512meg GTX is sold as Limited Edition.
Havent seen any other brand yet.

[Cowboy X]

Originally Posted by Chalnoth

Well, consider that's a 50% increase in memory bandwidth and a 20% increase in core speed, so the performance increase should vary between 20% and 50%, depending upon the limitations of the program being benchmarked (assuming the program isn't limited by geometry or CPU...so in other words, at high res). I'd say most of the time it'll be on the lower range of that, given that the 7800 GTX isn't hugely bandwidth limited.

You cannot discount the fuzzy logic that is applied to Nvidia's clock speeds . Is 550 really 550 ........ or is it 550 on 1 part of the chip and much lower speeds elsewhere ? For that reason raw clocks on Nvidia's current products don't mean much without seeing the benchmarks .

[trinibwoy]

Originally Posted by Cowboy X

You cannot discount the fuzzy logic that is applied to Nvidia's clock speeds . Is 550 really 550 ........ or is it 550 on 1 part of the chip and much lower speeds elsewhere ? For that reason raw clocks on Nvidia's current products don't mean much without seeing the benchmarks .

Not really since we pretty much know that the bottleneck will be in the TMU's/pixel-shaders and not the vertex shaders so we dont really care what the rest of the chip is clocked to.

[_xxx_]

Oh, I thought you were talking about 1800XT

That's probably just some fancy extra cooler or ViVo or some such making it into Lim.Ed.

[Arty]

Originally Posted by Matasar

No Leadteks 512meg GTX is sold as Limited Edition.
Havent seen any other brand yet.

EVGA 7800 GTX512 Black Pearl KO Edition with waterblock cooler, wonder what the clocks are going to be and price.

[Matasar]

LeadTek Geforce 7800GTX Extreme, 512MB DDR3, TV-Out, 2*DVI, HDTV, VIVO, 16X PCIe [Limited Edition]
Isnt all GTX VIVO ? And im pretty sure they all stick with the same Quadro cooler.
Where did you find the EVGA listed ?

[Ailuros]

Originally Posted by Cowboy X

You cannot discount the fuzzy logic that is applied to Nvidia's clock speeds . Is 550 really 550 ........ or is it 550 on 1 part of the chip and much lower speeds elsewhere ? For that reason raw clocks on Nvidia's current products don't mean much without seeing the benchmarks .

You have on a 7800GTX 430MHz for the SIMD channnels and ROPs and 470MHz for the VS units (whereby for some weird reason NVIDIA claims "only" 860 MVertices/s; theoretically the math would be 470*8 = 3760 / 4 = 940 MVertices/s).

Just because the geometry processors have a higher frequency it doesn't mean that the G70's have unpredictable performance. Raise both core and ram frequency on a standard G70 by say 15% and you will get roughly 15% more performance than on default. If the increase is at 30%, it's 30% more performance more or less on average. Where's the unpredictable factor exactly?

[SugarCoat]

Originally Posted by Ailuros

You have on a 7800GTX 430MHz for the SIMD channnels and ROPs and 470MHz for the VS units (whereby for some weird reason NVIDIA claims "only" 860 MVertices/s; theoretically the math would be 470*8 = 3760 / 4 = 940 MVertices/s).

Just because the geometry processors have a higher frequency it doesn't mean that the G70's have unpredictable performance. Raise both core and ram frequency on a standard G70 by say 15% and you will get roughly 15% more performance than on default. If the increase is at 30%, it's 30% more performance more or less on average. Where's the unpredictable factor exactly?

im wondering how far that 1:1 performance increase goes before it trickles down. And what effect the memory will have.

Lets say 430 to 550, 28% increase about, but the memory, from reference its going up 50%, or so its said, i expect it to be underclocked by 200MHz on a reference myself, rated 1800 but at 1600. Anyway, lets say best case, 50% increase on the memory.

What will that translate into when real world performance is measured, is my curiosity. Will it be close to 40%? Or merely ~25%. How much improvment will the memory give in the most taxing of bandwidth situations in todays titles and benchmarks? Can the core make use of it all? Can games? Etc...

The highest clocked GTX appears to be using 460 cores and 1300MHz. Everything else that was higher, is discontinued. All the AIBs have tried to make a gap between the $450-500GTX and this new card. That was very intentional, now all that remains to be seen is if it was done for sales or to remove the threat of 2 competing products from the same company.

[Hellbinder]

werent the rumors stating that today is the day?

[Rys]

Today is 6800 GS day. Fire up your browser in a week or so for the other one.

[RobertR1]

http://www.zipzoomfly.com/jsp/Produc...uctCode=322746

Check out the manufacturer spec. pdf. Only 470mhz clock? Ram is nice at 1700 though. Looks like Evga will be milking the "KO" badge again. Probably a 570/1780ish card with a $650+ price tag.

[CMAN]

Wow, the core is very dissappointing if that is correct. 570 MHz would sound better for the speed of the RAM.

[SugarCoat]

looks like the INQ was wrong again, i really doubt eVGA's base model is any lower then anyone elses. 550 seemed way too high to be in decent yields.

[Matasar]

470 for "basic" card i guess. KO for insane clock.

[trinibwoy]

Originally Posted by SugarCoat

looks like the INQ was wrong again, i really doubt eVGA's base model is any lower then anyone elses. 550 seemed way too high to be in decent yields.

Hmmmm that doesnt gel with a lot of hints being dropped about higher-than-expected clocks. 470 would be extremely disappointing based on leaked info on higher voltage and tweaked process.

[RobertR1]

The link has been pulled.

[no-X]

The question is why...

edit: didn't Dave say, that ATi will not need XT-PE? 470/1700MHz GTX could be the reason why.