MP 7,1 Radeon VII in the 7,1 Mac Pro

[PowerMike G5]

I currently have a Radeon VII in my cMP that I am thinking of porting over to the 7,1. Although I was planning on the Radeon Pro Vega II option, I'm starting to wonder if it makes more sense to port the Radeon VII and even just install a 2nd one.

From my understanding, it seems that the Radeon VII is very close in spec to the Pro Vega II. Basically the same GPU with half the VRAM and 4 less CU, but otherwise the same hardware.

I could get another one for like $500 and have a dual setup that will be more overall power than one of the Pro Vega II options, and also save roughly $2000 in the process.

The money isn't the issue in this case, but already having one Radeon VII makes this an option for me now. What does the collective Mac Pro hive mind think about this here? Primary use will be for Adobe Premiere Pro video editing and color grading.

 

[ctishue]

I've been considering the same thing since it is basically just a bit over 1/5th the cost and would allow me to put more money into SSD and CPU out of the gate. Can always choose to sell it and upgrade to one of the MPX modules Apple offers later I guess.

 

[Pressure]

Remember that the clock speed is different, so they perform nearly equal despite the missing CU count (14.1 vs 13.8 single precision).

It's only if you benefit from the extra 16GB of HBM2 things will shift in favour of the Pro Vega II. Also the noise level will be different as the Apple part is completely passive cooled.

 

[Briax]

Will the Radeon VII be able to power a Pro Display XDR?

 

[PowerMike G5]

Yes some good points. I have noticed that I max out the 16GB VRAM usage in iStat when I'm doing certain video edits, but it hasn't really been an issue. I think the idea of adding a 2nd Radeon VII and saving roughly $2000 is starting to be quite tempting. Especially with some other EOY purchases I need to make that cost more than this Mac Pro already

 

[Pro7913]

Radeon Pro Vega 2 is a workstation version with more HBM2 memory. AMD usually charge 2~3 times more than the original version for their workstation version.

 

[PowerMike G5]

For those that know more about the drivers in MacOS, is it safe to assume that the Radeon VII and the Radeon Pro Vega II will use the same GPU driver? If so, then there really shouldn't be any difference outside of what the hardware specs are, yes?

 

[chfilm]

And what about the lack of infinity fabric when you use two of these as opposed to two Vega IIs?
And does the Radeon 7 have any TB ports?

 

[goMac]

Will the Radeon VII be able to power a Pro Display XDR?

No.

The XDR requires Thunderbolt. Thunderbolt displays can only be driven from an MPX module.

 

[PowerMike G5]

There are no thunderbolt ports on the Radeon VII. That's one of the advantages of the MPX module with the Pro Vega II.

The lack of Infinity Fabric is something to think about, but its highly app dependent. For Adobe Premiere, it uses a single GPU for acceleration and playback and dual GPUs for acceleration and exports. I think for my needs at the moment, Infinity doesn't play a big role.

The interesting thing about this idea too is that I would have the 580X by default with the purchase, so I'd be able to put all 3 GPUs in here (2 Radeon VIIs + 580X) and still save about $2K.
[automerge]1576342300[/automerge]

No.

The XDR requires Thunderbolt. Thunderbolt displays can only be driven from an MPX module.

Yes, so this requires the Radeon Pro Vega II at minimum to use the XDR.

I'm very happy with my LG 4k DCI monitor, so plan to continue using that in addition to my Flanders Scientific monitor connected to my AJA Kona. I have the color accuracy monitor parts covered

 

[Z28McCrory]

I’m very curious about this too.

I don’t currently own a VII, but had considered ordering my 7,1 with the base GPU and buying a pair of VII’s for a lot less $$$$ than the Vega II upgrade.

Surely a pair of VIIs would equal (or considerably outperform) the single Vega II?

My use case is primarily photo and video. The video is “Only” 4K, but with a lot of grading and effects, so I think the 32GB on a single Vega II would be wasted on me, but I could benefit a lot more from the processing power of a pair of VII’s.

