MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)

[deconstruct60]

Or better to put one (or both) external storage on the IO Card? Being just an x4 device will that throttle full TB3?

@deconstruct60 do you have an opinion?

The storage isn't a problem in most casts as long as the TB controllers on the Vega II solo are feed the "proper" x4 PCI-e v3.

For example, Thunderbolt bay 6 . ... 9,200Mb/s https://eshop.macsales.com/shop/thunderbay-6/thunderbolt-3-raid

You could daisy chain two of those on the same bus and not have much of a problem. There are some NVMe SSD options though that would want to segregate.

But the 5K displays in that example are more "drama" than the storage.

 

[thisisnotmyname]

The storage isn't a problem in most casts as long as the TB controllers on the Vega II solo are feed the "proper" x4 PCI-e v3.

For example, Thunderbolt bay 6 . ... 9,200Mb/s https://eshop.macsales.com/shop/thunderbay-6/thunderbolt-3-raid

You could daisy chain two of those on the same bus and not have much of a problem. There are some NVMe SSD options though that would want to segregate.

But the 5K displays in that example are more "drama" than the storage.

I have TB3 storage that will saturate TB3 (RAID 10 NVMe arrays). Guess in a week or so I'll start plugging into different ports and see if it throttles anywhere

 

[moab1]

Same is true if ban PCI-e data consumer/producer on a port that is going to host video. More bandwidth for the video. ( For the 6K display there should be nothing else on that daisy chain. Video or PCI-e data ).

So if Apple tells you what to do with one subtype on port pairing there more to it than just that. It is also implicitly telling you about info for the other port in that pairing.

how do I configure the following on my new Mac Pro?? (Vega II Solo card)

2 x XDR Displays (OR 1 XDR & 1 5K/4K display)
1 x 4K Projector (HDMI) via AV Receiver
1 x TB3 Raid
1 x TB2 Raid (thunderbolt 2 -> 3 adaptor) with 2 other boxes daisy chained to first

I'd also like to add additional NVMe storage on PCI slots. does that make a difference in what can run? (Looking at the Micron 15.36TB SSD's (four of them at some point) via Highpoint 7120 controller)

 

[deconstruct60]

how do I configure the following on my new Mac Pro?? (Vega II Solo card)

2 x XDR Displays (OR 1 XDR & 1 5K/4K display)
1 x 4K Projector (HDMI) via AV Receiver
1 x TB3 Raid
1 x TB2 Raid (thunderbolt 2 -> 3 adaptor) with 2 other boxes daisy chained to first

You can use the HDMI connector on the Solo only if you have the 4K as the second monitor. (on TB bus 0 since it shares with the HDMI connector. ).

If you use 2 XDR or a XDR and 5K then the card is basically 'spent' in terms of edge video connection. ( both of those suck up two DisplayPort (DP) streams.). One two stream consumer has to go on TB bus 0 and the on bus 1 . Can't use adjacent ports. ).

https://support.apple.com/en-us/HT210316

The solo sends one DP to each of the standard TBv3 port pairs on the Mac Pro [ because they get only one those TBv3 ports are useless for XDR or 5k ]. You can hook a DP-to-HDMI adapter off an I/O card TB port.
The second port there will also be useless for video out at this point.


The TB3/TB2 devices can go on the Solo card on each of the bus pair sockets can't use now because of the video. They don't have to but those second bus pairs are not useless for video output so might as well stick some PCI-e data only devices on those.

I'd also like to add additional NVMe storage on PCI slots. does that make a difference in what can run? (Looking at the Micron 15.36TB SSD's (four of them at some point) via Highpoint 7120 controller)

The available PCI-e slots shouldn't impinge upon what is hanging of the Vega II solo card. The I/O may have some bandwidth sharing depending upon which PCI-e slot you put the card into.
( P.S. for the 7120 you'll need to find somewhere to put the drives since the card doesn't host the drives itself. At the moment nobody has a reasonable bracket for 4 2.5" drives. ).

 

[deconstruct60]

I have TB3 storage that will saturate TB3 (RAID 10 NVMe arrays). Guess in a week or so I'll start plugging into different ports and see if it throttles anywhere

Which is why those might be better inside. But yeah for sneaker-net moving around RAID 10 NVMe arrays TBv3 works more than reasonably well.

