An observation about ATi's physics demonstration

MrNasty

Limp Gawd
Joined
Apr 11, 2006
Messages
221
Look here

Look at this:

1951physic3.jpg


Now look at this:

1951physic2.jpg


Notice the x1600 looks as though it's jutting out (check the PCI retention bit) - it looks as if it's just wedged into the gap between the two x1900's.

Also notice the fact that the second x1900 is not connected via the dongle in xfire but the x1600 is.

Is it just me or is this a really weird configuration?

UPDATE


Here's a shot of ATi's x1900 physics demo - spot the different mobo and connection method?

For lazy clickers:
ati-meatstack.jpg


No difference in height with these cards. If that x1600 was connected, why didn't they just use this mobo? What aren't we being told?
 
That is weird... maybe there is no PCIE slot there and they made it look "real"


this would be worrysome for their claims if this is true.
 
MrNasty said:
Notice the x1600 looks as though it's jutting out (check the PCI retention bit) - it looks as if it's just wedged into the gap between the two x1900's.

Also notice the fact that the second x1900 is not connected via the dongle in xfire but the x1600 is.
Another monitor seems to be plugged into the X1600. The XF dongle looks like it's just hanging (the thick cable).

If the system was running when that picture was taken I think the X1600 is on some kind of slot adapter. Maybe the slot between the XF cards was x1/x2/x4 sized and needed an adapter to plug in a x16 size card. I would have just chopped the end off the slot. ;)

(edit) I think it was just badly worded:
The system features three video cards, two X1900 and one X1600. However, the Cross X1900 was hooked into the X1600 and the remaining X1900 wasn't connected to the system. Even without CrossFire enabled, the demo was extremely impressive.
The picture does show that the XF dongle isn't connected, but the X1600 isn't "hooked" into the X1900. That's a separate video cable. I wonder which card the physics were running on though. It may have been on the second X1900.

Another angle from the [H] article: http://enthusiast.hardocp.com/image.html?image=MTE0OTU2MjEzNkxLSnh5NWZGZzRfMV84X2wuanBn

Yes, it's likely an adapter between the slot and card because the middle slot is x1: http://www.hkepc.com/bbs/itnews.php?tid=608109
 
Still strange, check it:

If it only needed a 1x slot to do HW physics why don't ATi just make an adaptor for all the 1x boards that are already out there? Why make a whole new 3 slot mobo?
 
MrNasty said:
Still strange, check it:

If it only needed a 1x slot to do HW physics why don't ATi just make an adaptor for all the 1x boards that are already out there? Why make a whole new 3 slot mobo?
Read the ATI throws down the gauntlet thread/article in the ATI forum. ATI wants a bigger chipset market share, so they are going to push their own solutions. :p

A x1 PCI-E slot has about 4x the sustained bi-directional bandwidth of a PCI slot: 250MB/s sustained in each direction at the same time (500MB/s total) vs ~115-120MB/s total usable sustained R/W bandwidth on PCI.
 
I have done :) - it still seems they could do this with their current chipset, which would add value to those who already have them and at the same time pimp them to those that don't (not as if they would enable it for nForce boards :p lol) - thus increasing their share with current AND next gen products, as opposed to next gen products only...

While I don't really like the idea of buying a graphics card for physics myself, obviously not having to buy a new mobo would be an incredible plus point for most...

edit: the bandwidth wasn't really an issue - a 1x slot is still damn fast :) it was more a question aimed at whether the card would actually work full stop in a 1x slot - cos as you know PCIx is upwards compatible, but I've never looked at downwards compatibility...
 
MrNasty said:
While I don't really like the idea of buying a graphics card for physics myself, obviously not having to buy a new mobo would be an incredible plus point for most...
I just want to see how Havok FX shakes out vs PhysX hardware. The GPU manufacturers are just getting annoying now.

They have been hyping it even before Havok FX was finished and so far have even not provided a public demo since Havok FX was released. That probably means that driver support isn't really ready and that's the annoying part. They don't "have" a better solution available if it's not ready yet.
 
Wait, lemme get this straight. The X1900 and X1600 were in crossfire, and the other X1900 was processing physics? Yet ATI claimed the physics were being done on an X1600. I smell foul play somewhere... :mad:
 
pxc said:
I just want to see how Havok FX shakes out vs PhysX hardware. The GPU manufacturers are just getting annoying now.

They have been hyping it even before Havok FX was finished and so far have even not provided a public demo since Havok FX was released. That probably means that driver support isn't really ready and that's the annoying part. They don't "have" a better solution available if it's not ready yet.

I can solve that problem for you:
Name one title that supports ATI/HavockFX eye-candy physics? ;)

Terra - I havn't heard a single developer talk about them using in a game...
 
InorganicMatter said:
Wait, lemme get this straight. The X1900 and X1600 were in crossfire, and the other X1900 was processing physics? Yet ATI claimed the physics were being done on an X1600. I smell foul play somewhere... :mad:

The Eye Candy Wanna-be Physics has been filled with PR-Bull from day one...

Terra - Sad part is how many belives the PR-Bull and just accpt it... :rolleyes:
 
Just saw the demo.
Did any notice this part:

Q: When will it be available?
A: Late this year, maybe early next year.

So come 2007 this is out(maybe)...but still not a SINGLE game in sight :rolleyes:

Terra...
 
