Would need a new connector spec to support the weight. CPU socket mounts are way sturdier than PCIe slots.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Videocardz: NVIDIA GeForce RTX 3080 pictured? (now thats different)
- Thread starter Snowdog
- Start date
Would need a new connector spec to support the weight. CPU socket mounts are way sturdier than PCIe slots.
BinarySynapse
[H]F Junkie
- Joined
- Feb 6, 2006
- Messages
- 15,103
You would support the card with the case, not the PCIe slot.
jeremyshaw
[H]F Junkie
- Joined
- Aug 26, 2009
- Messages
- 12,511
It is basically what Nvidia has done with their SXM form factor in their DGX servers (3-4 generations already, starting with Pascal), and somewhat with the MXM form factor for laptops (since spun off as its own industry standards organization, and basically dead now). OCP has done something similar, but mostly for NIC.
First picture is a DGX server with 8 Nvidia GPUs, second picture is one of the GPU cards (different generation).
Connector is a dense mezzanine on the backside. All power goes through the two main connectors on the backside.
Last edited:
It wouldn't need to be supported by the PCIe slot. The edge connector would deal the electrical connection. Since the card is lying flat you just need support under the rest of the card.
If you stick with PCIe (using an edge connector) then you can still use standard GPU cards that exist today.
Well look at that, lol. Solution already exists, just need to bring it to the masses.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,315
Reverse the GPU, add a 90 degree connector, connect it to the top of the motherboard, and mount it directly to the case above the motherboard with its own cooling system the same way you do the CPU.
The problem with coming up with a solution isn't hard, the problem is getting it done 5 years ago and making it the future standard back then.
The problem with coming up with a solution isn't hard, the problem is getting it done 5 years ago and making it the future standard back then.
jeremyshaw
[H]F Junkie
- Joined
- Aug 26, 2009
- Messages
- 12,511
Well, in addition to what I posted above about Nvidia's work stretching back to Pascal (GP100 in particular), OCP is also trying to make something similar to Nvidia's SXM form factors, but with a somewhat more fixed spec. Nvidia has been a part of OCP since 2017, and Mellanox since 2012. Though I think Nvidia proper is only interested in the chiplets part of OCP and Mellanox on the NIC part of OCP, neither on the SXM competitor. The group that is developing the SXM competitor is basically just Facebook, MSFT, and Baidu.
Comixbooks
Fully [H]
- Joined
- Jun 7, 2008
- Messages
- 22,012
I hope they have a few million produced at launch the thing is going to be in such short supply. Unless they have in been in production for months.
I mean you can't even buy a PSU let alone find a 2080ti for sale unless it's used. Or the price is going to be about 2K per card.
Last edited:
Eh? Plenty of 2080ti around msrp out there. You can even get reference ones on Nvidia's site.
Armenius
Extremely [H]
- Joined
- Jan 28, 2014
- Messages
- 42,084
That definitely gives more explanation as to how the heatsink assembly is supposed to work.
Odd. It looks like angled fins in the middle of the card may be blocked from fan air from either fan. They just look like passive fins.
The part of of the HS closest to the monitor connectors, looks fairly traditional, with heat-pipes running to where the 2nd fan cools the second fin set. Here were it gets different. The fan is on the other side, and appears to blow through the heat sink. The blowing through the heat-sink makes sense, but it actually seems disadvantageous to put it on the backside. Blowing through from back to front, pushes the hot air under the card, trapping it where it may be pulled into the front side fan.
If they put the fan on the front with the blow through design, then it would help ventilate the hot air trapped under the card.
Anyone see something different?
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,315
Nope. It's one pass-through cooler inside the case and one semi-exhaust fan that blows down on the chip (up in the case) with some of the air going out the back.
Armenius
Extremely [H]
- Joined
- Jan 28, 2014
- Messages
- 42,084
I think you have it backward. The fan on the bottom of the card is intake and the one on the top is exhaust.
I hope you are correct. But looking at the way the "top" fan blades are scooped, it looks more like a clockwise turning intake, than it does a CCW turning exhaust.
Armenius
Extremely [H]
- Joined
- Jan 28, 2014
- Messages
- 42,084
Every PC fan I've worked with has rotated counter-clockwise. It would be odd if it was the opposite in this case.
Gigabyte has a current gimmick with counter-rotation fans. What I am looking at in fans, is the which way the scoop points. Picture the blades as the scoop part of a spoon. Every fan I have seen have seen rotates such that the concave scoop catches the air, not the convex side pushing it. Here is the Gigabyte, with one rotating clockwise, and one rotating counter-clockwise(both rotate in the direction of the scoop):
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,315
Armenius
Extremely [H]
- Joined
- Jan 28, 2014
- Messages
- 42,084
Look closer at the fan on the top part of the card. I think the lighting is confusing you. The top of the blades are concave if you look a little closer. Look at the blades on the top-right quadrant of the fan.
I'm all in on that one. My desktop computers I've built recently are hugely quieter than the days of 80mm two wire fans.
That would solve a lot of problems, but we only get what the makers of these products decide we get. What baffles me is the GPU can consume more power than the CPU, but runs with a smaller heat sink. Does the GPU make less heat or does it just have to tolerate more. Seems to defy physics.