 

[PowerMike G5]

It is app dependent, but my assumption is a pair of Radeon VIIs will considerably outperform a single Pro Vega II. From what I gather, it seems that a single Radeon VII will pretty much equal a single Pro Vega II already, outside of the extra VRAM, given they are both closely the same hardware.

In a use case with Adobe, it looks like the dual Radeon VII's will increase performance, given it will use both GPUs fully for things like accelerated exporting, etc. I believe timeline playback will only use one GPU, but it should use both for accelerated effects. And in this particular instance, since we'd have to get the base GPU, you could have 3 GPUs and spend about $2000 less. It's a tempting proposition.

I guess the final thing is if they both use the same driver from Apple, then it does make more sense to go with 2 Radeon VIIs.

 

[_Kiki_]

Are you sure if it will work ? Isn't 3rd party GPUs blocked by EFI ?

 

[PowerMike G5]

Are you sure if it will work ? Isn't 3rd party GPUs blocked by EFI ?

I don't see why not. I'm already using the GPU internally in the 5,1 cMP with no issues. Apple's FAQ on the 7,1 also states that the GPUs that work as eGPUs will work internally in the new Mac Pro as well.

In fact, they even show doing this in their official marketing pictures of the Mac Pro.

 

[defjam]

I've been considering the same thing since it is basically just a bit over 1/5th the cost and would allow me to put more money into SSD and CPU out of the gate. Can always choose to sell it and upgrade to one of the MPX modules Apple offers later I guess.

Isn't it refreshing to be able to do this once again? Thank you Apple for realizing your mistake with the 2013 Mac Pro and correcting it!

 

[deconstruct60]

....

The interesting thing about this idea too is that I would have the 580X by default with the purchase, so I'd be able to put all 3 GPUs in here (2 Radeon VIIs + 580X) and still save about $2K.

Probably not. You are probably not boing be able to pull power from the MPX bay 1 and put it into a card in MPX bay 2. Even if can also probably not going to get usage of the MPX bay 2 8-pin power feed not to conflict with the second PCI-e slot in Bay 2.

Illustrative is right on Belkin power cable page.

https://www.apple.com/shop/product/HMUJ2ZM/A/belkin-aux-power-cable-kit-for-mac-pro

See how the cables come out and around to the card above. That soaks up space. I know folks have been hand waving about how you'll be able to snake cords thru multiple 90 changes of directions with zero cable bend, but that is probably not true.

That second x16 slot in BAY is good for perhaps a PCI-e SSD card, but a full length , max double wide, relatively high card is probably going to have major problems finding a source to plug into.


Two Radeon VIIs and 'eject' the 580X ? yes that will work. ( but back to not driving the XDR or thunderbolt ..... without some giant, Rube Goldberg kludges ).

Yes, so this requires the Radeon Pro Vega II at minimum to use the XDR.

No. The 580X can run the XDR just fine off of the default system TBv3 ports. MBP 16" can run XDR. About everything released in 2019 can run the XDR. (the iMac Pro can run it in 5K mode. Not particularly cost effective but illustrative of extent of "minimal" requirements here. )


It will run nicer with an Vega II, but it isn't a minimal requirement.
[automerge]1576356604[/automerge]

Yes some good points. I have noticed that I max out the 16GB VRAM usage in iStat when I'm doing certain video edits, but it hasn't really been an issue. ...

Do those project correlate with certain attributes of the projects. For example 8K video transcoding spikes higher usage. If that is a long term growth trend for you workload demographics then will be a bigger deal over time.

Similar issue with size of resolution screen dealing with. ( size and scope/complexity of what is in active view).

There are some apps though than can barely make use of a single GPU (or its VRAM). If that's the primary tool then yeah there isn't much "bang for the buck" in Vega II. But the apps that following closely on the leading edge of Metal there is a substantive difference. Even more so if can shift data over to the VRAM size. ( two Vega II coupled together is 64GB of space. ]

 

[PowerMike G5]

Probably not. You are probably not boing be able to pull power from the MPX bay 1 and put it into a card in MPX bay 2. Even if can also probably not going to get usage of the MPX bay 2 8-pin power feed not to conflict with the second PCI-e slot in Bay 2.