Terra said:
I can solve that problem for you:
Name one title that supports ATI/HavockFX eye-candy physics? ;)
Havok FX was only announced in March and release was just this quarter. I have a problem with the hype, not whether or not it's going to be used. Wasn't it a bit longer than 2-3 weeks after the rebranded (and non-hardware accelerated) PhysX SDK was released before games started appearing? It was more like over 6 months.

But if that's the game you want to play, here's one: Hellgate: London, Flagship Studios (former Blizzard north employees).
“With Havok FX we can explore new types of visual effects that add realism into Hellgate: London. Given the widespread installed base of GPU’s and the incredible performance of the new NVIDIA GeForce 7900 boards, Havok FX was a natural choice."
-Tyler Thompson, Technical Director, Flagship Studios
 
MrNasty said:
Still strange, check it:

If it only needed a 1x slot to do HW physics why don't ATi just make an adaptor for all the 1x boards that are already out there? Why make a whole new 3 slot mobo?
A more interesting question is this:
When PhysX first was announced the ATi/ nVidia people said that a separate physics card would require immense bandwidth to the graphics cards in order to tell them where to render the vertices. The amusing part is that now they are saying "we really don't need all that bandwidth after all".
 
pxc said:
Havok FX was only announced in March and release was just this quarter. I have a problem with the hype, not whether or not it's going to be used. Wasn't it a bit longer than 2-3 weeks after the rebranded (and non-hardware accelerated) PhysX SDK was released before games started appearing? It was more like over 6 months.

But if that's the game you want to play, here's one: Hellgate: London, Flagship Studios (former Blizzard north employees).

HavokFX got 1 game(2 actually)?
AGEIA got +100.
If I where to use geek terms I would say +100FPS beats 2FPS :p

Terra - Look BEYOND the PR...
 
drizzt81 said:
A more interesting question is this:
When PhysX first was announced the ATi/ nVidia people said that a separate physics card would require immense bandwidth to the graphics cards in order to tell them where to render the vertices. The amusing part is that now they are saying "we really don't need all that bandwidth after all".

I guess they changed PR-tatics...

Terra - Wonder what they will say next :rolleyes:
 
Terra said:
HavokFX got 1 game(2 actually)?
AGEIA got +100.
Out of those "+100" NovodeX/PhysX games in development, only a handful will probably be published*. Few of those will do anything beyond eye candy and CPU physics for a good reason: PPU cards do not have a large enough installed base and the current adoption rate is small.

* Only "licensed developers" are allowed to download the SDK. I'll bet a good portion of the list is hobbyists who just wanted a free sdk. I have done things like that before. :p The available + announced games list is only 23 items long after 8 or more months. Havok FX has only been released for a couple of weeks.
 
I updated my original post - compare the x1600 testbed to the x1900 test bed. There are no dongles in sight at all on that new photo. Why didn't ATi use this board with the x1600 test? why does that board need to have the x1600 connected at all?

What is up here? I smell fish now. Lots of fish.
 
MrNasty said:
I updated my original post - compare the x1600 testbed to the x1900 test bed. There are no dongles in sight at all on that new photo. Why didn't ATi use this board with the x1600 test? why does that board need to have the x1600 connected at all?
Those are 2 different motherboards. I can see several component differences. The first one is the older model I posted a link to and the new one you posted seems to be one with 3 real x16 slots. That's not really the problem.

I also think there's something rotten about the "X1600" demo. My first thought was only XF physics was working (one X1900 rendering and one X1900 for physics), as Havok announced months ago (i.e. maybe ATI drivers weren't working with XF + a 3rd card... in "2+1" mode). But if the 3 X1900 system was also running a demo, it might be because the demo didn't run smoothly on the X1600 on the first system pics you posted.
 
perhaps, its just a hoax... maybe the 1st x1900 is rendering, while the secon x1900 does the Havok FX, and the x1600 is just sitting there...

or perhaps its a propriotary setup for havok fx. maybe it requires the dongle for the FX and they are running CrossFire without a dongle? I dont even have an interest in those over $300 video cards... I'm not even considering an x1800... so I dont know if thats possible on x1900's
 
pxc said:
Those are 2 different motherboards. I can see several component differences. The first one is the older model I posted a link to and the new one you posted seems to be one with 3 real x16 slots. That's not really the problem.

But they are both from the demonstrations at computex - check the dates out...

If they are 2 different boards (which is plain to the eye) for 2 different markets etc. and there was a reason to be touting both fair enough but as far as I can see there isn't one. Also alarmingly absent from the demonstration was ATi's 2 card solution and validation for the 'great if true' free overhead on one card physics processing.

All this just adds up wrong and it would be really nice to have an explanation from ATi now, I think.
 
Ah, mystery solved! I couldn't exactly remember what board the second one was but now I know.

The first board is a XPRESS 3200 board with a x1 x16 x1 x16 PCI-E slot layout (as I mentioned above) and the second board is an Intel Bad Axe with 3 x16 size slots (x8 x8 x8?). It's a bit silly to ask why ATI would want to show the XPRESS 3200 motherboard. That should be obvious.

I still suspect the first system was running physics on the X1900. Why would they disable XF and plug a monitor cable into the X1600 otherwise? It could be a driver problem but this is still just hot air since they don't release the demos to reviewers or the public.
 
Back
Top