Grebuloner
[H]ard|Gawd
- Joined
- Jul 31, 2009
- Messages
- 2,043
I noticed the same thing, then went agoogling for comparisons. The 2080ti (among other custom cards) has a big section of fins in the center that aren't blown on by the fans. They're vertical and hidden under the shroud so not only are they not getting much air over them, there really isn't space for convection, either. At least with the 3080 cooler, the fins are exposed which should add some real potential to them, even if up-side down.
On the other hand, I'm still having trouble on where the power connectors are. Why can't we have shots of the card rear? (would also help with that 300+W rumor...). There's that gap in the middle fin stack, but it just doesn't look big enough for two 8-pin connectors to fit...
Azrak
[H]ard|Gawd
- Joined
- Oct 4, 2015
- Messages
- 1,114
I don't think there is any air flow between the two fans on the heatsink. They are independent of each other. One blows through the heatsink - you can see through it in the latest picture. The other fan blows on top of the GPU and heatsink and onto the PCB.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,315
CPU die size is typically a lot smaller, so the heatsink has to do a lot of work to cool a smaller patch. They do run at different temps, usually, too.
Yeah fans always turn in the direction of the concave scoop. Anything else wouldn’t make sense for moving air.
It’s not clear to me either why both fans aren’t on the same side with the second one blowing air up toward the CPU and exhaust fans in a typical case.
Maybe it’s easier to route the heat pipes with the offset config.
Sounds reasonable, has to be something making a difference.
Which is exactly what I said. Though there seems to be some disagreement about which direction of the top fan is concave.
Not sure how. The concavity of the edge of the blade, follows the concavity of the scoop of the blades, and indicates a clockwise intake fan.
Edit. Image to make it clear, fans rotate in the direction, of the focus of the concave. The fan near the i/o panel rotates counter-clockwise, and is an intake. The concave for the other fan points in the other direction, and thus rotates clockwise, and is also an intake.
Though, it does seem odd that if they are both intakes, why not use the same blades and turn in the same direction. Maybe it's a temporary blade set and it really is an exhaust, but so far the blades indicate intake.
Attachments
Last edited:
Yes I was agreeing with you.
Looks real to me. Guess we're getting them a little sooner than expected! Wonder how long the card is.... I can only fit 11".
Thats what she said and it greatly disappointed me.
Which is about the same as I was stating earlier.
This isn't much removed from a normal dual fan GPU HSF. The biggest gain is flow through design, which should increase efficiency.
The questionably part, is why put the fan on the back? You could do the same thing with the fan on the other side. More of a gimmick??
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,315
To push air flow toward the fan that acts as the exhaust.
If you have a big case with a good push-pull setup on the front and the back it might not make a difference; if anything, it might worsen performance. But in a small case with restricted air flow, the fans are effectively blowing in the same direction together.
Armenius
Extremely [H]
- Joined
- Jan 28, 2014
- Messages
- 42,084
Apologies, I couldn't picture it in my head. We haven't seen the PCB design, yet, but when we do it may answer the question of why they put the fan on the back.
GoodBoy
2[H]4U
- Joined
- Nov 29, 2004
- Messages
- 2,760
So that middle portion of the heatsink has no airflow?
GoodBoy
2[H]4U
- Joined
- Nov 29, 2004
- Messages
- 2,760
The PCB seems pretty small .. where is all of the power circuitry? Starting to think this might be a transition to HBM2 memory.
Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,566
This is exactly what I am thinking as well. Then the 3090/3080ti/Titan will use GDDR6x memory.
Armenius
Extremely [H]
- Joined
- Jan 28, 2014
- Messages
- 42,084
Power circuitry is on a separate board based on what we've seen so far. Don't think it's HBM2. You can't get 11GB memory with HBM, as far as I know. 12GB would be three half stacks, and I can't see what that would look like on the interposer. And I can't see them moving to 16GB yet. There is still enough room on a half PCB for GDDR memory.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,315
From this photo I think it is a passive section that redirects air that isn't exhausted out of the case through the expansion slot vent out toward the sides where it gets pulled back in through the first fan.
Think of it like a book with a pull fan on the cover and a push fan on the back. Then folded open.
GoodBoy
2[H]4U
- Joined
- Nov 29, 2004
- Messages
- 2,760
Could be. It's a little hard to tell how far thru the center of the card the fins protrude, in the inner areas. Those on the edge in that second are full card width. The could even taper from thin on the sections' edges, to a single full width fin in the center. That would force the air out thru them, as well as separate the airflow between the front and back fans.. which would be a good idea actually.
Edit: Looking again at the picture, it appears in the center area the fins do not go all the way to the other side, and thinking about it, the other side of these fins will be the circuit board. Also, there is a fin on the GPU side of this heatsink section, that basically prevents the air to flow from the GPU fan to these fins. so I think these fins exhaust air from the rear fan, out the side of the card, and air from the GPU fan never even passes thru them, and solely exits the back of the case. Both fans are intakes. Rear fan blows thru card and out the front side of the card, GPU fan goes out the back of the card.
Last edited: