NV45 spotted @ Anand

[Lezmaka]

Well perhaps there's been something going on behind the scenes that we're not aware of and BTX PSUs (+ maybe newer ATX PSUs) will come with a dedicated, 3x2 connector for GPUs. Until then I'm sure an adapter will allow use with current units.

Looks like they've replaced the white MM header with on-board contacts too.

Just found a small blurb about the 2x3 connector on page 32 of this pdf

NOTE
Details of the 2x3 “Optional Power Connectorâ€￾ mentioned in the ATX 2.03 Specification are omitted from this design guide until such time as the signals on that connector are more rigidly defined.

Edit:

Hmmm, going by the description on page 21 of the ATX 2.03 spec, I don't see how it could be used for a video card. From my limited knowledge/comprehension, it seems it's supposed to be for unpowered IEEE-1394 (FireWire) devices.

4.2 Optional Power Connector
In addition to the main power connector, a 2-by-3 optional connector from the power supply can be used in an ATX system for such ancillary functions as fan monitoring, fan control, IEEE-1394 power source, and a remote 3.3 V sense line. Although this connector is not required for ATX compliance, it adds benefits that are compelling for a full-featured system:
• The fan monitor features add the ability to monitor and detect fan failures.
• A built-in fan control allows the motherboard to request fan shutdown when the system goes into a sleep or suspend mode.
• Fan speed control is possible to allow for slower fan speeds during low power usage.

...

4.2.4 1394 V Pin
This pin on the optional connector allows for implementation of a segregated voltage supply rail for use with unpowered IEEE-1394 solutions. The power derived from this pin should be used to power only 1394 connectors. The output of this power rail is dependent on the 1394 compatibility required. Use of this power rail for motherboard or other power needs may have unpredictable results, because power for 1394 devices is not required to be regulated and may provide voltage levels between 8 and 40 volts. See the applicable IEEE-1394 specification for details on the specific power requirements for this voltage rail. If this power rail is implemented, it should operate such that the main PS_ON# signal must be asserted low for power to be delivered at this connector.

Hopefully someone here with more knowledge can come up with something from this...

 

[anaqer]

Another change we noticed was this heatsink over some of the power circuitry on NV45:
http://images.anandtech.com/reviews/shows/computex/2004/nv45update/heatsink.jpg

Looks to be exactly the same as the one 6800U, no?

 

[anaqer]

Currently it seems that NV45 is running slower than NV40, but there may be plans for NV45 to be clocked slightly higher.

Anything to add, Baron?

 

[CJ]

Baron - do you know how much the core clock is being upped?

Well if you click the anand link, it says 450.

Urhm in the updated article it also lists the speeds... it says that the NV45 is running at slower clockspeeds than NV40: 350 Mhz core, 1000 Mhz mem.

First of all, it seems that at least some of the samples of NV45 that NVIDIA handed out are running at 350MHz with 1GHz GDDR3 memory, which is slower than the current NV40 GPU.

http://www.anandtech.com/video/showdoc.html?i=2067

 

[Evildeus]

This part was intriguing, but i don't think we will see anything new.

We can't help but wonder if NVIDIA has done any internal tuning on NV45 that would make performance at the same clock, go up. Unfortunately, architectural details are sparse at best, but we're working hard to get them if we can.

 

[nutball]

Is there any reason at all why early boards being shown to the unwashed public should be clocked at anything resembling their intended clock-speeds?

I mean OK if it could be clocked at 600 or 700Mhz, maybe show it now to put the sh*ts up ATI, but even then there might be marketing reasons not to do so (who'd buy a 6800, for a start?)

I just wonder if people aren't reading too much into that clock-speed. :?

 

[Dave Baumann]

It's also good to see that someone is finally making a dual-DVI X800 (Asus card shown later in the article). I know quite a few people have been asking for one of these for some time.

Ahhh, I heard about this. Look at the X800 XT reference images at Tech Report - the reference also has dual DVI. I had heard they were supporting integrated dual DVI with R42x, and it looks like R423 may have it.

 

[Joe DeFuria]

Currently it seems that NV45 is running slower than NV40, but there may be plans for NV45 to be clocked slightly higher.

Anything to add, Baron?

I'm not baron, but I would think that the added heat / power from the bridge chip is certainly not going to help the NV45 clocks...

