SPCR

NVIDIA's picoITX Atom+GeForce 9400M reference design.

Atom is honestly fast enough for many tasks, delivering the performance of a mainstream notebook from 4 years ago. The problem is that there are some applications that are commonplace today that can’t run on Atom. HD video playback isn’t possible on Atom + 945G platforms because the CPU isn’t fast enough to decode high bitrate video (much less H.264) and the chipset doesn’t support HD video decode acceleration.

NVIDIA saw an opportunity with Atom. Intel had a very popular CPU, that could be used in many more environments if it could only be paired with a more powerful chipset. Enter the GeForce 9400M.

Source -- http://www.anandtech.com/video/showdoc.aspx?i=3478

Very promising. It would make a heck of an upgrade for a new Mac Mini.

dougz wrote:Very promising. It would make a heck of an upgrade for a new Mac Mini.

Techreport did some testing of the ASUS N10Jc-A1 (Atom N270 + 9300M GS) and found that most games were severly CPU limitedto the point of being unplayable. A Geforce 9300 variant stripped down to 8 shaders might be plenty. Reducing the shaders wouldn't effect the HD capabilities. But if nVidia gets the MCP79 below the 9.3W TDP of the 945GSE + ICH7M, I won't complain.

I would like to see gaming tests done with a 2GHz Atom--even if Intel doesn't sell it at that speed, it is certainly capable of it. As an overclock it mainly provides more FSB and memory bandwidth (DDR2 667). I don't place much hope in a dual core Atom for increasing game performance. Even faster FSB and memory (DDR2 800) would likely be better.

TechReport wrote:All of these were met with varying degrees of failure. Team Fortress 2 in particular ran horribly, even at its lowest detail settings.

Team Fortress 2 will run on my overclocked ASUS Eee PC 100HA for a few seconds and then crash with what appears to be a sound driver issue. The less demanding Portal did the same. Yes, many drivers were tried.

As an alternative to Atom + 945G, Nvidia's Ion is pure win. I'd be willing to say that this platform can achieve <10W idles. This combo looks so amazing.... full HD capable, Pico-ITX, <10W idles.... the possibilities are endless. File servers can now play Blu-Ray. I'm looking forward to this.

QuietOC wrote:most games were severly CPU limited to the point of being unplayable. A Geforce 9300 variant stripped down to 8 shaders might be plenty. Reducing the shaders wouldn't effect the HD capabilities. But if nVidia gets the MCP79 below the 9.3W TDP of the 945GSE + ICH7M, I won't complain.

I'm not a gamer, so video, power, noise and price are my interests. This design appeals to me. (Thanks for the article link.)

Apple is getting a premium for the new MacBooks with the NVIDIA chip, but there are a significant number of other enhancements so it is hard to get a feel for chipset pricing.

Does anyone have an feel for the price difference between the GeForce 9400M and the Intel 945GC? (The GS is supposedly rarely used because of its cost.)

I've always loved the Atom design, but Intel has tried to avoid cannibalizing other products, resulting in sub-par Atom implementations. (Number of DIMMs, 1 GB memory, 945 video, GC/GS pricing.) Thankfully, that seems to be changing.

I don't think you can really consider anything more than very, very basic gaming on such a small, low power system so lack of performance in this field doesn't really worry me.

IMHO, this 'Ion' platform on a reasonably specified Mini-ITX board would be absolutely ideal for creating a small low-priced HTPC which can play pretty much anything thrown at it but which has enough power for relatively speedy web-browsing and basic word processing etc.

With reasonable heatsinks, I suspect such a device could be passively cooled without too many problems as well.

dougz wrote:Does anyone have an feel for the price difference between the GeForce 9400M and the Intel 945GC? (The GS is supposedly rarely used because of its cost.)

Well the 945GC is basically worthless. I believe the entire netbook platform (N270 + 945GSE + ICH7M) is $80 in 1000 unit quantities (probably less now). The cheapest Geforce 9300 motherboard is $95 after a $20 rebate.



That is a huge, expensive chip compared to the 740G/SB700 pair (I have pictures!)

I see the potential problems being: this chipset hasn't been low wattage at all (lower than a G45/G35 isn't saying much), and it is expensive. Maybe nVidia will strip it down and/or underclock it and sell it a lot cheaper?

