ada lovelace NVIDIA’s 4-slot Founders Edition GeForce RTX 4090 Ti/Ada TITAN graphics card has been pictured

[StDragon]

I find it a bit ironic that Nvidia (and other players it the GPU market) essentially undercut SGI (Silicon Graphics Incorporated) only to follow in their footsteps in form-factor. At some point, I could see Nvidia developing their own Onyx or Octane style stand-alone GPU workstation. Complete with their own PSU and water cooling loop in one case. At the moment, they've got rack mounted solutions however.

[leadeater]

5 minutes ago, StDragon said:

At some point, I could see Nvidia developing their own Onyx or Octane style stand-alone GPU workstation. Complete with their own PSU and water cooling loop in one case. At the moment, they've got rack mounted solutions however.

They do, Nvidia already sells workstations.

 

https://www.nvidia.com/en-us/data-center/dgx-station-a100/

 

[StDragon]

17 minutes ago, leadeater said:

They do, Nvidia already sells workstations.

 

https://www.nvidia.com/en-us/data-center/dgx-station-a100/

 

lol, admittedly that's new on me. But also comes to no surprise.

 

And the price is "Contact us". 

 

 

[leadeater]

15 minutes ago, StDragon said:

lol, admittedly that's new on me. But also comes to no surprise.

 

And the price is "Contact us". 

hah yup, it's not cheap. That's the second generation of that workstation, first one was water cooled rather than refrigerant.

[DildorTheDecent]

16 minutes ago, StDragon said:

lol, admittedly that's new on me. But also comes to no surprise.

 

And the price is "Contact us". 

A servethehome article suggests $199K for a single DGX A100.

 

[leadeater]

8 minutes ago, DildorTheDecent said:

A servethehome article suggests $199K for a single DGX A100.

 

That's a lot more expensive system. It has 8 A100 rather than 4 and very expensive PCIe switches. Think more around the $80k mark for this workstation, it could be less.

[Arika]

2 hours ago, StDragon said:

And the price is "Contact us". 

"If you have to ask, you can't afford it"

[Chiyawa]

Huh, how much power does RTX 4090 Ti use? I mean, with that massive cooling, I wouldn't be surprise if this thing can chug down 750W in burst mode.

[WhitetailAni]

10 hours ago, Chiyawa said:

Huh, how much power does RTX 4090 Ti use? I mean, with that massive cooling, I wouldn't be surprise if this thing can chug down 750W in burst mode.

The dual 12-pins (or maybe triple? who knows) will require enough power to cause a nation-wide power outage at max load. They will also cause your neighborhood to disappear in a giant explosion if you misalign the connector by a single micrometer.

 

Here at NVidia, we push the envelope of what's possible.

[Bitter]

Can't wait to see one of these on LN2. Only thing cooler would be if they had some on package HBM like the data center chips so you can just cool it all at once and get some insane bandwidth to match the core performance. I predict a 600W TBP before any OC.

[Senzelian]

I like how the guy put the card upright on the floor to take the picture. That really shows how tall the bloody thing is.

[RVRY]

I doubt Nvidia will be releasing this card anytime soon.

I assume they designed it just in case RDNA 3 was anywhere near competitive with the 4090 so marketing could still claim "the world's fastest graphics card is made by Nvidia".

[Bitter]

11 minutes ago, Senzelian said:

I like how the guy put the card upright on the floor to take the picture. That really shows how tall the bloody thing is.

I see flipflops and socks....is Linus the leaker?

[NumLock21]

So I'm seeing the ports are stacked up on one another. Most I've seen are 2 DVI ports or 2 HDMIs, but 4. Them soldering must be something.

[HenrySalayne]

4 hours ago, NumLock21 said:

So I'm seeing the ports are stacked up on one another. Most I've seen are 2 DVI ports or 2 HDMIs, but 4. Them soldering must be something.

But you do know the PCB is supposed to be flipped by 90° and thus parallel to the motherboard?

I don't know why the 3rd picture is nowhere to be found. That's allegedly the cold plate:

 

[leadeater]

1 hour ago, HenrySalayne said:

But you do know the PCB is supposed to be flipped by 90° and thus parallel to the motherboard?