Illustrative is right on Belkin power cable page.

https://www.apple.com/shop/product/HMUJ2ZM/A/belkin-aux-power-cable-kit-for-mac-pro

See how the cables come out and around to the card above. That soaks up space. I know folks have been hand waving about how you'll be able to snake cords thru multiple 90 changes of directions with zero cable bend, but that is probably not true.

That second x16 slot in BAY is good for perhaps a PCI-e SSD card, but a full length , max double wide, relatively high card is probably going to have major problems finding a source to plug into.


Two Radeon VIIs and 'eject' the 580X ? yes that will work. ( but back to not driving the XDR or thunderbolt ..... without some giant, Rube Goldberg kludges ).




No. The 580X can run the XDR just fine off of the default system TBv3 ports. MBP 16" can run XDR. About everything released in 2019 can run the XDR. (the iMac Pro can run it in 5K mode. Not particularly cost effective but illustrative of extent of "minimal" requirements here. )


It will run nicer with an Vega II, but it isn't a minimal requirement.
[automerge]1576356604[/automerge]


Do those project correlate with certain attributes of the projects. For example 8K video transcoding spikes higher usage. If that is a long term growth trend for you workload demographics then will be a bigger deal over time.

Similar issue with size of resolution screen dealing with. ( size and scope/complexity of what is in active view).

There are some apps though than can barely make use of a single GPU (or its VRAM). If that's the primary tool then yeah there isn't much "bang for the buck" in Vega II. But the apps that following closely on the leading edge of Metal there is a substantive difference. Even more so if can shift data over to the VRAM size. ( two Vega II coupled together is 64GB of space. ]

Good point. But the 580x is listed as a half-height MPX module that enables the use of the other slot that would be covered by the regular MPX module. Without seeing it though, I can't tell whether the 580x is setup in a way where this means there's just space within the MPX module open for the other slot, or if it really is just a half height module. Have you found any info about that?

Also with the PCIe Utility app back for this 7,1 MP, I wonder if we can dynamically assign the lanes again. Which means we'll have flexibility to route the cards as needed, given the space.

 

[goMac]

Are you sure if it will work ? Isn't 3rd party GPUs blocked by EFI ?

To be clear, third party GPUs were never blocked by EFI. Apple just used an older version of EFI that couldn't do boot screens with standard PC cards.

The 2019 Mac Pro uses the current UEFI standard, so it's fully compatible with PC cards.

It's just Apple moving to the current standard in the 9 years since the 2010. Not them removing a block.

 

[aaronhead14]

I’ve actually been thinking the same thing; however, not having those additional Thunderbolt 3 ports might be the dealbreaker for me.

 

[deconstruct60]

Good point. But the 580x is listed as a half-height MPX module that enables the use of the other slot that would be covered by the regular MPX module. Without seeing it though, I can't tell whether the 580x is setup in a way where this means there's just space within the MPX module open for the other slot, or if it really is just a half height module. Have you found any info about that?

You can use the other double wide slot in Bay 1 ... your just not getting high power. So if it is a bus powered ( 75W) card it would work fine. The problem is that the 580X is using at least a power of that aux 8 pin power of that bay. ( the 580X is not a 75W card. Far from it. ). So "all" of the power isn't available on MPX Bay 1 aux power. I strongly suspect that Apple just flips that whole aux power off once something is slotted and drawing power from the MPX connector in that bay. ( Folks keep proposing that there is some complicated circuit distribution tap dance that Apple has implemented. My bet is that it is a lot simpler than that either all option A or all option B).

