Will there be a new Mac Pro in 2015?

[flat five]

Nope I dont see it, he posted something

All I read was 2 or 3 of you jump on him because ...

scroll up.. not even too far..

• "here's the computer i'm contemplating -- thoughts?"
• five people offer thoughts.. not 2 or 3
• 2 positive / 3 negative
• blegg replies multiple times to the negs and zero times to the people trying to engage in positive conversation..


you still don't see it? it's plain

 

[ManuelGomes]

Please stop this non-sense, no more arguments.
I guess he has several opinions already, regarding what was asked. A decision should always be made according to the needs of each person.
If you feel comfortable with that decision, if you feel that rig is good for what you have to do, go for it.
Let's just give it a rest you guys...

koyoot, that could be good news, and maybe why Apple has waited all this long for the update, but I find it hard to believe that they would base the entire Dxx0 line on Fury silicon, specially since no more than 4GB is available per card. That would mean all cards are based on Fury and all have 4GB but with different levels of cut down chips, or Fury would be the lower end the higher end parts would be Grenada with 8GB, which is unlikely as well since Fury is supposed to be faster than Grenada, even the Nano I believe.
But if it has a dynamic throttling system, that's great news.
It's said that the power delivery system for Fury is well overrated, lots of headroom if needed.
For the nMP of course it needs to be constrained to the 145W or so but since the thermals seem so good it's possible that a good compromise can be achieved.

 

[lowendlinux]

I think that 4GB of HBM is going to be less of an issue than people think.

 

[ManuelGomes]

Bad news, it seems Fury lacks HDMI 2.0 after all, DP1.2a is there but HDMI is limited at 1.4a according to someone at AMD. That's a step back I'd say.
Hard to believe, all this time in the works and it comes crippled, or at least not up to the latest standards.
It was bad enough that DP was also not ready...

Yes, 4GB should be more than enough for most but it's been a clear differentiator for the various GPU levels in Apple gear.
I could see all options with 4GB, but that would mean Apple going back at least in the D700 card, from 6GB to 4GB, which I believe in unlikely.
Still, looking at the benchmarks (for those who still do and care) for resolutions up to 4K Fury with 4GB does well, but 5K and higher (8K for those who want/need that high a resolution) it comes down quickly and the cards with more mem get the cake.
For me 4K is more than enough, 5K is already something I don't really care about but Apple seems to think 5K is much better than 4K, even if it's just to make their stand over 4K. That makes me think that maybe 4GB is too limited for an Apple GPU, but maybe I'm wrong. But with this HDMI issue now, it only gets worse.

 

[koyoot]

I don't believe it would be that way. I think GPU with HBM would be an addition to the line of GDDR5 GPUs. So: D310, D510, D710 and for those who want FirePro Fury.

D310: Tonga, D510: Grenada, D710 Grenada XT. And FirePro Fury or something made of cut down Fiji.

Also, people are missing the point. Tonga is pretty good at 5K with 4 GB of Memory. And Fury X will be also good with that, especially with double pixel fillrate from Tonga technology. Tonga has 32 ROPs and Fury has 64. And Tonga with 32 ROPs has higher pixel fillrate than R9 290X with 64 ROPs. The technology is different thanks to Color Compression.
Also, Metal is designed to get rid of situations when you ran out of GPU memory.

 

[tuxon86]

I think that 4GB of HBM is going to be less of an issue than people think.

It may be a problem for those woirking in 3D modelling where your scene as to all fit in vram. This is why the 12gig cards exist. The R9 390 and 390x will come with 8gig vram. HBM2 will also come in 8gig flavor.

 

[lowendlinux]

It may be a problem for those woirking in 3D modelling where your scene as to all fit in vram. This is why the 12gig cards exist. The R9 390 and 390x will come with 8gig vram. HBM2 will also come in 8gig flavor.

I'll worry about when it indeed becomes a problem.

 

[coccosoids]

Don't know how readable this is, and sorry if it screws up the page formatting.

What do people think of this build? It's for Maya and Maxwell Render / V-Ray.

GPU is a little underpowered but hey, I can upgrade later.

Thoughts?

<edit> It's not entirely clear from the image but that is TWO processors in this build. That's not £1500 for a single 8core CPU!

