AMD Radeon "Big Navi" PCB Allegedly Pictured

[btarunr]

Here's what is possibly the first picture of an AMD Radeon RX 6000 "Navi 21" reference PCB, which the company could use across its RX 6900 XT and RX 6800 series graphics cards. The partially built PCB has a possibly large ASIC at the center (masked out), surrounded by eight GDDR6 memory pads, confirming a 256-bit wide memory interface. There are two 8-pin PCIe power connectors, and a 16-phase VRM solution powering the card's 320 W typical board power. Display connectors on the board appear to include two DisplayPorts, one HDMI, and a USB type-C. There are also a handful headers handling the fans and possible ARGB LED illumination of the card. We also spy a 2-way switch, which could indicate dual-BIOS.



View at TechPowerUp Main Site

 

[silentbogo]

Looks legit, and it seems to be the same layout as previous leak (and matches AMD renders of the card from recent presentations).

 

[Valantar]

I really hope retail boards aren't as unnecessarily tall as that ... I get that they need board space for power planes to deliver the kind of current needed by a GPU like this, but that still looks excessive.

 

[FinneousPJ]

A 6800 non XT(X) with dual BIOS and triple power connectors would be my dream board. I imagine you could grab a Russian bootleg bios and turn it into a monster

 

[ratirt]

At the Guru3D, the op mentioned that this board is almost identical with the Sapphire Nitro+ 5700XT.

 

[Chrispy_]

Damn; I was hoping the reference PCB was general-purpose enough to allow for blower style coolers, but there are too many components where the radial fan would need to be for that to happen.

Before the flame war over blower coolers kicks off, I *know* they're worse than open designs with multiple fans for cooling the GPU but it's sad that the Big Navi won't have a blower option because sometimes getting 200W+ of heat directly out of the back of the PC is really heckin' useful.

For very quiet/silent builds, or for SFF builds where space is at a premium, the blower's single fan can cope well, depending on where you set the power limits, and you don't have to worry about adequate space, additional intake/exhaust clearance, and additional case cooling requirements for the motherboard, RAM, CPU caused by having an open GPU cooler.

 

[Valantar]

Damn; I was hoping the reference PCB was general-purpose enough to allow for blower style coolers, but there are too many components where the radial fan would need to be for that to happen.

Before the flame war over blower coolers kicks off, I *know* they're worse than open designs with multiple fans for cooling the GPU but it's sad that the Big Navi won't have a blower option because sometimes getting 200W+ of heat directly out of the back of the PC is really heckin' useful.

For very quiet/silent builds, or for SFF builds where space is at a premium, the blower's single fan can cope well, depending on where you set the power limits, and you don't have to worry about adequate space, additional intake/exhaust clearance, and additional case cooling requirements for the motherboard, RAM, CPU caused by having an open GPU cooler.

There have been blower versions of the past two generations of RTX cards from at least Gigabyte and Asus, both of which also make AMD GPUs. GB even has a blower RTX 3090. If the demand is there, they'll make one. There are good arguments for not designing a reference card for a blower though, as that design forces a ton of board space to be left essentially unused to make room for the fan, which makes the design very inefficient for the majority (non-blower) cards it will actually be used in.

(As for the "blower is better for SFF" argument, that is extremely dependent on your case, and not typically true with current SFF cases (Ncase, Ghost S1, Formd T1, etc.) provided you set them up correctly with the required fans. But there are of course cases where a blower is better.)

 

[_JP_]

At the Guru3D, the op mentioned that this board is almost identical with the Sapphire Nitro+ 5700XT.

I'm not seeing how "almost identical" fits in the complete different board component layout, besides the "traditional placement" for VRMs next to the chips they feed, which applies to every recent card made in the last ~10 years. :\

 

[Chrispy_]

There are good arguments for not designing a reference card for a blower though, as that design forces a ton of board space to be left essentially unused

A good point well made.

(As for the "blower is better for SFF" argument, that is extremely dependent on your case, and not typically true with current SFF cases (Ncase, Ghost S1, Formd T1, etc.) provided you set them up correctly with the required fans. But there are of course cases where a blower is better.)

Well yes. I wasn't referring to all SFF cases, just those where the CPU and GPU occupy the same general volume of the case. A lot of newer SFF cases move the GPU to a completely seperate airflow zone via a PCIe ribbon cable, and are specifically designed that way so that you can use a GPU with an open-air cooler.

 

[ratirt]

View attachment 173029
View attachment 173028

I'm not seeing how "almost identical" fits in the complete different board component layout, besides the "traditional placement" for VRMs next to the chips they feed, which applies to every recent card made in the last ~10 years. :\

I've misspoken. Similarities not almost identical. It is a different chip and so it has to be slightly different.

 

[Turmania]

When will they send the review samples to tech sites?

 

[Chrispy_]

When will they send the review samples to tech sites?

Probably a week before launch or so, which isn't the 28th, it's an undisclosed date in November

Next wednesday is just the official unveil and at this point quite a lot of what's going to be unveiled has already leaked.

 

[Valantar]

View attachment 173029
View attachment 173028

I'm not seeing how "almost identical" fits in the complete different board component layout, besides the "traditional placement" for VRMs next to the chips they feed, which applies to every recent card made in the last ~10 years. :\

I've misspoken. Similarities not almost identical. It is a different chip and so it has to be slightly different.

The only similarities I see are the general outline of the PCB including the positioning of I/O and power connectors, as well as the rough position of the GPU. RAM is entirely different, VRMs are entirely different ... This is not "slightly different". The only thing even warranting the comparison IMO is the tall PCB with the notch for the power connectors.

 

[jesdals]

The only similarities I see are the general outline of the PCB including the positioning of I/O and power connectors, as well as the rough position of the GPU. RAM is entirely different, VRMs are entirely different ... This is not "slightly different". The only thing even warranting the comparison IMO is the tall PCB with the notch for the power connectors.

considering the mounting holes for the cooler, it looks like a new design?

 

[Jism]

I really hope retail boards aren't as unnecessarily tall as that ... I get that they need board space for power planes to deliver the kind of current needed by a GPU like this, but that still looks excessive.

It all makes sense. You distribute the power among the chip, not one way in. If you do it one way in it requires alot of current to pass through a few layers of PCB ... Not really what you want.

300W GPU is pretty much 240A at 1V ~ 1.2V.

 

[Valantar]

It all makes sense. You distribute the power among the chip, not one way in. If you do it one way in it requires alot of current to pass through a few layers of PCB ... Not really what you want.

300W GPU is pretty much 240A at 1V ~ 1.2V.

Yes, but that's only between the die and VRMs. There's only 12V outside of that, and the VCore voltage planes do 100% guaranteed not extend to the top of the PCB - there is no reason for them to do so. So then you have 25A instead.

considering the mounting holes for the cooler, it looks like a new design?

Obviously, yes. If it's a 6000-series board, it would be downright shocking if they could re-use an older design.

 

[jesdals]

Yes, but that's only between the die and VRMs. There's only 12V outside of that, and the VCore voltage planes do 100% guaranteed not extend to the top of the PCB - there is no reason for them to do so. So then you have 25A instead.


Obviously, yes. If it's a 6000-series board, it would be downright shocking if they could re-use an older design.

At guru3d they mention that it could be a drop in gpu board

 

[Valantar]

At guru3d they mention that it could be a drop in gpu board

You mean one of those socketed ones they use for testing and binning? I don't see any evidence of a socket (those sockets are huge), and besides, those still need to be designed around at least a series of ASICs. It would be very surprising if a board made for Navi 1X fit Navi 2X chips.

 

[Caring1]

I thought USB-C was being dropped from GPUs for whatever reason.

 

[Totally]

At the Guru3D, the op mentioned that this board is almost identical with the Sapphire Nitro+ 5700XT.

I'm lookina at it in my case. They're not even close.


edit:

I've misspoken. Similarities not almost identical. It is a different chip and so it has to be slightly different.

Similar doesn't even cut it close either the vrm layout is completely different and there has been some power delivery added between the gpu die and i/o, ram arrangement is different, and the pcb is very sparsely populated also. There are more differences than similarities and the differences are obvious.

 

[ratirt]

I'm lookina at it in my case. They're not even close.


edit:


Similar doesn't even cut it close either the vrm layout is completely different and there has been some power delivery added between the gpu die and i/o, ram arrangement is different, and the pcb is very sparsely populated also. There are more differences than similarities and the differences are obvious.

I'm not saying it is similar or identincal but at Guru3d they have described the board that way. It has ti be different slightly it is a new chip. Maybe they were talking about the layout or that this board will be nitro+.

 

[Valantar]

I'm not saying it is similar or identincal but at Guru3d they have described the board that way. It has ti be different slightly it is a new chip. Maybe they were talking about the layout or that this board will be nitro+.

Or maybe they just took a quick look at it, saw a similar-looking outline (and ignored or didn't look closely enough at everything else), and made a poorly thought-through point in an argument that was pushed out early to gain clicks? You seem to be arguing that "they said it, so it must be true in some regard." In this case it's quite obvious that it isn't.

 

[ratirt]

Or maybe they just took a quick look at it, saw a similar-looking outline (and ignored or didn't look closely enough at everything else), and made a poorly thought-through point in an argument that was pushed out early to gain clicks? You seem to be arguing that "they said it, so it must be true in some regard." In this case it's quite obvious that it isn't.

I don't argue nor I say it is true. I said they've said that and I've posted it here for others to review. What you do with it it's on you. Is it similar or not? I couldn't care less.

 

[Valantar]

I don't argue nor I say it is true. I said they've said that and I've posted it here for others to review. What you do with it it's on you. Is it similar or not? I couldn't care less.

But people are saying that it clearly isn't very similar, yet you keep contradicting them, and trying to explain what Guru3D might have meant. That is arguing. If you don't care, it's rather odd that you keep retorting to the people saying that their statement is wrong. But never mind that, it's plenty clear at this point that Guru3D made an off-the-cuff statement based on a very high-level visual similarity (essentially nothing more than the PCB outline, power connectors and I/O placement) and that even a little scrutiny has shown it to be untrue. Let's move on, yeah?

 

[ratirt]

But people are saying that it clearly isn't very similar, yet you keep contradicting them, and trying to explain what Guru3D might have meant. That is arguing. If you don't care, it's rather odd that you keep retorting to the people saying that their statement is wrong. But never mind that, it's plenty clear at this point that Guru3D made an off-the-cuff statement based on a very high-level visual similarity (essentially nothing more than the PCB outline, power connectors and I/O placement) and that even a little scrutiny has shown it to be untrue. Let's move on, yeah?

I don't care. I just posted it cause I found it. Yeah let's move on.