For those who think this is for HD video:

TechReport wrote:Curious to see whether the Intel integrated graphics would fare much worse, I tested the same videos on both the GeForce and the GMA 950, swapping HDMI output for the N10J's own display. Surprisingly, there was little difference in playback performance between the two graphics options. The GMA's CPU utilization was about 5% higher with my MPEG2 clip, but otherwise CPU utilization was unchanged from the previous round of testing. There was also no difference in CPU utilization between the GeForce and GMA when playing back standard-definition DivX content, which pegged the CPU at roughly 15%.

QuietOC wrote: For those who think this is for HD video

I guess I'm having trouble understanding why this should be the case. From the second page of the AnandTech piece:

That's 27% CPU utilization on an Intel Atom processor when playing back a 18Mbps 1080p H.264 scene

Note that this is the very same 9400M that’s in the new Apple notebooks, not a watered down version, the clocks, features and performance are the same (although presumably OEMs could choose to underclock the graphics core for particularly power/heat sensitive applications).

Could there be driver or OS differences between the articles?

Here's an article on the use of a $20 Sempron LE-1150 (single core 2.0 GHz) with $30 ASUS GeForce 8400GS passively-cooled card playing back HD video with the new (for Linux) Video Decode and Presentation API for Unix -- http://www.phoronix.com/scan.php?page=a ... _gpu&num=1

Not a low power board (38 watts) but it is suggested that the PCI version could be used on an Atom system for HD video.

Without VDPAU the low-end Sempron was facing a near-constant utilization rate of more than 50% and frequently would spike to 100% usage. When switching to VDPAU, the CPU utilization started out at just under 20% and never went beyond that. Most of the time the CPU utilization was less than 10% and would just jump when restarting the H.264 video file playback....

We are very pleased with the results from this latest round of VDPAU testing. We were successful in playing high definition video files on Linux in multiple formats all from a $20 AMD Sempron processor and $30 NVIDIA GeForce graphics card. Though if these were encrypted video files and using say a Blu-Ray disc, this likely wouldn't be possible since the processor would have a very hard time keeping up, but anyways Linux is currently lacking such a player.

Stay tuned to Phoronix for more information on NVIDIA's VDPAU as it matures as well as details on AMD's XvBA as soon as it launches. You can share your video playback experiences under Linux in the Phoronix Forums.

Even more interesting would be Atom 330 + 9400M combination with GigE, 2-4 SATA ports, 2 SODIMM slots and a PCIe x1 slot. It could be a WHS system or a media center or an email / web browser system.

Make it so...

-D

derekva wrote:Even more interesting would be Atom 330 + 9400M combination with GigE, 2-4 SATA ports, 2 SODIMM slots and a PCIe x1 slot. It could be a WHS system or a media center or an email / web browser system.

No room on a Pico-ITX. No point for a reference board.

Make production version a Mini-ITX or Micro-ATX. Prefer a 330 dual core, preferably with dual channel memory.

angelkiller wrote:As an alternative to Atom + 945G, Nvidia's Ion is pure win. I'd be willing to say that this platform can achieve <10W idles.

I'll be very disappointed if this isn't substantially less than 10W at full load. My Via Pico ITX is under 11W under load as measured on the PSU input (which incidentally is nowhere near the ARM 11 it replaced, which was ~3W at full load).

|Romeo| wrote:

angelkiller wrote:As an alternative to Atom + 945G, Nvidia's Ion is pure win. I'd be willing to say that this platform can achieve <10W idles.

I'll be very disappointed if this isn't substantially less than 10W at full load. My Via Pico ITX is under 11W under load as measured on the PSU input (which incidentally is nowhere near the ARM 11 it replaced, which was ~3W at full load).

But can the Via decode high-bitrate H.264 video?

I think your expectations of <<10w load are unreasonable for such a feature-rich platform. I wouldn't be surprised if it could achieve <10w idle, though.

So much for the nvidia-via platform. It would have been nice if it were the nano in there.

