Another MX fiasco?

[Ante P]

Well I dunno, the chipsets name just scares me: NV18.
Accoring to two french sites it will be equivelent GeForce 5 MX.

"One learns on HardWare.fr that it would be extremely possible that the next chip of nVIDIA, known until now under the name of code NV18 bears the name of GeForce5 MX This information is not at all official (for the moment) but would come from one of the many graphic board manufacturers nVIDIA."

I sure hope it's not a beefed up GF4MX with AGP8 support as some people seem to belive..
That would be a disaster.

And I hope it's not some cheapskate "GF3MX" with only DX8.0 support.

Well, time will tell.

 

[pascal]

My guess worst than that. Probably:

0x - TNT class
1x - DX7
2x - DX8
3x - Dx9

 

[Dave Baumann]

I sure hope it's not a beefed up GF4MX with AGP8 support as some people seem to belive..

Every rumour or report I've seen on both NV18 and NV28 says they are nothing more than their current GeForce 4's with AGP8X support.

 

[Randell]

what about the rumours regarding extra ~20M transistors on the nv18/28?

 

[Dave Baumann]

There aren't any. The only place thats been seen are from the slides at the conference.

 

[Randell]

well those slide caused rumours then thast what I meant. AGP 8x wouldn't add all those extra transistors would it?

 

[kid_crisis]

"what about the rumours regarding extra ~20M transistors on the nv18/28?"

Dave's right, there aren't any rumors about that. Jen Hsun Huang stated specifically at the July 11th CC that NV28 has 86 million transistors, and NV18 has 81 million transistors. When the CEO states it, I wouldn't call it a rumor.

As to what those extra transistors are for, there is no way that it requires an extra 23 million transistors to implement AGP-8x. I guess we'll find out next week.

Surpisingly, we seem to know more right now about NV30 than we do about NV28 and NV18.

 

[Randell]

"what about the rumours regarding extra ~20M transistors on the nv18/28?"

Dave's right, there aren't any rumors about that. Jen Hsun Huang stated specifically at the July 11th CC that NV28 has 86 million transistors, and NV18 has 81 million transistors. When the CEO states it, I wouldn't call it a rumor.

As to what those extra transistors are for, there is no way that it requires an extra 23 million transistors to implement AGP-8x. I guess we'll find out next week.

Surpisingly, we seem to know more right now about NV30 than we do about NV28 and NV18.

how does that jive with Daves statement above? i.e. everything he's heard or read states GF4/MX plus 8xAGP?

 

[alexsok]

"what about the rumours regarding extra ~20M transistors on the nv18/28?"

Dave's right, there aren't any rumors about that. Jen Hsun Huang stated specifically at the July 11th CC that NV28 has 86 million transistors, and NV18 has 81 million transistors. When the CEO states it, I wouldn't call it a rumor.

As to what those extra transistors are for, there is no way that it requires an extra 23 million transistors to implement AGP-8x. I guess we'll find out next week.

Surpisingly, we seem to know more right now about NV30 than we do about NV28 and NV18.

How come next week?

If I'm not mistaken, GF4TI had only a 3 million transistors advantage over GF3 and it had many additions (I've read on some forum about them and they were pretty significant), so I can't imagine why would it take 23 million transistors to implement AGP 8x support?!

Besides, in current games, it has next to zero advantage over AGP 4x, so I doubt that's the only addition in NV28.

 

[KimB]

It sounds to me like this could possibly be a radical departure from their previous naming scheme.

The other possibilities I could think of would be:

1. eDRAM (to lower costs, not necessarily to improve performance...might particularly make sense for a laptop part).

2. More pipelines than the previous parts.

3. Massively-increased cache sizes.

 

[alexsok]

It sounds to me like this could possibly be a radical departure from their previous naming scheme.

The other possibilities I could think of would be:

1. eDRAM (to lower costs, not necessarily to improve performance...might particularly make sense for a laptop part).