We'll see, but strongly suspect that the slot the full sized MPX module blocks is allocated switched to also 'feed' the TBv3 controllers on the modules blocking the slot. Otherwise it would be pretty much wasted bandwidth since physically blocked anyway.

 

[moab1]

You can hook a DP-to-HDMI adapter off an I/O card TB port.
The second port there will also be useless for video out at this point.

Thanks! So I'm clear, all three devices will run in the following set-up?

Solo card:
HDMI: rendered useless with the following Bus use
Bus 0: XDR Display 1
Bus 0: TB3 Raid
Bus 1: XDR display 2 (or 5K display)
Bus 1: TB2 Raid

4K HDMI Projector - attach via DP-to-HDMI Adaptor in the "I/O Card TB Port (currently built into PCI Slot 8?)

 

[Stephen.R]

Thanks! So I'm clear, all three devices will run in the following set-up?

Solo card:
HDMI: rendered useless with the following Bus use
Bus 0: XDR Display 1
Bus 0: TB3 Raid
Bus 1: XDR display 2 (or 5K display)
Bus 1: TB2 Raid

4K HDMI Projector - attach via DP-to-HDMI Adaptor in the "I/O Card TB Port (currently built into PCI Slot 8?)

You could just run one XDR display from the I/O card, and use the regular HDMI port on the graphics card itself.

That's mostly what the "MPX" second connector is about (besides power): routing the DP streams to the other TB3 ports in the system.

 

[deconstruct60]

You could just run one XDR display from the I/O card, and use the regular HDMI port on the graphics card itself.

Not with a Vega II solo card. A 580X or Vega II Duo? Yes. The W5700X and Solo options 'short' the default TBv3 sockets on video feeds. Only give them one instead of two. Don't have two, can't drive the XDR.
[automerge]1576217291[/automerge]

Thanks! So I'm clear, all three devices will run in the following set-up?

Solo card:
HDMI: rendered useless with the following Bus use
Bus 0: XDR Display 1
Bus 0: TB3 Raid
Bus 1: XDR display 2 (or 5K display)
Bus 1: TB2 Raid

4K HDMI Projector - attach via DP-to-HDMI Adaptor in the "I/O Card TB Port (currently built into PCI Slot 8?)

Yes, if out to spend max XDR money that is the way to go ( have to spend more to hook up the projector. )

If want to spend more money can buy a Duo and then shift one of the XDRs off the card edge and still get to use the HDMI local. But a DP-to-HDMI dongle is waaaaaaay cheaper if don't absolutely need the additional GPU horsepower.

 

[chfilm]

Errr... On the Mac Pro 2013 the HDMI connector was shared on one of the Thunderbolt Bus (TB controller). I suspect that is still true. Actually yes.

" ... If you connect a display to the HDMI port on your MPX module, the Thunderbolt 3 ports on Bus 0 can support one additional display at 4K or lower resolution. ..."

https://support.apple.com/en-us/HT210316


So presuming 6 DP streams max on Vega. In that case it would be the TB bus that is closest to the HDMI connector. [ the Radeon Pro WX9100 has a 6 min Displayport chips. It is Vega10 chip implementation but AMD didn't add anything there with Vega20 since it was mainly targeted for high compute products. I can't find evidence that Vega supports seven streams. ).

The Vega II solo has 2 DP ports going to each TB controller ( otherwise couldn't host a 6K on one of the part) and has two DP ports to the base TBv3 ports on the Mac pro ( one to each respective TB controller ... so they can't host the 5k-6k screens. ). That would leave zero leave for the HDMI port. (just like the Mac Pro 2013. ). So they 'steal' one from a TB controller for the HDMI port ( there is a DisplayPort switch upstream of the TB controller input and flip off the TB feed when the HDMI is active. ).

The current 5k monitors still soak up two DP streams.

If use the HDMI then loose one stream. A HDMI dongle of the I/O card's TB port would work. (there is a 1 stream being sent that way).



They put the same HDMI 'stealing one stream' restriction on the DUO too. I suspect just trying to maximize reuse on the edge port design work. Didn't "have to" do that with the other DP streams available. but it would mean trace rerouting on that part of the card so they probably don't want to. the "extra" two being feed into the MPX connector are on the other half of the card. That is much easier to do with little trace routing impact.

