PDA

View Full Version : Quadro FX3000 mini-preview (by Baron)


The Baron
07-22-03, 12:51 PM
So, some of you have probably heard of the Quadro FX3000 that was announced today. I was going to write a little preview bit about it, but, of course, my motherboard died and prevented me from accessing everything I wrote down during the conference call and the presentation I was sent. So, going from memory, here's the account of the call.

It was on July 11, and I was on the phone with Charlie Rasch, the Quadro product line manager, and Jerry something-or-other (I want to say Orbach, but that's just the guy from Law and Order, and for the life of me I cannot remember the guy's name), the Quadro PR manager.

Anyway--the FX3000 press release is here. (http://www.beyond3d.com/forum/viewtopic.php?t=6990)

And now for random things I remember:

It's based on the NV35 core, but the FX2000, based on the NV30 core, is still in production.

Core/memory speeds are 400/425; even with the lower speeds than the FX2000, SPEC benchmarks were significantly higher.

They're not designed to replace the current Quadro line--they're designed to basically let NVIDIA enter the visualization/powerwall market (more on that in a bit).

Two models: FX3000 and FX3000G. The FX3000 has dual DVI outputs (one, I believe, is called a double DVI output or something like that?), occupies two slots, has a slightly modified version of the FX2000 cooler (it's apparently quite a bit quieter, while cooling the same amount, and of course there's no DDR2 to get absurdly hot), and lists for $2295.

FX3000G adds a framelock and a genlock. Genlock allows the card to sync to a video signal, and the framelock allows multiple cards to work in tandem by attaching Cat 5 cables to the cards. I believe they said that the bandwidth is great enough to allow 16 cards (in different systems) to work together at the same time.

So... the framelock is the interesting bit. It synchronizes rendering--however, I don't believe that it is the equivalent of SLI. I BELIEVE that every card still outputs to its own DVI monitor (meaning, if you have 5 cards, you have 5 displays). But, maybe Uttar wasn't too far off with his whole "NV3x core has 3dfx-style SLI elements" essay.

I mentioned that the FX3000 line was designed to put NVIDIA in the powerwall market (you know, those huge projected displays). The big way they're going about this is what they call a "single system powerwall."

Basically, instead of having the graphics card send an absurdly high resolution image to a device which divides the output among multiple projectors, the card will split the signal itself and allow you to use 2 random projectors and display a much higher resolution. Falls into the "Dude, why didn't I think of that?" category.

Jerry and Charlie emphasized that the card would be of great benefit to markets that extensively used volumetric rendering (primarily oil industry and medical imaging). Charlie stated that the increased memory bandwidth (256-bit bus) would be extremely beneficial. However, he also hinted that a move to PCI Express would be coming--I would expect the Quadro NV40 card (and possibly also the NV40) to appear in PCI Express form (the NV40 would probably also appear in AGP form, but we'll see) and that the nForce 3 boards will have PCI Express support by the end of the year. (Possibly nF3 Pro 250?)

NVIDIA is still pushing Cg pretty hard, and it looks like it's gaining support in the workstation market since it works with OGL.

Now, the stuff that just came up at the end during the little Q&A session:

With regards to NV36--they would not talk about "future products" but I think it was Charlie who said that it would make sense to have an NV36 Quadro.

From the sound of it, nF3 Pro 150 will be out in both an Opteron and Athlon64 form by the time Athlon64 is released.

There should be a big tasty announcement at SIGGRAPH (it made me EXTREMELY happy, but the info is not from NVIDIA and, as far as I know, is not yet public, so I can't share specifics).

When they asked how NVIDIA was viewed in the enthusiast market, I mentioned that the driver problems were taking a toll on its reputation. Happy, Dig? :p

I'm sure there's stuff I'm forgetting, and I'll try to have a real preview up sometime.

Overall, it was an extremely insightful experience--thanks Charlie, Jerry, and Susan Austin (former Quadro PR manager who set the whole thing up). I look forward to hearing from NVIDIA again... ;)

Hanners
07-22-03, 12:57 PM
Originally posted by The Baron
...the framelock allows multiple cards to work in tandem by attaching Cat 5 cables to the cards. I believe they said that the bandwidth is great enough to allow 16 cards (in different systems) to work together at the same time.

Well, that gets my vote for 'cool idea of the week'. :D

Now, someone tell me why I can't hook up my 9800 Pro directly to my laptop via Cat 5 to play games? ;)

dlee96
07-22-03, 06:36 PM
Does anyone know if/when Linux drivers will be available for the genlock capabilities on the 3000G?