That MPX Bay 2 x16 slot might be able to get away with an longer cable and take the 6 pin from slot 5 and run it down sure. But you are not going to feed a 250+ W kind of card like that. Maybe some overly heavyweight I/O fabric card. Or maybe a entry level video card. Or an HDX audio ( there is more to life than simply just high power draw video cards. )

This document though says 6-pin is only suppose to be for slots 5-8. And each MPX Bay 8 pin subsystem is only for that respective MPX bay. I'm sure folks are going to ignore the docs and attempt to hack away. but for what it was designed to do ... that is documented.

The Mac Pro is basically optimized in the 3-4 GPU space toward the MPX. There may be some two mainstream cards that will outpace a single Vega Solo or single Duo. But 3-4 GPU space Apple has pretty much placed themselves in the more pragmatically advanced position. ( hence scoop up more of the folks with the bigger budgets. )

Also with the PCIe Utility app back for this 7,1 MP, I wonder if we can dynamically assign the lanes again. Which means we'll have flexibility to route the cards as needed, given the space.

Apple talks about some flexibility but it is never going to be send everything to anywhere. I think it is mainly going to be mostly about effectively turning off slots 6 and 7 because not really using them. Or if are using 6 and 7 how to dribble out the substantially heavy over subscription you have in a 'fair' way. The relegation from the secondary slots in the MPX Bays ( 2 and 4 ) probably will simply just happen automatically when they are covered up and that diversion is only to the MPX connector.


P.S. the PCI-e slot document also labels Slots 1 and 3 as "Use First" which is probably indicative that those do not have PCI-e switches and are hard wired to the CPU. Those are exactly what the MPX modules will only plug into.

 

[deconstruct60]

To be clear, third party GPUs were never blocked by EFI. Apple just used an older version of EFI that couldn't do boot screens with standard PC cards.

The 2019 Mac Pro uses the current UEFI standard, so it's fully compatible with PC cards.

UEFI standard includes secure which some cards don't really follow. the notion that anything for from budget bin is going to work across all boot options is going to stick. ( the boot camp into Windows ... if that was all super , lowest common denominator standard stuff would just work.... and it doesn't. )

There is more coverage now, but dropping down the lowest of low Windows tolerated kludges (and/or sneaky fall backs to BIOS) ... I wouldn't hold my breath on that.

 

[goMac]

UEFI standard includes secure which some cards don't really follow. the notion that anything for from budget bin is going to work across all boot options is going to stick. ( the boot camp into Windows ... if that was all super , lowest common denominator standard stuff would just work.... and it doesn't. )

There is more coverage now, but dropping down the lowest of low Windows tolerated kludges (and/or sneaky fall backs to BIOS) ... I wouldn't hold my breath on that.

Apple notes that 32 bit EFI cards won't work. And you're right, cards that don't have proper UEFI implementations probably won't work.

Fortunately most people here are concerned with newer cards that are based on reference designs with proper UEFI implementations. But yeah. I wouldn't assume that if you grab a real budget card or some random card from 5 years ago, it's gonna work.

Very unlikely that Apple included any UGA or traditional BIOS ROM support. I could _maybe_ see UGA if Apple wanted to make sure older Mac cards worked.

 

[Z28McCrory]

On another thread someone posted this screenshot of some of the "bugs" with using a VII as an eGPU. Do you think there is potential for the same issues if a VII is used internally in a Mac Pro?

 

[PowerMike G5]

On another thread someone posted this screenshot of some of the "bugs" with using a VII as an eGPU. Do you think there is potential for the same issues if a VII is used internally in a Mac Pro?

View attachment 882850

Possibly. But I have been using it in the 5,1 with no issues whatsoever. Sleep and fans work as expected. Maybe that's more indicative, given its running on an internal PCIe slot in this instance.

 

[PowerMike G5]

Starting to find some pics of the 580X MPX module. It looks like it doesn't block the power connection ports for either MPX bay and powers itself fully from PCIE slots 1 & 3.

So in regards to running 2 Radeon VIIs, it seems physically possible. But as deconstruct mentions, the question remains whether pulling 2 8-pin power from the bay also using the Radeon MPX module is within limit?