Ok... how annoying- so I have to either daisychain the HDMI via thunderbolt Adapter begins one of my raids again like I used to on the trashcan or go through one of the E/A ports.. grrrr
Thank you!

 

[deconstruct60]

Ok... how annoying- so I have to either daisychain the HDMI via thunderbolt Adapter begins one of my raids again like I used to on the trashcan or go through one of the E/A ports.. grrrr
Thank you!

if Apple had not stuck the "extra' TB controller on the top it wouldn't be a problem. Would have the same three controllers that the MP 2013 had. 6 DP streams would feed into one. (you could put one of the XDR on the I/O card.)

But the DUO and the 580X configuration opened the door for the four baseline TBv3 sockets. With one or two Duos installed it isn't an issue. So same theme with this new Mac Pro ; more money spent fewer compromises.

 

[thisisnotmyname]

if Apple had not stuck the "extra' TB controller on the top it wouldn't be a problem. Would have the same three controllers that the MP 2013 had. 6 DP streams would feed into one. (you could put one of the XDR on the I/O card.)

But the DUO and the 580X configuration opened the door for the four baseline TBv3 sockets. With one or two Duos installed it isn't an issue. So same theme with this new Mac Pro ; more money spent fewer compromises.

I wonder if they'll get smart with the dual Solos and route 2 to the IO card and 2 to the top
[automerge]1576221339[/automerge]

if Apple had not stuck the "extra' TB controller on the top it wouldn't be a problem. Would have the same three controllers that the MP 2013 had. 6 DP streams would feed into one. (you could put one of the XDR on the I/O card.)

But the DUO and the 580X configuration opened the door for the four baseline TBv3 sockets. With one or two Duos installed it isn't an issue. So same theme with this new Mac Pro ; more money spent fewer compromises.

Isn't that IO card going to be choked by being in a PCIe x4 slot? That's less bandwidth than a native thunderbolt3 controller supports.

 

[deconstruct60]

I wonder if they'll get smart with the dual Solos and route 2 to the IO card and 2 to the top

My guess is the DisplayPort lanes in the MPX Bay 2 are a 'dead end'. with some DP switching between the controllers ( i think two DP switches ) . I just don't think they added that trace complexity and addtional cost to the motherboard. I suspect that the internals TB controoler DP feeds all just simply run from the MPX Bay 1 connector. Basically cheaper and less "drama"

[automerge]1576221339[/automerge]


Isn't that IO card going to be choked by being in a PCIe x4 slot? That's less bandwidth than a native thunderbolt3 controller supports.

The TBv3 controller only takes PCI-e x4. If there were two controllers? Yes. . Is there some bleed of the USB controller and audio stream yeah. But more than likely that x4 for that slot is provisioned off of some switch that is hooked to the other slots too. so if drop something in those slots can also get some bandwidth bleed.

That is why it is probably a decent place to push "overflow" video to.

I have a suspicion that the top two ports might even more thinned out on bandwidth though if switched-share there goes highly concurrent load. ( and at least one of those x16 slots is thinned out also. Probably the second one in MPX Bay 2 )

 

[thisisnotmyname]

My guess is the DisplayPort lanes in the MPX Bay 2 are a 'dead end'. with some DP switching between the controllers ( i think two DP switches ) . I just don't think they added that trace complexity and addtional cost to the motherboard. I suspect that the internals TB controoler DP feeds all just simply run from the MPX Bay 1 connector. Basically cheaper and less "drama"







The TBv3 controller only takes PCI-e x4. If there were two controllers? Yes. . Is there some bleed of the USB controller and audio stream yeah. But more than likely that x4 for that slot is provisioned off of some switch that is hooked to the other slots too. so if drop something in those slots can also get some bandwidth bleed.

That is why it is probably a decent place to push "overflow" video to.

I have a suspicion that the top two ports might even more thinned out on bandwidth though if switched-share there goes highly concurrent load. ( and at least one of those x16 slots is thinned out also. Probably the second one in MPX Bay 2 )

I'm missing something fundamental, PCIe v3 x4 is only good for a little under 4GB/s and TB3 is good for 5GB/s (minus overhead) so how can a single TB3 controller only need x4?