I don't know why the 3rd picture is nowhere to be found. That's allegedly the cold plate:

 

Thing is this layout actually doesn't make any sense. So you have the PCB now flat parallel with the motherboard yet you still have to slot in to a PCIe slot... so the PCB has a 90 degree bend in it? That's seems like an unnecessary manufacturing challenge to me. Like sure the photo looks quite legit but it still makes no sense to me at all. It's a cooler designed more for an SXM slot than PCIe but it's too long for that and has fans which an SXM cooler would not.

 

Also the left hand side in this photo is where the video outputs are supposed to be but there doesn't appear to be an slots or cutouts or anything visible to accommodate that. Sure the photo has that part cropped out but there is just enough there to see that the fins go down past where the video outputs would be, and the cold plate appears to go all the way to the end as well.

 

Basically wtf, this is weird and will be real interesting if it ever comes out.

 

Or we are seeing the back side and it's rotated and the right hand side of the photo is where the video outputs are?

[HenrySalayne]

8 minutes ago, leadeater said:

Also the left hand side in this photo is where the video outputs are supposed to be but there doesn't appear to be an slots or cutouts or anything visible to accommodate that. Sure the photo has that part cropped out but there is just enough there to see that the fins go down past where the video outputs would be, and the cold plate appears to go all the way to the end as well.

It's a little weird to navigate this cooler assembly, but the cut-outs for the connectors (thus the back of the GPU) are on the right. Don't be fooled that you can see the cooler fins, it's the supposed top of the card (since there is no PCB), it's an all see-through / blow-through heatsink, probably symmetrically top to bottom.

19 minutes ago, leadeater said:

Or we are seeing the back side and it's rotated and the right hand side of the photo is where the video outputs are?

I missed the last sentence. Yes!

 

15 minutes ago, leadeater said:

So you have the PCB now flat parallel with the motherboard yet you still have to slot in to a PCIe slot... so the PCB has a 90 degree bend in it?

More of a 90° riser. Techpowerup (the source of the picture) is talking about several daughter-boards.

 

 

After all the design is more than questionable. The laser engraving on the back would be upside down, if the card is mounted in a normal computer case, And the stand-offs to screw the GPU die to the cold-plate are beyond the memory chips. That's extremely far away and I don't think we have ever seen anything like this.

[leadeater]

50 minutes ago, HenrySalayne said:

More of a 90° riser. Techpowerup (the source of the picture) is talking about several daughter-boards.

Doesn't matter what it is PCIe devices are secured and held up via the slot so whatever they are doing the slot must support the cards weight and it has to do it from the edge. So  the only practical way I can see that working is that it's a PCIe connector soldered through the board, likely with reinforcing around it which might mean it sits a bit higher than a normal card and will be slightly offset from the slot with a tiny amount of overhang above the slot.

 

[HenrySalayne]

9 minutes ago, leadeater said:

So  the only practical way I can see that working is that it's a PCIe connector soldered through the board, likely with reinforcing around it which might mean it sits a bit higher than a normal card and will be slightly offset from the slot with a tiny amount of overhang above the slot.

We could see some form of contact springs while the PCIe connector is just a tiny PCB with contact pads on the back. It's hard to say if this would satisfy signal integrity concerns, but the new CAMM module does a similar thing.

If the entire PCIe connector turns out to be just a 3 cm deep PCB and it is securely screwed into the frame of the graphics card, I don't see any structural issues. It's actually a better solution than larger PCBs, because the shear force is the same but the bending force is much lower. So it's much less likely we see any cracks forming in the PCIe connector and it might be easily replaceable.

 

But who knows if this "leak" is actually genuine.

 

10 minutes ago, leadeater said:

indeed!

[Heliian]

Will it run forspoken though? 

 

I guess this chonky mf is what happens when you no longer lean on sli.

[leadeater]

1 hour ago, HenrySalayne said:

We could see some form of contact springs while the PCIe connector is just a tiny PCB with contact pads on the back. It's hard to say if this would satisfy signal integrity concerns, but the new CAMM module does a similar thing.

The issue is the weight, not figuring out how to deliver a PCIe connection and signal. Today right now you have a PCB that is vertical and ridged from the slot and is supported by the cooler itself together offering the strength required.

 

Now you have a board, riser or anything otherwise coming out of the PCIe slot and must connect to the the cooler but does not go up the back of the cooler? Where is the structural rigidity and support coming from. ~2kg being supported on a few mm pivot point, yea not happening. The slightest jostle and that's going to snap right off.