-dlee

The Baron
07-22-03, 06:51 PM
I was informed that Linux drivers and some sort of an API would be available for the card, the framelock, and the genlock.

DSC
07-23-03, 01:37 AM
Baron,

The QuadroFX 3000(400/425) is not clocked lower than the FX 2000(400/400).

Will Nvnews have a full preview/review on the FX3000/3000G? :D

The Baron
07-23-03, 01:39 AM
Will Nvnews have a full preview/review on the FX3000/3000G?
Review? Only if they want to provide me with a Xeon, software, couple of projectors, and other random stuff.

Preview? Yeah, once I get my normal computer back. Blah.

And doh, see, this is why I need my notes. ;)

GlowStick
07-23-03, 02:10 AM
i want to review a powerwall.

i am really impressed of this 'attach cat 5 to video card' dose this mean that the video card has a cat5 port on it?!!?

I hope it uses regular network protocalls so you can use it with routers, not just that the card happens to use cat5 cables.

I love networking, its so fun!

ricercar
07-23-03, 05:26 AM
Originally posted by GlowStick
idose this mean that the video card has a cat5 port on it?!!?

I hope it uses regular network protocalls so you can use it with routers, not just that the card happens to use cat5 cables.

The FX 3000 has two RJ45 ports. I heard from a NVIDIA demo developer that you'll blow your router/NIC if you plug the FX 3000 into a CAT5 network, so it appears to have power as well as data on the interconnects.

Also heard Friday night that some folding at home people have a large number of NV35 GL cards linked and custom-programmed to do the folding calculations in OpenGL or something. This parallel processing group of NV35GL GPUs are the 12th fastest supercomputer involved in the Folding project. Expect an NVIDIA press release on this soon.

The Baron
07-23-03, 09:39 AM
The FX 3000 has two RJ45 ports. I heard from a NVIDIA demo developer that you'll blow your router/NIC if you plug the FX 3000 into a CAT5 network, so it appears to have power as well as data on the interconnects.
Yup. It would blow out a network card or router; however, they've apparently added some protections against that from happening.

Bert
07-23-03, 01:38 PM
Originally posted by The Baron
So... the framelock is the interesting bit. It synchronizes rendering--however, I don't believe that it is the equivalent of SLI. I BELIEVE that every card still outputs to its own DVI monitor (meaning, if you have 5 cards, you have 5 displays).

Yep, that's what it is. If you run a seamless projection (two or more projectors side by side) then even if you get it to lineup perfectly, you will see a seam. This is because the cards are out-of-sync, so one card may be a frame behind. Very disturbing.

It's described in this PDF (http://www.quantum3d.com/PDF/whitepapers/channelsync_wp.pdf) , for example. Until now, Quantum3D was the only one offering genlock/framelock for NVIDIA chips.

And it's not only for projection: I have a monitor with 4 parallel DVI inputs that displays 4 tiles, each one driven by one DVI output. It really helps to up the fill-rate when you run at 3840x2400 pixels. But if the signals are not synced, the seams are visible. Very disturbing.

Btw, does anybody know a non-Quadro FX with dual DVI output?

The Baron
07-23-03, 01:43 PM
Originally posted by Bert
And it's not only for projection: I have a monitor with 4 parallel DVI inputs that displays 4 tiles, each one driven by one DVI output. It really helps to up the fill-rate when you run at 3840x2400 pixels. But if the signals are not synced, the seams are visible. Very disturbing.
/me drools

It mentioned 3840x2400 in the presentation.... Ohmygod. I want.

Too bad it's 9 grand for the monitor and another 3 grand for the FX3000G ;)

prodikal
07-23-03, 04:00 PM
Originally posted by The Baron
Review? Only if they want to provide me with a Xeon, software, couple of projectors, and other random stuff.


lol ask the guy in this forum iwht dual 3.06 xeons and 6 gig of ram to over and test it :D

GlowStick
07-23-03, 06:52 PM
im really dissapointed that they wont plug into a network grrr!

but at least, they use cables that are CHEEP and readily avalible, so you dont have to buy propritory **** that will stop being produced in like a year or something etc.

wilson
07-24-03, 01:37 AM
Originally posted by Bert

And it's not only for projection: I have a monitor with 4 parallel DVI inputs that displays 4 tiles, each one driven by one DVI output. It really helps to up the fill-rate when you run at 3840x2400 pixels. But if the signals are not synced, the seams are visible. Very disturbing.