 

[deconstruct60]

I'm missing something fundamental, PCIe v3 x4 is only good for a little under 4GB/s and TB3 is good for 5GB/s (minus overhead) so how can a single TB3 controller only need x4?

Because PCI-e isn't Thunderbolt . If you are encoding and mulutplexing data you need to transport at a speed higher than what you are encoding.

There is no good reason at all why they need to be exaclty the same. There are two different data streams.

Trying TB trying to gets its arms around DP v1.3/1.4 is bigger priority than trying to crank the baseline PCI-e source.

The Intel® JHL7540 Thunderbolt™ 3 Controller

datasheet: https://thunderbolttechnology.net/sites/default/files/18-241_Thunder7000Controller_Brief_FIN_HI.pdf

"... 4 lanes of PCI Express Gen 3 – Great for the fastest storage, External Graphics (eGfx), 10 Gb Ethernet, and more – Provides the Lowest latency for PCI Express audio ..."


The input source feed for PCI-e into the TB controllers has been x4 since day zero. All the way from TB v1 to v3. The controller implementation over time have evolved to pass through a higher percentage of he x4 over time. v3 jumped to passing/encoding PCI-e v3 is about the closet to 100% of all the implementations.

P.S. this is a bit older implemenation but most of the basics of what is going on is still the same.

https://www.tomshardware.com/reviews/thunderbolt-performance-z77a-gd80,3205-4.html

There is a conversion aspect to the Thunderbolt switch. Raw PCI-e data stream is just not burped onto the wires.

 

[Stephen.R]

Not with a Vega II solo card. A 580X or Vega II Duo? Yes. The W5700X and Solo options 'short' the default TBv3 sockets on video feeds. Only give them one instead of two. Don't have two, can't drive the XDR.

The Apple support doc literally says this will work (emphasis mine):

If you install a single Radeon Pro Vega II MPX Module, you can connect up to six displays using Thunderbolt 3 in these configurations:

• Two Pro Display XDRs with resolutions of 6016 x 3384 at 60Hz connected to any two of the following locations: Bus 0, Bus 1, or the Thunderbolt 3 ports on the rear of your Mac Pro. Connect one display for each location.

so I have to either daisychain the HDMI via thunderbolt Adapter

No. Read the document, specifically the part under "Connect displays to a single Radeon Pro Vega II MPX Module". It literally says you can use any two of the TB3 ports on the back of the Mac.

 

[vailr]

Possible new "Mac Pro Junior" hardware, disguised as "XBOX Series X"?

Microsoft says that the Series X will target 4K/60fps performance with Zen 2 and RDNA architecture from AMD, leveraging hardware-accelerated ray tracing, GDDR6 memory, and NVMe solid-state storage.
https://www.theverge.com/circuitbreaker/2019/12/13/21020149/xbox-series-x-pc-specs-analysis

 

[fuchsdh]

Possible new "Mac Pro Junior" hardware, disguised as "XBOX Series X"?
View attachment 882749
Microsoft says that the Series X will target 4K/60fps performance with Zen 2 and RDNA architecture from AMD, leveraging hardware-accelerated ray tracing, GDDR6 memory, and NVMe solid-state storage.
https://www.theverge.com/circuitbreaker/2019/12/13/21020149/xbox-series-x-pc-specs-analysis

It does feel a little like they were inspired by the tube Mac Pro. Will be interested in the teardown to see if they drew on the thermal core idea at all.

 

[ekwipt]

Possible new "Mac Pro Junior" hardware, disguised as "XBOX Series X"?
View attachment 882749
Microsoft says that the Series X will target 4K/60fps performance with Zen 2 and RDNA architecture from AMD, leveraging hardware-accelerated ray tracing, GDDR6 memory, and NVMe solid-state storage.
https://www.theverge.com/circuitbreaker/2019/12/13/21020149/xbox-series-x-pc-specs-analysis

Could be a distinct possibility we will see a revamped Apple TV Gaming next year to compete with the Xbox and PS5, Apple wants to be in the centre of every loungroom, Apple TV was a great start, but if you want to use VR and more complex games, they'll have to use AMD APU systems

 

[OkiRun]

It does feel a little like they were inspired by the tube Mac Pro. Will be interested in the teardown to see if they drew on the thermal core idea at all.

Competition is great for all of us. It drives innovation which = more and (better) choices.

