einstein_314
04-02-04, 05:19 PM
Just found this over at inquirer.
WE WONDER whether there was any irony intended by Nvidia holding its NDA 'Editors day' briefing yesterday, April 1st.
Since we first broke the news that NV40 would be a .13µ, 200+ million transistor, sixteen pipeline part, since we first reported on a first hands on look at an early NV40, a few further snippets have winged our way.
This time, one of our Nigerian sources gives us some insight into what they saw, as Nvidia debriefed them on the NV40's marchitecture.
Apparently Nvidia say that what its been showing its NDA gagged partners is not really what it will bring to market.
Firstly, the thermal control system. The pictures that have circulated the Web in the past week only give rise to speculation as to whether Nvidia will introduce another NV30'esque 'DustBuster' as only a bare PCB is shown.
The NV40 engineering samples shown at CeBIT were rather different than the 175 million transistor part photographed.
For start they are on a green PCB which to those in the know, kind of gives the game away to Nvidia's stormtroopers as to perhaps one or the other of two partners whose board has been photographed...
Also as Fudo mentioned earlier, those NV40 engineering samples are pretty big, and aside from the dual power connectors on the AGP variant – apparently the PCI Express version will require only one power connector – the board had to accommodate a fairly dramatic thermal control system.
This, like Nvidia NV30, NV35GL and NV38 means that the engineering samples will - on most mainboards, though not on the tasty ABIT AN7 nForce2 based mainboard we're evaluating at the moment, encroach on the first PCI slot.
Nvidia seems to have told more than one of its partners that the final production board might not have this requirement, which could be good news for small form factor PC owners. Some, in wanting a high-end 3D accelerator, have had to either go for an MSI 5950 Ultra – which deviates from the NV38 reference heatsink design and is much slimmer, or forget Nvidia and go with an ATi Radeon 9800XT.
What is particularly dramatic about the NV40 engineering samples heatsink, is that it also seems to require the implementation of heatpipe technology to speed the transfer of heat away from what is clearly a toasty core.
If this were a passive solution then our take is that a good chunk of the the market would be rejoicing, but it's not – it's an active solution which uses a fairly substantial centrifugal 'blower' – as per the reference 5950 Ultra (NV38) - and, we're told, it's not the quietest.
This surprises me somewhat, as following the embarrassing, almost global condemnation of the NV30 reference design, Nvidia – I was confident - would have paid particular attention to the fan acoustics. Still we'd like to point out that what is being described here is Nvidia's NV40 engineering samples, and if what is suggested is true, things could well change in production.
In contrast the ATI R420 engineering samples that have been rolled out, again a .13 micron part, boast an active heatsink for the VPU, which is somewhat similar in design to that on the Radeon 9800XT (R360), but even smaller.
Smaller by design, we think, to prevent heat transferring across the heatsink surface and increasing the temperature of the much cooler running (than both DDR-1 and GDDR-2) GDDR-3 memory modules.
We're confident, that the ever cost conscious ATI, will not have any heatsinks on the memory modules of ATI X800, and we suspect that the rear memory modules on most NV40 boards won't either.
Secondly, we've had several reports that Nvidia's back door action included demonstrations of Unreal Tournament 2004, but that during these demonstrations, there were noticeable "artefacts".
When we spoke to Nvidia about our findings, they were immediately dismissed as an immature driver issue which had now been addressed. Of course this is entirely feasible, and we hope that this is the case.
However we can't help wondering whether in an effort to impress the impressionable, the engineering samples memory clock had been adjusted to dizzy frequencies. Frequencies that the Samsung GDDR3 video memory employed, really couldn't hack. µ
here (http://www.theinquirer.net/?article=15115) for the article.
I just can't see nVidia making that mistake again. Must be the inquirer dreaming up things again.
WE WONDER whether there was any irony intended by Nvidia holding its NDA 'Editors day' briefing yesterday, April 1st.
Since we first broke the news that NV40 would be a .13µ, 200+ million transistor, sixteen pipeline part, since we first reported on a first hands on look at an early NV40, a few further snippets have winged our way.
This time, one of our Nigerian sources gives us some insight into what they saw, as Nvidia debriefed them on the NV40's marchitecture.
Apparently Nvidia say that what its been showing its NDA gagged partners is not really what it will bring to market.
Firstly, the thermal control system. The pictures that have circulated the Web in the past week only give rise to speculation as to whether Nvidia will introduce another NV30'esque 'DustBuster' as only a bare PCB is shown.
The NV40 engineering samples shown at CeBIT were rather different than the 175 million transistor part photographed.
For start they are on a green PCB which to those in the know, kind of gives the game away to Nvidia's stormtroopers as to perhaps one or the other of two partners whose board has been photographed...
Also as Fudo mentioned earlier, those NV40 engineering samples are pretty big, and aside from the dual power connectors on the AGP variant – apparently the PCI Express version will require only one power connector – the board had to accommodate a fairly dramatic thermal control system.
This, like Nvidia NV30, NV35GL and NV38 means that the engineering samples will - on most mainboards, though not on the tasty ABIT AN7 nForce2 based mainboard we're evaluating at the moment, encroach on the first PCI slot.
Nvidia seems to have told more than one of its partners that the final production board might not have this requirement, which could be good news for small form factor PC owners. Some, in wanting a high-end 3D accelerator, have had to either go for an MSI 5950 Ultra – which deviates from the NV38 reference heatsink design and is much slimmer, or forget Nvidia and go with an ATi Radeon 9800XT.
What is particularly dramatic about the NV40 engineering samples heatsink, is that it also seems to require the implementation of heatpipe technology to speed the transfer of heat away from what is clearly a toasty core.
If this were a passive solution then our take is that a good chunk of the the market would be rejoicing, but it's not – it's an active solution which uses a fairly substantial centrifugal 'blower' – as per the reference 5950 Ultra (NV38) - and, we're told, it's not the quietest.
This surprises me somewhat, as following the embarrassing, almost global condemnation of the NV30 reference design, Nvidia – I was confident - would have paid particular attention to the fan acoustics. Still we'd like to point out that what is being described here is Nvidia's NV40 engineering samples, and if what is suggested is true, things could well change in production.
In contrast the ATI R420 engineering samples that have been rolled out, again a .13 micron part, boast an active heatsink for the VPU, which is somewhat similar in design to that on the Radeon 9800XT (R360), but even smaller.
Smaller by design, we think, to prevent heat transferring across the heatsink surface and increasing the temperature of the much cooler running (than both DDR-1 and GDDR-2) GDDR-3 memory modules.
We're confident, that the ever cost conscious ATI, will not have any heatsinks on the memory modules of ATI X800, and we suspect that the rear memory modules on most NV40 boards won't either.
Secondly, we've had several reports that Nvidia's back door action included demonstrations of Unreal Tournament 2004, but that during these demonstrations, there were noticeable "artefacts".
When we spoke to Nvidia about our findings, they were immediately dismissed as an immature driver issue which had now been addressed. Of course this is entirely feasible, and we hope that this is the case.
However we can't help wondering whether in an effort to impress the impressionable, the engineering samples memory clock had been adjusted to dizzy frequencies. Frequencies that the Samsung GDDR3 video memory employed, really couldn't hack. µ
here (http://www.theinquirer.net/?article=15115) for the article.
I just can't see nVidia making that mistake again. Must be the inquirer dreaming up things again.