Only 2-stripe mode works with non-genlocked source right now. 4-stripe mode requires genlocking.

Btw, does anybody know a non-Quadro FX with dual DVI output?

Quadro4 980XGL is officially supported by IBM. ATI FireGL X1 supposedly can do 30Hz in 2-stripe mode. Matrox has 2 special-version Parhelia for the T221.

Bert
07-24-03, 07:33 AM
Originally posted by wilson
Only 2-stripe mode works with non-genlocked source right now. 4-stripe mode requires genlocking.

Yep, we had not have success yet at the 4-stripe mode. Still hoping for the new firmware that was promised for "summer". It also is supposed to support the tile-mode (2x2@1920x1200).

Originally posted by wilson
Quadro4 980XGL is officially supported by IBM. ATI FireGL X1 supposedly can do 30Hz in 2-stripe mode. Matrox has 2 special-version Parhelia for the T221.

We're running with the 980XGL at the moment, which works surprisingly well. It's like ~20 Hz in 2-stripe mode, and we even solved the tearing problem between the two halves, which was still very much noticable. I have a nice 3D app that is tuned for low fill-rate which runs at 20fps fullscreen easily :D. I doubt X1 can do much better then 20 Hz because that's just the limit of dual DVI :|. In quad DVI it's supposed to be 43 Hz.

We had the Matrox G200MMS in there initially. Sh*tty performance (although we even did 3D via VNC :D), then had very lengthy troubles of getting DVI cables instead (they're harder to get than gold dust, and as pricey :wtf: ).

wilson
07-24-03, 01:59 PM
Originally posted by Bert
YWe're running with the 980XGL at the moment, which works surprisingly well. It's like ~20 Hz in 2-stripe mode, and we even solved the tearing problem between the two halves, which was still very much noticable. I have a nice 3D app that is tuned for low fill-rate which runs at 20fps fullscreen easily :D. I doubt X1 can do much better then 20 Hz because that's just the limit of dual DVI :|. In quad DVI it's supposed to be 43 Hz.

The X1 has been verified to drive the T221 at 30Hz in 2-stripe mode. The maximum refresh in 4-stripe/tile mode is 41Hz.

Have you tried changing the EDID of the monitor to do 25Hz? Most likely your monitor is set to 20. I have done 25Hz using the Gainward 4600 w/dual-DVI and SoftQuadro.

This doc tells you how: http://www.idtech.co.jp/en/support/pdf/EDID-EOa_13.pdf

Also, try the 980XL bios from IBM's site.

The QFX2K has no tearing problem although overlays don't always work when straddling the 2 halves. I can use a card with better fill rate.

Bert
07-24-03, 05:39 PM
Interesting. Do you know how ATI gets a 50% bandwidth improvement on the DVI channels? Sounds like magic ;-)

Thanks for the EDID tip. Haven't seen this at IBM's site. Actually, what is the relation of IDTech and IBM? Are you affiliated with either?

wilson
07-24-03, 07:12 PM
Originally posted by Bert
Interesting. Do you know how ATI gets a 50% bandwidth improvement on the DVI channels? Sounds like magic ;-)

Not magic, just good engineering. ATI uses DVI drivers that meet/exceed the single-link DVI spec. 30Hz at 1920x2400 + blanking is very close to 165MHz.

Thanks for the EDID tip. Haven't seen this at IBM's site. Actually, what is the relation of IDTech and IBM? Are you affiliated with either?

IDTech is a joint venture of IBM and a Taiwanese LCD manufacturer named Chi-Mei. IBM Research developed the "Bertha" and IDTech manufactures the display. I assume Chi-Mei contributed some manufacturing know-how.

I'm not affiliated with any of them. Just a satisfied customer. I run a list called IBM_T2X_LCD on Yahoo Groups.

Bert
07-25-03, 09:04 AM
Thanks again! I just subscribed to that list and browsed through the archives. Nice info in there. Maybe you should get IBM to put a link up from their drivers page :cool:

TriangleMan
08-07-03, 11:06 PM
:confused:

I haven't seen anything about an anticipated date. What does NVIDIA's release mean?

That they are:
1. Desigining it
2. Going to design it.
3. OEM's should already be making it?

Really, I want to know where and when I can buy several.

TM