Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

New MacPro predictions? Hopes? Fears?

Jeff Kilgroe said:
(snip) And while the Mac Pro trash can doesn't get a lot of love around here, it's been the best selling Mac Pro. Ever. It is very highly regarded in many pro environments. (snip)
Click to expand...

Thanx to Jeff for noting this, it's easy to get tunnel vision and forget just how demanding our particular use case is. That said, I dearly hope that a 7,1 MacPro will be released in the next year that leverages multiple newer technologies to make it powerful enough for 8K R3D workflows. The components required are either already on the market or dropping soon. Whether it's worth Apple's time to make a high spec iteration of the MacPro for what we do is another question altogether. - especially if they continue to eschew nVidia/CUDA.

Cheers - #19
 
After speaking with someone I respect at one of the SW manufacturers - he told me:

"For playback performance, we still need to do tests but its looking like you will need a Linux dual 20 core CPU and four or more Titan X Pascals for reasonable 4K performance of the 8K source clip."

We need to build a system that can handle the actual 8K, not a 4K "proxies" of the 8K clips. End of this month can not come soon enough. Bring it on Apple...
:reddevil:
 
Jeff Kilgroe said:
...Not so sure that Samsung thing is legit. OTOH, it may be just another attempt at copying Apple's successes. And while the Mac Pro trash can doesn't get a lot of love around here, it's been the best selling Mac Pro. Ever.
Click to expand...

Curious where you read this? Apple has never released sales data to that level of detail to my knowledge.
 
8k workflow ability is years ago (think, how many pixels per second is 6k hispeed compared to 8kp24). Next year or two we might have 10kp50 going on Red's oversampling, and you should expect upto 16k holographic from elsewhere in the lefe time of the unut). We are way, way, way behind.
 
Peter Majtan said:
After speaking with someone I respect at one of the SW manufacturers - he told me:

"For playback performance, we still need to do tests but its looking like you will need a Linux dual 20 core CPU and four or more Titan X Pascals for reasonable 4K performance of the 8K source clip."

We need to build a system that can handle the actual 8K, not a 4K "proxies" of the 8K clips. End of this month can not come soon enough. Bring it on Apple...
:reddevil:
Click to expand...

Then, can I suggest again: What about a thread on building an mini blade system that is expandable for whatever comes along also useful for 3D (extra build volume, even on kickstarter). It is probably cheaper to make a render farm of cheaper parts, with Red contributing custom mass distributed pattallel drivers/plugins (that they can then on license as general versions for non red systems. Thus covering their costs and making some profit) (note free version handles only Red footage, everybody else pays from $100-$1000). I'm talking about keeping it in redcode as long as possible.

Does that seem like a plan. Basically I'm also expecting the ability to print my more advanced designs at an atomic level one day, so I have an need again for mo power, and to run my crude simulations at the atomic level (crude being using the least processing as possible to get reliable mechanism modelling). Suddenly, we are looking at a future with a big market for very high end use, and a host of new professionals.

I'll tell you what. I have been sick a lot, but was planning to contact Nvidia about making an Arm+cuda based parallel processing machine that was 1kw of power consumption (and bigger versions for research). Even 200w is pretty hefty processing power. 7nm limit is fast approaching, so such a machine could be built before it gets here. With riscV I would upgrade that to riscv for even higher performance. Before anybody says this or the other about the ti be replaced TX1, compare it on a performance per watt basis to intel (early testing put it up against Intel i3 in early Linux (likely unoptimized, and without the needed custom drivers) 10w with unused cuda. Using cuda coprocessing how many i7's is that likely to blow away at 10w, let alone with optimised of the Linux. Very tantalizing what is coming.

So, let's go a step further and think about this, if a kuckstarter was done, a modular desptop blade farm could be manufactured in partnership with a leading case/hardware manufacturer, at the sane time why not massive parallel RiscV+X coprocessing units (maybe with a partner like Nvidia with cuda, and using their fast interlink technology, or pci-e, whatever from board to board). Everything canbe a pci-e like/or extra high card to eliminate need for more then one motherboard (if at all). Such things could also plug into pci-e slots of a normal PC case, or hook externally to a pcie slots in a conventional PC (this is mainly if PCIe is chosen as the card interconnect, as Nvidias etc is faster).
 