I would go for higher freq Xeons. They're great for pure bucket renders though.

 

[tuxon86]

I'll worry about when it indeed becomes a problem.

For many it does... This is why they buy such card with that much vram.

 

[koyoot]

It is not that big problem, when HBM is only a High Bandwith, low latency cache for data stored in System RAM and used by Mantle/Metal APIs. HSA architecture in full chart, here. I think bigger picture can be seen, now where Apple wants to go.

And why people complain about new Mac Pro .

 

[fuchsdh]

For many it does... This is why they buy such card with that much vram.

Yeah it's funny to see how fundamentally gaming PC and workstation PC consumers totally ignore each other when it comes to video cards Ironically recently with the 4K gaming trend is the first time I've seen publications start noting how important the extra VRAM becomes.

 

[koyoot]

http://wccftech.com/amd-radeon-r9-n...ngle-8pin-connector-sff-design-faster-hawaii/

Pretty amazing is this card. I wish it could end in next Mac Pro. What worries me is the fact that its only 4 GB GPU. If we think about it, current games can be optimized to use at max 2 GB of VRAM at 4K. Critical word here is "current". For future games it will be really not enough. Maybe not for less demanding games from Blizzard, Riot, etc. The other problem is Pro segment. For professional work is a gigantic problem. You can run out of VRAM very fast. However, here comes hope with HSA, and something like Metal, as well as for games, but 4 GB of VRAM is still 4 GB of VRAM.

A problem, and I would say: "designed obsolence".

 

[lowendlinux]

Yeah it's funny to see how fundamentally gaming PC and workstation PC consumers totally ignore each other when it comes to video cards Ironically recently with the 4K gaming trend is the first time I've seen publications start noting how important the extra VRAM becomes.

No the gamers are complaining too in fact everyone is and that's exactly the reason I'm going to wait and see. The future of AMD as an independent company rest on this gen of cards and the '16 gen of processors. The mistake that both the pro camp and the gaming camp are missing is that AMD has probably thought of a way to dump a greater then 4GB frame buffer into 4GB. Folks act like this is the first time AMD has made a card, so I'll just wait and see what's actually up, look at the performance, then worry from there.

 

[deconstruct60]

Bad news, it seems Fury lacks HDMI 2.0 after all, DP1.2a is there but HDMI is limited at 1.4a according to someone at AMD. That's a step back I'd say.

How is DP1.2a a step back? Did previous AMD cards have DP1.3? No. HDMI 1.4a might be a bit of step back since some AMD cards supported HDMI 1.4b. I suspect this card does 1.4b but since "b" is just 3D 1080p @ 120Hz( which isn't "hot" anymore ) 1.4a is about as descriptive. But no AMD card ever had HDMI 2.0 so how can they be going backwards from a point they were never at?

[ from the informative tech specs aspect AMD's website blows chunks. ]


Technically with the right firmware support DP1.2a can support a HDMI 2.0 converter.
"...
Can I connect a DisplayPort 1.3 device to a new television that supports HDMI 2.0?
... Some DisplayPort 1.2a systems will most likely be upgradable through firmware to support this new feature.
..."
http://www.vesa.org/faqs/

The missing nugget is related to copy protection ( HDCP v2.2 ). There are likely OS stack hooks that are required for that which may have been foggy as Fury went to design feature freeze. Similar to why not surprising no DP 1.3 is present also.

Hard to believe, all this time in the works and it comes crippled, or at least not up to the latest standards.
It was bad enough that DP was also not ready...

Since this is hardware implementations ( hooked into the basic circuit design) you don't get standards implementation the next month after standard goes final. Sometimes get things like "pre n" or "pre ac" Wifi because the difference is in the firmware or driver and not hardcoded in the circuits.

The devices downstream have a easier time with with HDCP v2.2 because they basically just consumers. Decode to throw on the screen is probably easier than encode and transmit. The transmit part is where may loose control of the content. Similar with "pass through" components which take HDMI in and send it right to HDMI out as a switch. Not a point of origin.

I could see all options with 4GB, but that would mean Apple going back at least in the D700 card, from 6GB to 4GB, which I believe in unlikely.

