Different core steppings and the changes

[Glonk]

I just noticed that my NV25 is an "A3" stepping (I think that's the right term), but many screenshots I've seen of older NV25s say A2 on them. Likewise, my NV20 is an "A5", but I'm not sure if there were production models before A5 either.

My question is: Is there any noticable performance/feature differences between the steppings, or do they exist just to try to get yields higher? Are higher steppings more likely to overclock better?

 

[alexsok]

I just noticed that my NV25 is an "A3" stepping (I think that's the right term), but many screenshots I've seen of older NV25s say A2 on them. Likewise, my NV20 is an "A5", but I'm not sure if there were production models before A5 either.

My question is: Is there any noticable performance/feature differences between the steppings, or do they exist just to try to get yields higher? Are higher steppings more likely to overclock better?

Well, if I recall correctly, there was a difference between NV25 "A2" stepping and "A3" stepping, concerning support for some feature, although I don't remember which one...

Maybe someone knows?

 

[GetStuff]

I don't remember reading any official comments on that so I just assumed the new revision was to increase yields.

 

[Glonk]

I've searched on Google and couldn't find anything about the differences between A3 and A2 NV25 chips.

 

[mat]

would be interesting if there is a difference... my Ti4200 is an A2 stepping

 

[KimB]

Well, I do know that the NV20 A5 stepping fixed an issue with 3D textures...I'm not aware of any hardware bugs that are unresolved by drivers in the NV25.

 

[WOOFLE]

"steppings" are usually minor revisions of the core to tweak certain aspects of the core taht aren't reaching the manufacturers expectations. Sometimes they are to fix actual bugs, but usually by the time they reach the general public this is not the case and that's a whole other story from the "normal" process of it.

They are not at all uncommon. In the CPU world Intel announce their new steppings like the P3s with Cb0, Cc0, Cd0. AMD don't usually go announcing them, but you can tell by reading the code on the chip when there are major changes. The more common reason for new steppings are to increase processor yield (how many working chips come out of a wafer bake) and to improve the attainable clock speeds.
You see when say Intel released the P4 or AMD releasles the Palomino they don't just start designing the next project or major revision they are also constantly trying to improve what they can get out of their current designs in terms of yield and clockspeed. Because if tey can make more CPUs, or faster ones they can make more money.

These core revisions usually shuffle around and or add in some more transistors to allow for yield or binning (the process of sorting out fast CPUs from slow ones) improvements. The sort of things they do are that if a CPU design is being unable to clock higher because one small section is recieving what I suppose you could best describe as "jitter" in it's signals.

Just say that there are 2 transistors that need to send their power on to another two transistors to complete an operation and that they both need to arrive as close together as possible to stop the CPU from crashing by screwing up. Maybe the initial design will work fine at aaa MHz, but once they are trying to get the CPU to run at bbb MHz they find that although the signals are still only arriving out of sink by 0.X of a second that the increased ratio of 0.X of a second to the faster clock cycle has caused regular crashing. The engineers then have the challenge of manipulating the minute 'wires' within the processor so that the signals arrive close enough in time proportionately that the CPU doesn't crash.

Hmm I'm not an expert on this (I've only done the most basic of digital electronics), but I hope it helps you understand what steppings are about. BTW when overclockers talk about differnct steppings, they actually often only mean different batches, not steppings. A common misonception is that the words are interchangeable.

 

[alexsok]

Well, I do know that the NV20 A5 stepping fixed an issue with 3D textures...I'm not aware of any hardware bugs that are unresolved by drivers in the NV25.

Yes, i was 3D textures...

And it was a hardware bug, not sure if it was resolved in drivers, but it was resolved at "A3" stepping.

 

[AlexUnwinder]

Hardware accelerated line antialiasing in OpenGL is disabled in NV25/A2. It's disabled at the driver level so I don't know if it's just a marketing or A2 has buggy AA implementation.

 

[KimB]

Hardware-accelerated line antialiasing is supposed to be one of the features of the workstation-level cards, I thought. After all, in games, I don't really see much reason to have line AA.