 

[AidenShaw]

Will be interested in the teardown to see if they drew on the thermal core idea at all.

Or to see if they avoided repeating that mistake.

 

[fuchsdh]

Or to see if they avoided repeating that mistake.

I don't see where the mistake to repeat here would be—the Xbox is going to be built for one spec that will at worst never produce more heat or draw more power, and more likely will become more power-efficient with die shrinks if they produce newer models. There's no unexpected thermal corner to be backed into—the benefits of consoles versus PCs.

 

[AidenShaw]

the Xbox is going to be built for one spec that will at worst never produce more heat or draw more power

Isn't this Apple's fundamental mistake with the thermal core - expecting that TDPs will only go down?

 

[mattspace]

Could be a distinct possibility we will see a revamped Apple TV Gaming next year to compete with the Xbox and PS5, Apple wants to be in the centre of every loungroom, Apple TV was a great start, but if you want to use VR and more complex games, they'll have to use AMD APU systems

VR is still-born on Apple platforms.

1. AMD can't produce realtime 3d muscle that's competitive with Nvidia - look at Macsurfer, they link to articles fawning over the gaming (VR equivalent) performance of the 5700 as being "Nvidia beating", except it's "beating" a 1660 - it's laughable.
2. There's no Mac culture of GPU upgrading and swapping, and VR is still in the "eat all the GPU possible" stage.
3. Thunderbolt is an expensive, and low performance option for hosting GPUs, as compared to putting them in a computer.
4. macOS has nothing to bring to the table, because you don't interact with the host operating system on a VR station.

All of the above means that VR developers aren't bothering to target macOS as a deployment platform. That isn't going to change until Apple delivers them customers, and they aren't going to deliver them customers until they have an xMac-class machine, and get Nvidia back in the picture.

 

[thisisnotmyname]

Because PCI-e isn't Thunderbolt . If you are encoding and mulutplexing data you need to transport at a speed higher than what you are encoding.

There is no good reason at all why they need to be exaclty the same. There are two different data streams.

Trying TB trying to gets its arms around DP v1.3/1.4 is bigger priority than trying to crank the baseline PCI-e source.

The Intel® JHL7540 Thunderbolt™ 3 Controller

datasheet: https://thunderbolttechnology.net/sites/default/files/18-241_Thunder7000Controller_Brief_FIN_HI.pdf

"... 4 lanes of PCI Express Gen 3 – Great for the fastest storage, External Graphics (eGfx), 10 Gb Ethernet, and more – Provides the Lowest latency for PCI Express audio ..."


The input source feed for PCI-e into the TB controllers has been x4 since day zero. All the way from TB v1 to v3. The controller implementation over time have evolved to pass through a higher percentage of he x4 over time. v3 jumped to passing/encoding PCI-e v3 is about the closet to 100% of all the implementations.

P.S. this is a bit older implemenation but most of the basics of what is going on is still the same.

https://www.tomshardware.com/reviews/thunderbolt-performance-z77a-gd80,3205-4.html

There is a conversion aspect to the Thunderbolt switch. Raw PCI-e data stream is just not burped onto the wires.

Where I'm getting lost is that we apparently have more endpoint bandwidth than backhaul. I'm sure I'm being slow on the uptake and missing something very basic here but if we can support data at 40 Gb/s (5 GB/s minus overhead) across the TB3 endpoint but the controller is bused into the rest of the computer on PCIe v3 x4 which only supports ~3.9 GB/s we have a delta of ~1.1 GB/s. I'm missing how we can support that 40 Gb/s speed when the bus the controller is on doesn't. Sorry to make you explain this like you're talking to a five year old, what am I missing here?

 

[fuchsdh]

Isn't this Apple's fundamental mistake with the thermal core - expecting that TDPs will only go down?

But it's a console, so that's an entirely reasonable assumption to make. Either the thermals remain exactly where they are when you produced it, or (like previous versions of the Xbox 360 and Xbox One) they release later refinements on the same specs, creating smaller, less power-hungry machines.

Apple's failure was not building enough wiggle room so they could fit hotter GPUs if necessary, or enough headroom for the entire thing to comfortably run maxing all side of the core. That shouldn't be an issue for Microsoft even if they adapted the same idea (especially since you're running games, so everyone knows that you're pegging the GPU and CPU constantly for hours on end.)