As well as CFW interfacing (Cheap Fat Wide-Speed). As a cheap alternative to cutting edge interfacing techniques, you can simply use more lines of an existing cheap standard. For instance, more PCI-e lines from blade to blade for communications. Another example, not for this case, but for external use for instance, one could parrallel arranged TB 3/4 ports, or have a extendable TB port the same except with an extended section for more lines). So, 1+ Tb/s TB, or many Tb/s PCIe, is possible with some syncing delays. Its not a hardware sophisticated system, requiring a lot of software handling, but each card is able to do most of its work on card, as the data is highly parrallelable lots of local data in our space. The new mega mesh technologies are looking towards high distributed clould server data, and henceforth are over featured for out use, and over featured means performance/density/cost loss somewhere. Basically fancy cloud database, regular server, and parallel computing requires different mesh standards for optimization. A standard above could accomodate the use of any mesh standard at one time, with bridges to the others, so not an issue. Manufacturers of parts decide which interfacing standard to use, and people use comparable parts.
 
Did I forget, my case design for the Nvidia was way cooler than the MacPro, something you would be proud to have in your living room.
 
Thankyou for the heads-up Peter.

That is a lot of 'juice'.

+) This must surely be software limitation in not being able to effectively use the GPUs?
+) Perhaps it highlights that the end to end pipeline should be re-architected

REDs democratising of 8K could just give Apple the business justification to more heavily invest in this - especially if 40 Intel cores + 4 State of the Art GPUs are required in Linux.

When Apple update FCP for 8k R3D source material, with 4K UHD delivery, Apple will have an '8k target' to hit for MacOs.

I've been wondering how I would feel if a 64 core Apple machine came out built into a iMac.
'leaks' for the A10x core performance are equivalent to Intel's XEON.

Could Apple build a reference quality screen (the main cost), and throw in a 64 core chip that could probably handle the CPU side of 8k?
Would it sell?

AJ


Peter Majtan said:
After speaking with someone I respect at one of the SW manufacturers - he told me:

"For playback performance, we still need to do tests but its looking like you will need a Linux dual 20 core CPU and four or more Titan X Pascals for reasonable 4K performance of the 8K source clip."

We need to build a system that can handle the actual 8K, not a 4K "proxies" of the 8K clips. End of this month can not come soon enough. Bring it on Apple...
:reddevil:
Click to expand...
 
I find it hard to believe that 8K will require 40 cores and 4 Titan Pascals. I get realtime playback of 6K full debayer on an overclocked 10 core and one Titan.
 
Peter Majtan said:
After speaking with someone I respect at one of the SW manufacturers - he told me:

"For playback performance, we still need to do tests but its looking like you will need a Linux dual 20 core CPU and four or more Titan X Pascals for reasonable 4K performance of the 8K source clip."
Click to expand...

I disagree. Since I can play back dual streams of 6K R3D in a 4K timeline onto a 4K monitor just fine on less than half those specs ( 3 year old dual 10-core, 980TI GPUs).

Not sure about Apple. I hope they realize it's time to either lead or get out of the way entirely. If they're not going to offer up some new systems as the appropriate Intel CPUs come out, they may as well exit the PC space entirely at this point. It sucks and I really don't want them to, but they have completely lost sight of the pro market. I think too much focus on this whole "post-PC" mantra and building devices predominantly for consumption. Can't say it's because they're greedy. If they were truly greedy, they would realize they could be ruling the production end of the industry and not just the consumer end.

DJ Meyer said:
Curious where you read this? Apple has never released sales data to that level of detail to my knowledge.
Click to expand...

I have my sources. OTOH, the numbers are skewed. The 2013 nMP is the longest-lived Mac Pro model. The trash can has obviously not out-sold the composite of all models of Intel towers that came before it. What I find fascinating about the trash can is how small the secondary market is for them is. They have held their value quite well -- not on these forums -- but the ones I've sold on eBay are always gone in a flash and usually for more than I figured I'd get for them. Only a small handful of these stay on eBay perpetually and those are usually the goofy or strangely overpriced auctions. In certain professional and Edu circles, the 2013 Mac Pro is still the go-to system, even though it's been on the market for 2 years, 10 months and one week.
 
Jeff, playing back is one thing. Being able to actually do something with it in real-time is another. Think your average node tree in DaVinci. I have no reason to doubt the source...
You are right that the current nMP is a fantastic machine for many professions and that is why I think they will keep it in the lineup and just do an incremental upgrade of the innards.
But from what I can read between the lines from my own connections at Apple - they are working on a new workstation with drastic re-design of both HW and aesthetics.
This is why I think it will come in rack-mount form and separated bay for specialised cards. Resurrection of the xServe while giving it a power to handle almost anything via the expansion.