2. More pipelines than the previous parts.

3. Massively-increased cache sizes.

I still think it's too little changes for a 81million transistors part...

 

[KimB]

8MB of eDRAM from TSMC would take up 64 million transistors.

 

[alexsok]

8MB of eDRAM from TSMC would take up 64 million transistors.

Then what about NV28?

 

[Nagorak]

I think the fact that they have been so quiet about these parts is a pretty good indication that they're nothing all that new. I really doubt the newer NV28 is going to be much more competition for the R300 than the current Ti4600. But we can hope since it might drive down prices... (then again if both ATi and Nvidia are content asking high prices...)

 

[snk]

The slides showing NV28 and NV18 were extremely blurry and hard to read. Number 5 looked like 6 (there is no NV26, only NV25), and 6 looked like 8 (NV25 has 63, not 83 million transistors). Finally, number 3 looked very much like 8 as well (there is no NV80). You can see for yourself:

http://www.forum-3dcenter.net/vbulletin/attachment.php?s=&postid=324971

Based on this, I am 100% sure that NV28 and NV18 have 65M and 31M transistors respectively, 2M and 3M up from their predecessors.

 

[kid_crisis]

Based on this, I am 100% sure that NV28 and NV18 have 65M and 31M transistors respectively, 2M and 3M up from their predecessors.

Yes, that was the slide shown at the July 11th CC, and it clearly showed the following numbers at the CC (which Jen Hsun also stated verbally):

NV18: 81 million
NV28: 86 M
NV30: 120 M

If you take a second look at that slide you can see that it is a possible interpretation (and it also happens to be the correct one).

 

[Dave Baumann]

Yes, that was the slide shown at the July 11th CC, and it clearly showed the following numbers at the CC (which Jen Hsun also stated verbally):

He wasn't reading from the poor resolution slide was he?

 

[Nagorak]

Based on this, I am 100% sure that NV28 and NV18 have 65M and 31M transistors respectively, 2M and 3M up from their predecessors.

Yes, that was the slide shown at the July 11th CC, and it clearly showed the following numbers at the CC (which Jen Hsun also stated verbally):

NV18: 81 million
NV28: 86 M
NV30: 120 M

If you take a second look at that slide you can see that it is a possible interpretation (and it also happens to be the correct one).

You realize that the CEO is probably not directly involved with the development of any of these products right? Who's to say he wasn't just reading off the slide himself?

 

[kid_crisis]

He wasn't reading from the poor resolution slide was he??

Heh. Nope, it was clearly readable at the CC.

You realize that the CEO is probably not directly involved with the development of any of these products right? Who's to say he wasn't just reading off the slide himself?

Yeah, that's possible, and I suppose they could have had a typo on the slide. I'm hoping Jen Hsun is a little more on the ball than that. We should know for sure at the next CC on Thursday, Aug 15th.

 

[Dave Baumann]

I just think those numbers don’t seem to add up. Lets look at his anecdotally:

• By the look of that chart, the pattern of the older chips roughly shows relative performance on the y scale; you’ll note that for the new parts NV30 is much higher above everything (as we’d expect), but NV18 and NV28 are at the same y positions as NV17 and NV25 respectively. If the y scale does represent relative performance this would indicate that those extra transistors are not doing anything for performance.
• As has been pointed out, historically NV1x parts relate to DX7 class products – if this is the case with NV18 it would indicate that those extra transistors are also not for increasing the feature specification.
• Historically NVIDIA have introduced a low end offering and a high end offering; given the positioning on the graph and the naming of NV18 this would indicate it’s the low end part. This being the case doesn’t 81M transistors for the NV18 seem a bit too much? We know that NV30 is the only .13um product so NV18 is going to be .15um; ATi could only scale down to the low end with Radeon 9000 by scaling back many of 8500’s performance features to cut down on the 60M transistors it uses.

When you add to the fact that the only rumours anyone has on these products are added AGP8X support I think theres enough room to doubt what he/the slides says until something is actually announced.