 

1 hour ago, HenrySalayne said:

If the entire PCIe connector turns out to be just a 3 cm deep PCB and it is securely screwed into the frame of the graphics card, I don't see any structural issues. It's actually a better solution than larger PCBs, because the shear force is the same but the bending force is much lower. So it's much less likely we see any cracks forming in the PCIe connector and it might be easily replaceable.

Not is very clearly much worse. If its a separate PCB and slots in vertical like as is now then you have the issue of securing it to the cooler. If it's horizontal with a soldered in PCIe slot connector then you're putting on lot of stress on solder joints and those will 100% fail within the usable life of the card.

 

There is a damn good reason why PCBs are the way they are right now and do such a good job, single complete board with a very large area securing it to the cooler giving it a lot of strength. The PCI bracket screws are for stopping it sheering in the slot, this will not protect either of the above options.

 

You simply need a non-trivial amount of contact area between the PCIe connector and the cooler to get enough strength and if the back of the cooler is completely exposed then where is that support coming from?

 

Edit:

And to be clear my concern come from the total weight of the card and it's length.

[HenrySalayne]

1 hour ago, leadeater said:

You simply need a non-trivial amount of contact area between the PCIe connector and the cooler to get enough strength and if the back of the cooler is completely exposed then where is that support coming from?

I think you have the wrong idea.

The tiny part of the PCB of the PCIe connector inside the PCIe slot has to transfer all the forces from the card into the slot. Therefore the other end of the PCIe connector attached to the graphics card doesn't need to be any bigger. A PCB that's screwed down or clamped into a metal structure (like the frame of the GPU) can transfer much higher forces than the puny PCIe slot could even dream of.

1 hour ago, leadeater said:

Not is very clearly much worse. If its a separate PCB and slots in vertical like as is now then you have the issue of securing it to the cooler. If it's horizontal with a soldered in PCIe slot connector then you're putting on lot of stress on solder joints and those will 100% fail within the usable life of the card.

I don't think that's much of an issue, since PCBs are pretty flexible. You just need to secure the PCB before the solder joints (so the forces will not flow through the solder joints).

1 hour ago, leadeater said:

There is a damn good reason why PCBs are the way they are right now

Money!

A flat PCB with as few connectors as possible is simply the cheapest option.

[leadeater]

10 hours ago, HenrySalayne said:

The tiny part of the PCB of the PCIe connector inside the PCIe slot has to transfer all the forces from the card into the slot. Therefore the other end of the PCIe connector attached to the graphics card doesn't need to be any bigger.

It does need sufficiently large contact area. think about the forces involved of something around 2kg with movement, vibration and shock. It might be fine in isolation but computers get moved and shipped. They don't live in space under zero gravity and even then inertia still applies.

 

That connection point between any such PCB simply cannot be 10mm with only a few screws and not the entire length or good portion of the entire length of the card. The graphics card will snap that PCB very easily at either the screw points or potentially entirely.

 

Think about it logically. Get a thick piece of plastic or wood and a metal block or thick plate. Secure it to the wood using screws of glue starting at 10mm contact area and then waive this around. 100% the metal block aka the weight is coming off one way or another. Repeat until it doesn't, you'll find the contact are between the two materials to prevent this is non-trivial.

 

10 hours ago, HenrySalayne said:

You just need to secure the PCB before the solder joints (so the forces will not flow through the solder joints).

Again in isolation yes, in reality no. No matter what you do the slot connector and solder points will be weight/load bearing and over time those joints will get put under a lot of stress from that and movement and will 100% break. This is why in devices like UPS's connections between boards are either cables or solid metal bars with screws. They do that because it's actually a problem. Solder joints simply are not load bearing and can never be.

 

While you could use a plastic standoff or some other kind of reinforcement like I mentioned that will not remove movement and vibration forces, only compressive. Have a think about it, when is the last time you saw solder joints being used in any kind of load bearing capacity when decent weight is involved.

[CommanderAlex]

I guess in the Show off your Setup! thread, we'll have to start showing off our chunky double pole breaker to supply the amps to this. 

[CommanderAlex]

Also, it appears PG137 has made it on a shipping manifest. 

https://videocardz.com/newz/nvidia-pg137-titan-ada-gpu-with-48gb-memory-spotted-during-shipping

@BiG StroOnZ