It will be interesting to see what the supposed upcoming mac mini refresh will look like....the NVidia reference platform is certainly a possible direction. They just need to add an eSATA port or two and it's a great little low power system.

frostedflakes wrote:

|Romeo| wrote:

angelkiller wrote:As an alternative to Atom + 945G, Nvidia's Ion is pure win. I'd be willing to say that this platform can achieve <10W idles.

I'll be very disappointed if this isn't substantially less than 10W at full load. My Via Pico ITX is under 11W under load as measured on the PSU input (which incidentally is nowhere near the ARM 11 it replaced, which was ~3W at full load).

But can the Via decode high-bitrate H.264 video?

I think your expectations of <<10w load are unreasonable for such a feature-rich platform. I wouldn't be surprised if it could achieve <10w idle, though.

You know, I've never tried playing H.264 on it. Most of the time it runs monitor less which might not be the best way to enjoy video. For sure I won't ever see it stutter though

But in all seriousness, the heatsink and fan on the Via are not particularly over specified -and I see that the Ion has a very similar heatsink, which gives some upper limits on how much power the whole thing can take. And since the Atom is rather more frugal than the C7, the chipset must be getting bl**dy hot if the Ion board as a whole is drawing more than 10W. Remember that my numbers are form the input side of a 60W (WI) Pico PSU, that it's almost certainly not particularly efficient at ~1/6th of its rated power draw.

On a slightly different note, I notice with interest thst no one has published (or possibly not got from nvidia) a picture of the underside of the Ion board. Which I find odd/interesting given that there's an expansion connector of some kind there.

To CA_Steve: it looks like eSATA is on the daughter board.

|Romeo| wrote: I see that the Ion has a very similar heatsink, which gives some upper limits on how much power the whole thing can take.

Where do you see a heatsink? The pictures in the Anandtech article just show bare chips. I'm sure the Nvidia 9400M will have a heatsink in production versions, but a single-core Atom doesn't need one (although I bet production versions would go with a dual-core Atom).

|Romeo| wrote:On a slightly different note, I notice with interest thst no one has published (or possibly not got from nvidia) a picture of the underside of the Ion board. Which I find odd/interesting given that there's an expansion connector of some kind there.

Look at the second page of the Anandtech article. Near the middle of the page there is a picture showing the top and front views with labels. The connection on the underside is the SO-DIMM socket.

From your comments, I wonder if you are not confusing some other product you've read about with the Ion . . .

dougz wrote:

derekva wrote:Even more interesting would be Atom 330 + 9400M combination with GigE, 2-4 SATA ports, 2 SODIMM slots and a PCIe x1 slot. It could be a WHS system or a media center or an email / web browser system.

No room on a Pico-ITX. No point for a reference board.

Make production version a Mini-ITX or Micro-ATX. Prefer a 330 dual core, preferably with dual channel memory.

Sorry - that was assumed on my part. Pico ITX is fairly useless IMHO for anything other than embedded systems. Mini-ITX is about as small as I go (and even that can be somewhat of a PITA - what can I say, I like having expansion slots).

-D

jessekopelman wrote:

|Romeo| wrote: I see that the Ion has a very similar heatsink, which gives some upper limits on how much power the whole thing can take.

Where do you see a heatsink? The pictures in the Anandtech article just show bare chips. I'm sure the Nvidia 9400M will have a heatsink in production versions, but a single-core Atom doesn't need one (although I bet production versions would go with a dual-core Atom).

Look harder

http://www.anandtech.com/GalleryImage.aspx?id=4819

That is the Ion with its heatsink, mounted on the daughter board. As I say, a typical Pico ITX heatsink covering almost the entire board.

|Romeo| wrote:On a slightly different note, I notice with interest thst no one has published (or possibly not got from nvidia) a picture of the underside of the Ion board. Which I find odd/interesting given that there's an expansion connector of some kind there.

Look at the second page of the Anandtech article. Near the middle of the page there is a picture showing the top and front views with labels. The connection on the underside is the SO-DIMM socket.

From your comments, I wonder if you are not confusing some other product you've read about with the Ion . . .

I don't think so.

