The Official RV630/RV610 Rumours & Speculation Thread

Status
Not open for further replies.
The 8500 GT is based on the cheaper G86 core, not on the G84 (used in the 8600 GT and 8600 GTS).
Yup, I know that.

G86 being 128b has an theoritical advantage over RV610, if thats the score of a 64b RV610LE then AMD could possibly be not that far behind in the low-end segment.
 
Yup, I know that.

G86 being 128b has an theoritical advantage over RV610, if thats the score of a 64b RV610LE then AMD could possibly be not that far behind in the low-end segment.

This is what we know about both low end cores right now (yes, they're not direct equivalents).
The same CPU was used in both tests, apparently (Core 2 X6800).
Also, both cards probably have very early driver support, so it's difficult to ascertain much else from the little data provided.


RV610 LE
VR-Zone said:
at 1280x1024 resolution, 3DMark06 scores stood at 12xx

G86-300 (8500 GT)
VR-Zone said:
Its 3DMark06 scores (1280x1024) stood at 22xx


It would be interesting indeed to know more about 8400 GS and 8300 GS.
Also, are the "8200 GS" and "8100 GS" monikers reserved for ultra-low end cards, IGP's or specific mobile solutions ?
And is it really a RV610 "LE", or a standard RV610 ?
Who will have the cost advantage ?
A small 65nm core (AMD), or a cheaper PCB (this Nvidia 8500 GT looks like it's using a 7600 GS' PCB, just like the 8600 GT's PCB is a dead ringer for the old 7600 GT one).
And is a RV610 really smaller at 65nm than a G86 at 80nm, in terms of die size and power consumption ?


Too many questions, only a month or two to go...
 
Last edited by a moderator:
A small 65nm core (AMD), or a cheaper PCB (this Nvidia 8500 GT looks like it's using a 7600 GS' PCB, just like the 8600 GT's PCB is a dead ringer for the old 7600 GT one).
How do you know this is an or case? Were there images of this "RV610" board that a comparison can be made? If there are memory bus differences between them, wouldn't that also have some sway on the potential PCB costs as well?
 
How do you know this is an or case? Were there images of this "RV610" board that a comparison can be made? If there are memory bus differences between them, wouldn't that also have some sway on the potential PCB costs as well?

What makes you say that other editions of G86 (like the 8400 GS and 8300 GS) don't have a further "LE" version with a 64bit bus just waiting for release ?
And that it's not pin-compatible with the 7300 GS' PCB ?
 
This is what we know about both low end cores right now (yes, they're not direct equivalents).
Based on VR's reports so far:
8500GT RV610XT
8400GS RV610PRO
8300GS RV610LE
And is a RV610 really smaller at 65nm than a G86 at 80nm, in terms of die size and power consumption ?
RV610 is smaller, not sure about power consumption as it will depend on the clocks.

Also would a 64b PCB be cheaper to manufacture compared to 128b?


Assuming R600's die size is ~420mm2 on 80nm, and RV630's die size is ~160mm2 on 65nm... Then, if the architecture was very scalable (which you would expect it to be), it would be nearly exactly half a R600. It would certainly make sense if RV670, RV630 and RV610 were 75%, 50% and 25% of R600 respectively.
That makes perfect sense.

There was an earlier leak on RV610 die size, it was wrong on the gpu codename & bus width but you were spot on!

r6xxux1.png


I'm your new fan Arun. :yes:
 
What makes you say that other editions of G86 (like the 8400 GS and 8300 GS) don't have a further "LE" version with a 64bit bus just waiting for release ?
And that it's not pin-compatible with the 7300 GS' PCB ?
Your prior post was referencing the 8500GT board specifically, not some others. Given that you were talking about what is "known", why switch to what isn't? For instance, what would the score be for these configurations you are now drawing reference to?

As for the comment on whether these apparent scores are for the LE or now, for reference X1300 PRO scores around 1200 in 3DMark06 now. A generational step like that [previous top SKU performances moving to a lower SKUp, even accounting for bus difference, may not be unreasonable.
 
Based on VR's reports so far:

The X1300 XT is just a renamed (and shrunk) X1600 Pro, so you can't really count it as a separate version, performance-wise.
Unless you're willing to go the extra mile and count the 7600GS as a part of that competing family...
So, we can't be certain there will be three distinct RV610's.