Again - we are less then a month away from finding out. Let's see...

:sifone: P
 
Peter Majtan said:
Jeff, playing back is one thing. Being able to actually do something with it in real-time is another. Think your average node tree in DaVinci. I have no reason to doubt the source...
You are right that the current nMP is a fantastic machine for many professions and that is why I think they will keep it in the lineup and just do an incremental upgrade of the innards.
But from what I can read between the lines from my own connections at Apple - they are working on a new workstation with drastic re-design of both HW and aesthetics.
This is why I think it will come in rack-mount form and separated bay for specialised cards. Resurrection of the xServe while giving it a power to handle almost anything via the expansion.

Again - we are less then a month away from finding out. Let's see...

:sifone: P
Click to expand...

Party... At long last the blade form of the workstation they talked about. But with my design you could get rid of the rack and vertically mount on the desk (you vsn have it both ways, except you have to be careful with venting). Be interesting to see what options there are that can be applied to them to achieve this. It's always be a limitation the boxes, that requires you to go even more expensive specialist if you fint wsbt to diy. However I forgot, I think they should go vertically stsckabe with.the mini too. I proposed such for a media hub/bluray module years ago.
 
Jeff Kilgroe said:
I disagree. Since I can play back dual streams of 6K R3D in a 4K timeline onto a 4K monitor just fine on less than half those specs ( 3 year old dual 10-core, 980TI GPUs).

Not sure about Apple. I hope they realize it's time to either lead or get out of the way entirely. If they're not going to offer up some new systems as the appropriate Intel CPUs come out, they may as well exit the PC space entirely at this point. It sucks and I really don't want them to, but they have completely lost sight of the pro market. I think too much focus on this whole "post-PC" mantra and building devices predominantly for consumption. Can't say it's because they're greedy. If they were truly greedy, they would realize they could be ruling the production end of the industry and not just the consumer end.



I have my sources. OTOH, the numbers are skewed. The 2013 nMP is the longest-lived Mac Pro model. The trash can has obviously not out-sold the composite of all models of Intel towers that came before it. What I find fascinating about the trash can is how small the secondary market is for them is. They have held their value quite well -- not on these forums -- but the ones I've sold on eBay are always gone in a flash and usually for more than I figured I'd get for them. Only a small handful of these stay on eBay perpetually and those are usually the goofy or strangely overpriced auctions. In certain professional and Edu circles, the 2013 Mac Pro is still the go-to system, even though it's been on the market for 2 years, 10 months and one week.
Click to expand...

Educational is the word. I can imagine students and tevently ex-studrnts snapping this up, having been exposed to them and they being trendy and a step up from iMac.
 
Now, redcode is basically similar to cineform raw, so I was going to calculate the processing power based on that. A 2Ghz core duo was able to record fullhd at 24fps when they were making diy cinema cameras. So 16:9 8kp24 should be 16 times more or less that to encode. 32 core duo's, what we have now is faster per cycle. But they do higher frame rates and bit depths, so what does that add up to? It shouldn't be 40 faster per cycle and higher frequency modern cores (maybe 20 core chips are clocked 2ghz +, or is this phi territory plus normal cpu's) plus 4 titan cards, but if you include soecusl effects and 8k cinematic quality 3D then it may be.
 
Bob Gundu said:
Considering an updated FCPX is debuting soon which will be able to work with 8k footage, I suspect new hardware should follow.
Click to expand...

Where'd you hear that?
 
Hmm, 4ghz=74.6496MB/s (metric megabytes). Let's be generous, and say there is a CPU that can perform twice as much work per cycle. So, that is 2 GHz times 16 = 32ghz at 8kp24 12 bits. So, that is like 10 cores. 8kp50 12 bit = 155.52MB/s, so around 20 cores, 16 bits around 26.66 cores (note, this is software and does not include GPU). If it was 4-6ghz it would be less, but I doubt we are going to see many cores on a chip that fast near term. However, let's say if it 6ghz, then that is 13.333 cores, or 4 sockets of 4 (dies that still exist). I'll assume the cinrform figures were at 6:1, maybe 4:1.

Of course things don't scale like that in real life, but worth a stab.

So, what would make redcode so processor intensive after the upgrade it received years ago (we believe using cineform technology). It is big money differences if it can be done on less than 40 cores and 4 titan cards. How come it can't be one CPU chip and one titan with gp-gpu techniques?
 
You must log in or register to reply here.
Back
Top