I am aware of the SODIMM slot being on the underside of the board in the Pico ITX specification. However if you look at the pictures of nvidias reference system design, you can see the Ion mounted on a daughter board (or "PICO ITX CARRIER CARD" as nvidia seem to be calling it). You can also see that DC in power socket is on the daughterboard, whilst the Ion has a standard Pico ITX power supply connector (if you can really call it standard, but regardless it's the 12 way 2mm connnector). However there is no such connector on the daughterboard for using wires (and why would you anyway).

Add to this all the connectors on the daughter board (VGA, audio, digital audio, more USB, the power button, the battery) and the fact that there is no connector on the top of Ion for a power button, power LED, reset button, etc. and it becomes very clear that there is a fairly substantial connector on the underside of the Ion in addition to the SODIMM slot.

Which can't be seen in the side view picture. If a picture ever surface of the underside of the Ion, it will show what I think will be a fairly substantial connector.

I am quite interested in this board, so have pored over all the information release to date quite closely.

[editted for one glaring spelling mistake]

Looks like NVIDIA is ahead, for now, with Pure Video-like hardware acceleration for Linux users.

When NVIDIA had introduced VDPAU they did supply patches that added support for this API to the MPlayer and ffmpeg projects. The open-source community came along and provided VDPAU support for MythTV late last month. This work done entirely outside of NVIDIA ended up working out quite well and now MythTV takes full advantage of this NVIDIA creation.

The Xine community has now come along, however, and created their own VDPAU support.

While VDPAU continues to be adopted by different media players and users have been experiencing great results, AMD's competitor, which will be called X-Video Bitstream Acceleration, still has yet to see the light of day.

Source -- http://www.phoronix.com/scan.php?page=n ... &px=Njk0MA

I hope this lights a fire under AMD. NVIDIA drivers are still binary-only, which just doean't work out that well for Linux.

dougz wrote:Does anyone have an feel for the price difference between the GeForce 9400M and the Intel 945GC? (The GS is supposedly rarely used because of its cost.)

Bit-Tech reports that the Atom + 945GC + ICH7 is $50. I assume they mean the Atom 220, but they list the ICH7M, so something isn't right. They also claim the TDP is 12W so this is higher power than 945GSE + ICH7M (9.3W). The author clearly has no real knowledge of the current Atom netbook platform, so this whole article is probably worthless.

dougz wrote:Very promising. It would make a heck of an upgrade for a new Mac Mini.

It would be a huge downgrade for the Mac mini. The mini has a Core 2 Duo processor, which even in its outdated form would blow away the Atom. The 9400M would add 1080p playback, something the mini's GMA950 can't do, but would otherwise not be an improvement.

it's more likely that the mini refresh will be some low end core 2 duo + 9400 to align with the recent macbook. I still like the Ion reference design, though. Each have their place.

The Task Manager picture in the Anandtech article shows two CPU charts, does that mean the reference design uses the dual-core Atom 330?

Pate

Pate wrote:The Task Manager picture in the Anandtech article shows two CPU charts, does that mean the reference design uses the dual-core Atom 330?

Pate

It's hyperthreading.

fyleow wrote:It's hyperthreading.

Right, of course, I forgot Atom has hyperthreading.

Pate

|Romeo| wrote:

jessekopelman wrote: Where do you see a heatsink? The pictures in the Anandtech article just show bare chips. I'm sure the Nvidia 9400M will have a heatsink in production versions, but a single-core Atom doesn't need one (although I bet production versions would go with a dual-core Atom).

Look harder

http://www.anandtech.com/GalleryImage.aspx?id=4819

That is the Ion with its heatsink, mounted on the daughter board. As I say, a typical Pico ITX heatsink covering almost the entire board.

Ok, I'm with you now.

|Romeo| wrote: I don't think so.

I am aware of the SODIMM slot being on the underside of the board in the Pico ITX specification. However if you look at the pictures of nvidias reference system design, you can see the Ion mounted on a daughter board (or "PICO ITX CARRIER CARD" as nvidia seem to be calling it). You can also see that DC in power socket is on the daughterboard, whilst the Ion has a standard Pico ITX power supply connector (if you can really call it standard, but regardless it's the 12 way 2mm connnector). However there is no such connector on the daughterboard for using wires (and why would you anyway).