RV610 is smaller, not sure about power consumption as it will depend on the clocks.
If you're so certain, i honestly believe you. :)
Although..., i'm curious.
Which process is more expensive right now at TSMC ?
The "value" 80nm or the "value" 65nm ?
And why did Nvidia decide to make G84 and G86 at the 80nm half-node, when the G72 is being produced right now at 65nm ?

Also would a 64b PCB be cheaper to manufacture compared to 128b?

Well, if Nvidia is really willing to do it for even the low end 8300 GS, it must be ! :D
There's more to it than just bus width (surface mounted components, layers, mass production and reusable PCB's from previous generations -G86 will be the fourth pin-compatible core on these PCB's, after NV43, G73 and G84-, etc).
 
Last edited by a moderator:
The X1300 XT is just a renamed (and shrunk) X1600 Pro, so you can't really count it as a separate version, performance-wise.Unless you're willing to go the extra mile and count the 7600GS as a part of that competing family...So, we can't be certain there will be three distinct RV610's.
Well, its what VR have suggested so far. Only when the cards are launched, we will get to know such details.

If you're so certain, i honestly believe you. :)
Although..., i'm curious.
Which process is more expensive right now at TSMC ?
The "value" 80nm or the "value" 65nm ?
And why did Nvidia decide to make G84 and G86 at the 80nm half-node, when the G72 is being produced right now at 65nm ?
I linked to it right in my post, it's Arun peice. As for your second question, I believe there is an Arun's post on that very same topic; 65nm may not cheaper than 80nm.

Well, if Nvidia is really willing to do it for even the low end 8300 GS, it must be ! :D
There's more to it than just bus width (surface mounted components, layers, mass production and reusable PCB's from previous generations -G86 will be the fourth pin-compatible core on these PCB's, after NV43, G73 and G84-, etc).
Nvidia's AIB partners might already have lots of G7x PCBs lying around for this exact purpose.

Compiled the die sizes of what we got so far:
r6xxvg8xzr0.png

Just from looking at that list, RV630 might bring some serious pain on G84 and G86 on RV610.
 
As i said before, just look at VR-Zone's picture of the 8500 GT, and then compare it with a 7600 GS or a 7300 GT.
Likewise for the 8600 GT (not GTS), versus the old 7600 GT and 6600 GT.
The images were sufficiently detailed that you could tell they were exactly pin-for-pin compatible?
 
Compiled the die sizes of what we got so far:
r6xxvg8xzr0.png

Just from looking at that list, RV630 might bring some serious pain on G84 and G86 on RV610.

Where did you get the G86 core size ?
I've only seen a photo of the G84 die so far...

Even the mere difference between G80 and G84 heavily suggests a future "middle" player.


The images were sufficiently detailed that you could tell they were exactly pin-for-pin compatible?

The PCB ones ?
Sure.
Just look at capacitor and resistor placement, etc.
In fact, take a look at these two PCB's:

http://www.ocworkbench.com/2007/nvidia/8600GT/8600GT-1 copy.jpg

http://www.techpowerup.com/img/06-03-08/NV7600GT.jpg


One is a 8600 GT, the other is a 7600 GT.
See any resemblance ? ;)
 
Last edited by a moderator:
The PCB ones ?
Sure.
Just look at capacitor and resistor placement, etc.
In fact, take a look at these two PCB's
The PCB's may be similar, but that does that actually tell us the ASIC pinouts are exactly pin-for-pin compatible? Or that they have similar designs / thermal properties?
 
Is RV630 supposed to be the midrange line of cards? If so, I can't say I'd be surprised if the upper midrange cards approached the power envelope of previous gen lower high end cards.

Especially when you factor in 512 megs of memory.

And I'm sure including increased support for greater hardware acceleration for both H.264 and VC-1 doesn't come cheap. My memory is sketchy on this, but back when AVIVO came out I remember talk about certain decode functions being relatively complex and thus done in software. And it appears that ATI is now including more of those functions into the R6xx cores.

Then again, my memory might be completely faulty on that. :)

Regards,
SB
 
Status
Not open for further replies.
Back
Top