 

[Tim Murray]

Currently it seems that NV45 is running slower than NV40, but there may be plans for NV45 to be clocked slightly higher.

Anything to add, Baron?

I'm not baron, but I would think that the added heat / power from the bridge chip is certainly not going to help the NV45 clocks...

Hm, guess the location of the bridge chip explains why nobody knew it wasn't native PEG until now...

Anyway, it's still a few months away, and this isn't anything near a final board. Give it a couple of months.

 

[Joe DeFuria]

Anyway, it's still a few months away, and this isn't anything near a final board. Give it a couple of months.

Absolutely. I would have to think that nVidia is going to at minimum, target current NV40 clocks. However, I'm just pointing out that any "gains" they get from going to TSMC 0.13 low-k, is going to be offset to at least some extent by the presense of the bridge. Now you're pumping out more heat, and have yet another bus to manage, which makes higher clocks more difficult.

In other words, if NV45 was native PEG, I would expect nVidia to have a much easier time to get the clock speeds up, with a bridge, it's going to be more of a challenge.

 

[Xmas]

I don't think that bridge adds much of a challenge. Because if it did, why would they move it to the same package?

 

[Joe DeFuria]

I don't think that bridge adds much of a challenge. Because if it did, why would they move it to the same package?

Obvious: because it's easier on board manufacturers. Better for nvidia to manage HSI on the package, than for the board manufacturers to.

 

[Nebuchadnezzar]

http://www.anandtech.com/showdoc.html?i=2068

 

[trinibwoy]

http://www.anandtech.com/showdoc.html?i=2068

Well a lot of people must be disappointed. NV45 is irrelevant for all purposes and the strange black connector is also on ATI's part.

 

[Bjorn]

http://www.anandtech.com/showdoc.html?i=2068

Well a lot of people must be disappointed. NV45 is irrelevant for all purposes and the strange black connector is also on ATI's part.

It's just a name so i don't see what there is to be dissapointed about. The NV48 is going to be the real refresh part anyways. You could argue that NV41 or something like that would have been a better name but it's just internal so who cares.

 

[anaqer]

NV might run out of digits if it keeps up doing this, last time it had just two of 'em left (2 and 3)...

 

[Joe DeFuria]

It's just a name so i don't see what there is to be dissapointed about.

Well, the "disappointment" is that we all knew the NV45 was taped out and pretty much ready to go, so that a true NV40 "refresh" looked to be very close to imminent.

When the NV48 may actually appear is anyone's guess at this time.

I'm starting to wonder about "this fall's" line-up from both ATI and nVidia. I mean, in the high-end space both the 6800 ultra or XT PE will just barely be out in quantity (July?) by the time the "traditional fall lanuch time" occurs, which is usually first week of September. (This was a relatively LATE shipping of spring products.)

Me-thinks that this year's "fall refresh" at the high end is either going to be considerably delayed (to like November), or might be by-passed completely until next spring.

If nVidia doesn't manage to get an NV48 out for this fall, I'm betting we won't see an R420 refresh either....

 

[nutball]

Maybe the product cycles really are slowing. (Good job too IMO, if and when it happens).

 

[Evildeus]

Well, if we all knew that we should also have known that there were not much reason to be a real refresh. Moreover, there's not much room in front of Ati and Nv (perhaps a bit more for Nv but not much), so i don't think we will see some real improvement before 2006 now.

 

[Bjorn]

Well, the "disappointment" is that we all knew the NV45 was taped out and pretty much ready to go, so that a true NV40 "refresh" looked to be very close to imminent.

That's true, i forgot the "NV45 taped out at the same time as NV40" rumours.

I'm starting to wonder about "this fall's" line-up from both ATI and nVidia. I mean, in the high-end space both the 6800 ultra or XT PE will just barely be out in quantity (July?) by the time the "traditional fall lanuch time" occurs, which is usually first week of September. (This was a relatively LATE shipping of spring products.)

Me-thinks that this year's "fall refresh" at the high end is either going to be considerably delayed (to like November), or might be by-passed completely until next spring.

If nVidia doesn't manage to get an NV48 out for this fall, I'm betting we won't see an R420 refresh either....

I agree. Though the new information about the 0.11 micron process could possibly lead to a change of plans for the IHV's.