Add to this all the connectors on the daughter board (VGA, audio, digital audio, more USB, the power button, the battery) and the fact that there is no connector on the top of Ion for a power button, power LED, reset button, etc. and it becomes very clear that there is a fairly substantial connector on the underside of the Ion in addition to the SODIMM slot.

Ok, again I'm fully with you now. For some reason, when I read the article I never saw any link to the gallery and its pictures of the populated box. Now that I've seen them, your comments make perfect sense. Sorry for being a bit flip in my original response.

Configuration files buried within some versions of Mac OS X show evidence that the next iMac and Mac Mini will indeed be based on the NVIDIA MCP79 chipset....

Amongst other benefits, the new notebooks have much improved graphics card capabilities, which make them more suitable to take advantage of OpenCL technologies coming in Snow Leopard....

A configuration file found in the Mac OS X version that ships with the new MacBook and MacBook Pros reveal entries referencing unreleased "MacMini3,1" and "iMac9,1" models.

Sources --
http://www.macrumors.com/2008/12/19/evi ... -chipsets/
http://www.appleinsider.com/articles/08 ... psets.html
http://www.electronista.com/articles/08 ... 400m.leak/

dougz wrote:

Configuration files buried within some versions of Mac OS X show evidence that the next iMac and Mac Mini will indeed be based on the NVIDIA MCP79 chipset....

Amongst other benefits, the new notebooks have much improved graphics card capabilities, which make them more suitable to take advantage of OpenCL technologies coming in Snow Leopard....

A configuration file found in the Mac OS X version that ships with the new MacBook and MacBook Pros reveal entries referencing unreleased "MacMini3,1" and "iMac9,1" models.

Sources --
http://www.macrumors.com/2008/12/19/evi ... -chipsets/
http://www.appleinsider.com/articles/08 ... psets.html
http://www.electronista.com/articles/08 ... 400m.leak/

Did anyone doubt this? I think the only question is what the CPU is going to be. Some have suggested that they will "step down" to a dual-core Atom, making the Mini pretty much a Nettop. I think they'll stick with mobile C2D, though. Now Ion, even as presented here with only a single core Atom, would make perfect sense for Apple TV . . .

QuietOC wrote:For those who think this is for HD video:

"The GMA's CPU utilization was about 5% higher with my MPEG2 clip, but otherwise CPU utilization was unchanged from the previous round of testing. There was also no difference in CPU utilization between the GeForce and GMA when playing back standard-definition DivX content, which pegged the CPU at roughly 15%.

This combination should be ideal for HD video. DivX and mpeg2 are irrelevant; H.264 and also VC-1 are the important codecs and should be playable with this integrated graphics with DXVA.

What front end are you guys planning on using with this little box?

Boxee and XBMC don't support DxVA acceleration. Those are both excellent front ends but this platform would be nearly worthless with that software. Or at least not any better than a Xbox. SageTV would be a good choice and maybe even VMC. Linux now has hardware acceleration for h.264 with Nvidia cards also.

autoboy wrote:What front end are you guys planning on using with this little box?

Early days. XBMC and Boxee aren't using VDPAU yet.

Here's mplayer using an Atom 330 and Sparkle 8400GS PCI (fanless):

I was looking for a thread to latch on to to say SUCCESS! I got my
Atom 330 board coupled with the 512MB Sparkle Fanless 8400GS PCI card
to play videos with VDPAU! See specs here:
http://www.mythtv.org/wiki/index.php/VDPAU#User_results

I have a bit more customization to do, but the goal will be to have a
solid state (sdhc card) with the OS, and all mythtv content will be
stored on the backend.

So far I am extremely impressed with the difference that VDPAU made.
My sample videos (hdhr recordings) wouldn't even play (100% CPU used,
locked up mplayer). However with VDPAU and myth trunk svn r19298 with
vdpau enabled, worked like a champ, using only 10-20% of the CPU.

This is very promising. I can't wait for 0.22!

Scroll down for cited results (Sparkle 8400GS PCI & Atom 330)

Source -- http://www.gossamer-threads.com/lists/m ... 176#362176

Note: More info on VDPAU (cards, drivers, Myth support) at MythTV Wiki -- http://www.mythtv.org/wiki/index.php/VDPAU