Unless they finally go to support for Crossfire on PCIe. Two nanos would add up to 8GB.

The top end "Nvidia killer" card is going to be the dual model in the standard card market.

But if the HBM design got them better uniformity , better thermal control, and packed into a smaller space .... I can see Apple temporarily backsliding on top end VRAM space. The fact that the core of the GPU is just a package now means the custom part of the board that Apple has to do is smaller. Plug-in AMD's package and to a large extent just have routing to the ports, to PCIe power , and power management to do. GPU core and memory interconnects you get "for free". ( quite similar to CPU/GPU/northbridge/sothbridge fusion with CPU packages. Plug in the package and lots of stuff is just done. )


I think 8hi is possible with HBM1 but it just isn't affordable versus the mainstream 8GB DDR5 cards. For "Pro video" cards 'affordable' isn't as much of a constraint. Like the 12 core CPU option that is priced in the "if have to ask you can't afford it" range... the 8GB option could be in the same boat. Not sure what AMD is the "FirePro" space with the generation 1 HBM designs. They probably won't appear until early 2016. That is about the time the Mac Pro would see Xeon E5 1600 v4 availability as well.

HBM2 (denser memory chips ) will crack the problem but the timing for that and an updated Mac Pro are way out of sync.

 

[deconstruct60]

... The mistake that both the pro camp and the gaming camp are missing is that AMD has probably thought of a way to dump a <4GB frame buffer into 4GB.

Like Color compression?
"... The end result is that GCN 1.2 introduces a new color compression method for its ROPs, to reduce the amount of memory bandwidth required for frame buffer operations. ... "
http://www.anandtech.com/show/8460/amd-radeon-r9-285-review/3

That isn't going to turn 4GB of compressed into the equivalent of 8GB of uncompressed, but the "drop off" point is probably not the same as when pushing around more compressed data. More "horsepower" means the compression algorithms can get better ( can look harder at more data ).

The disconnect though is apps that are designed with PCIe v2 connections, no shared memory between GPU card and host (simpler, more restricted GPU memory controllers), and a statically pre cache everything possible in the VRAM mentality. They are not going to match with the more modern GPU features. There will be a limit to what AMD can do in the drivers if the overlying applications is chock full of assumptions that are a deep mismatch.

The Mac Pro with a "computation GPGPU" card is another wrinkle. Push the bulk of the math data to computation card and just pull answers to the graphics output card. May not need to stuff as much of the bulk pure math data onto the same card.

 

[tuxon86]

Unless they finally go to support for Crossfire on PCIe. Two nanos would add up to 8GB.

depends on the implementation of crossfire they use for the nano. Unless it change recently, the vram isn't combined across gpus. 2x 4gig card = 4gig vram available not 8. Supposedly Directx 12 will permit this on the windows side and maybe Metal on OSX. I haven't been following up Metal since my old 2010 i3 imac won't really benefit from it.

 

[Stacc]

I'm interested to see what the benchmarks end up being for Fiji. So far, it looks like a gaming oriented card though. Its basically 2 Tonga's stapled together with HBM, which has limited double precision compute performance. Double precision performance is what Tahiti (D500/D700) specialized in, and made them (at the time) number crunching monsters. Then again, Fiji may just have the raw horsepower that it does not matter. Its also memory limited, as everything points to the first generation of HBM being limited to 4 GB. This could be a tough sell. I see it going 1 of 3 ways. Apple could either

1. Offer alternative options for the high end. You can choose lots of memory with Hawaii/Grenada or get great performance with Fiji but with less memory.
2. Sell people on the fact that they don't need lots of GPU memory
3. Just offer Hawaii/Grenada on the high end.

Option 1 doesn't seem Apple like. They don't like confusing options, even if it makes sense from a consumer options perspective. Option 2 is possible, and if anyone can spin this its Apple. What I am not sure of is if this is possible from a professional Apps standpoint, i.e., how necessary is GPU memory is necessary or can they pull some tricks out to make it seem like a giant pool of memory. Option 3 is obviously the safe route, but offers less raw performance.

I am hoping for the mythical xMac, that slides in nicely between the Mac Mini and the Mac Pro in terms of a headless mac. Put it in a similar form factor as the Mac Pro, throw in one of those Nano Fiji's, top of the line Skylake processor in it and charge $2000-$2500 for it. Give it a thunderbolt 3 port that connects to a retina display and you are talking about an awesome machine. They could also get rid of the 4 core mac pro option and up the entry price to $3500 - $4000 and leave it as a true workhorse, professional machine. That leaves the iMac to get the broadwell/skylake processors that have a great integrated GPU and Apple doesn't have to worry about the thermal constraints of the all in one form factor.

AMD showed off its own idea of what they think is a unique form factor enabled by Fiji. It was interesting, but still looks like the crap that is pushed at PC gamers. I would love to see what Apple could come up with.

 

[koyoot]

When you have an API like Metal you completely don't need Crossfire for any purpose. And I would like people to keep that in mind.

 

[tuxon86]

I'm interested to see what the benchmarks end up being for Fiji. So far, it looks like a gaming oriented card though. Its basically 2 Tonga's stapled together with HBM,

Preliminary report seem to place it in the vicinity of the Titan X/980ti performances wise if we do believe those prelim reports... Of course the 4gig ram may be a minus for some compared to the 12gig of the Titan X or even the 6gig of the 980ti. I like the stock water cooler though. Supposedly it is rated for 500w max dissipation.

 

[tuxon86]

When you have an API like Metal you completely don't need Crossfire for any purpose. And I would like people to keep that in mind.

There not the same thing... Crossfire add a physical link between the GPU.

 

[koyoot]

I understand, but... what for? I know its a dumb question, but not only me would like to know the answer.

 

[tuxon86]

I understand, but... what for? I know its a dumb question, but not only me would like to know the answer.

Signal/clock synchronization between those GPU

 

[deconstruct60]

That would mean all cards are based on Fury and all have 4GB but with different levels of cut down chips, or Fury would be the lower end the higher end parts would be Grenada with 8GB, which is unlikely as well since Fury is supposed to be faster than Grenada, even the Nano I believe.

You are trying to rank order them in terms of a 'good , better, best' order. Could have two "top ends" where the fork is based on different criteria. If keep just 3 options then..

A plain Fury nano as a D510
A Fury as a D712 (computational speed over capacity )
A Grenada 8GB as D718 ( capacity over computational speed. )

So instead of improvement on a single dimension have a "good enough for most" and then two different better options. One toward capacity limits and the other toward computational limits.

If the fury cards make it easier to plug-in the package and get a new card then expanding to a 4 card mix wouldn't be too hard with fixed resources likely assigned to card R&D. They'd all need to merge with one thermal core interface though.

If the Mac Pro is going to move to a 2 (or more) year update cycle then skipping the new HDM wave is about on the same page as skipping TB v3 would be. It is a relatively long time if waiting until 2018 for an update.

 

[deconstruct60]

There not the same thing... Crossfire add a physical link between the GPU.

Crossfire can be run over PCIe v3. Yes they are physically connected on the same "bus", but it is not particularly added in on top of what already have done to insert both GPUs. The dual Fury board just has a PLX PCIe switch at the bottom. Have only seen some casual pictures, but I don't think there is a another whole set of traces between the packages.

 

[tuxon86]

You are trying to rank order them in terms of a 'good , better, best' order. Could have two "top ends" where the fork is based on different criteria. If keep just 3 options then..

A plain Fury nano as a D510
A Fury as a D712 (computational speed over capacity )
A Grenada 8GB as D718 ( capacity over computational speed. )

So instead of improvement on a single dimension have a "good enough for most" and then two different better options. One toward capacity limits and the other toward computational limits.

If the fury cards make it easier to plug-in the package and get a new card then expanding to a 4 card mix wouldn't be too hard with fixed resources likely assigned to card R&D. They'd all need to merge with one thermal core interface though.

If the Mac Pro is going to move to a 2 (or more) year update cycle then skipping the new HDM wave is about on the same page as skipping TB v3 would be. It is a relatively long time if waiting until 2018 for an update.

The Nano would be a good choice for the nMP since it's made to work in smaller box with a lower TDP and power usage. It is still to early to know what the performance hit the air cooled fury X will suffer compared to the water cool unit. I don't see how Apple could fit two water block inside the nMP...