Welcome to our community

Be a part of something great, join today!

Register 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 Mac Pro Announced at WWDC! - It's OFFICIAL!

Peter Majtan said:
For as long as you will keep making these hasty comments without putting things in perspective - you are bound to get responses like mine. I would like to see the "cheap gaming PC" you have mentioned to open up that 30GB texture set in Mari and work on it while being able to play the scene animation and see instant feedback, including all the image channels (such as bumb, specular, etc...)...
Click to expand...

When the Mac Pro comes out I'll do a comparison, OK?

And yes you can load amazingly large textures on a cheap gaming PC. Mari isn't loading the whole 30GB texture set onto the GPU. It dynamically loads what it needs. As long as you set things up so there is no bottleneck for it to load what it needs... you're good. The display of bump and specular is also not a particularly groundbreaking feat either these days.

Peter Majtan said:
But don't take my word for it - the guys from Weta and Pixar that made that presentation are facing daily challenges that 99% of us would never do. And they clearly have stated that Mari has never run smoother with such a set of data.
Click to expand...

Well, since you've worked in VFX, you'll know that older "power workstation" computers are not the fastest sometimes. Shiny new ones are often much faster. Especially for interactive tasks like painting.

It is a misconception to think that the big VFX houses are working on magically more advanced hardware or workstations. Most of the time, when a friend goes to work at a fancy VFX house, they come back with stories about horrible bottlenecks and outdated stuff... and really smart people who make up for that :) What makes a great VFX house great is the people. The hardware often is a little out of date, actually! Personally, every VFX house I've ever worked at has usually had slower equipment than my home systems. I was really surprised when IT gave me a PCIe SSD at work recently, actually.

So... you're comparing between (possibly older) systems that these guys have worked with... and a Mac Pro system containing stuff that's hasn't even been released yet. I'd EXPECT it to be faster for Mari painting work.

Is this supposed to be surprising?

But hey, let's find something we can agree on: I love The Foundry. I hope that this exposure helps encourage tons of people to buy the Mac version of Mari. I would like nothing more than for The Foundry to bathe in money - and pour it a little of into developing more awesome software. We can agree on that, right?

Bruce Allen
www.boacinema.com
 
Took time to think what apple has done to the pc market. Apple has succedeed in knocking, Hp and dell of the market. But not the whole pc market. Now the fight would be between apple vs gigabyte and asus. You know when u try to make a drastic change you fight your own. And they try to attack backn thereby opening an avenue for outsiders to come in. It happened with ios and android google helped the asians own the smartphone race. Android came to dominance becos apple loved the walled garden approach and microsoft was too slow which in this case is hp and dell. Look were motorola, palm are now. Now the smartphone is all about samsung, lg, huwaie. They gave better choice to consumers Same thing would happen here. Apple would always design its hardware around its software. Dat would put adobe in a dicey position and adobe would try to attack back. I predict adobe would start using all 32 threads in dual processors use all the resources gpu has to offer that would actually drive 4k adopotion cos 4k workflow would become really easy.they would actually try to make there stuff faster better than competiting apple product. by then Asus, Gigabyte supermicro would be pushing to the market far sexier designs . Making the asians compltely the producers of hardware . These companies would give far better options than apple. Now this is the catch if americans fail to accept them cos I know most europians will then most of american entertainment jobs would be outsorced abroad espically indians and chinese even europe they wold start taking all the vfx jobs, post production jobs in general bcos wen u give the asians space they democratize everything. Android was not popular becos it was open source but becos the asians embraced it. The fight would be between apple and adobe with the asians propagating the pc the truth is in the next 5yrs every movie out ther would have iron man/man of steel type of effects cos intel, Amd, Nvidia would continue pumping out more powerful chips there would a big boom in hardware among pros. U knw the way it stand now most powerful computers are luxry in places like china and india even a $1500 desktop cos but wen the asians take over, every one would have enough computing power and everybody would start rivaling hollywood. I think we are gradually come to the end of an era I think apple is the instrument with which american technology will be toppled. Just my personal opinionz.
 
thanks Bob woke up to your funny post ;)
 
Hahahhhahah @ Bob
 
Sergio Perez said:
Haven't gone through all the posts so sorry all if this was already posted, but where can I find this Pixar demonstration and talk with the new Macpro?
Click to expand...

This link references the video which is on Apple's Developer site:
http://www.macobserver.com/tmo/article/a-pixar-artist-on-testing-apples-new-mac-pro-w-out-seeing-it

Essentially, a Pixar texture artist says Mari runs better his beta Mac Pro than any other machine he's used. He says the machines given to Pixar people were housed/disguised in big boxes and he didn't actually see the machine until the conference. Pretty cool video!
 
Jeff Kilgroe said:
Yes. But there are two of those 20Gbps pipelines in each Thunderbolt port. Thunderbolt 2 also allows channels and even ports to be multiplexed for increased capabilities. Two Thunderbolt 2 ports multiplexed together has about the same throughput as a PCIe v2 X16 slot. Higher latency, of course, but still close to the same bandwidth. Each Thunderbolt 2 port is backed by 4 lanes of PCIe v3 at 985MB/s per lane. That's 3.9GB/s and real-world throughput is closer to 3.5GB/s, plus some overhead for the Thunderbolt host and whatnot. I don't think it's unreasonable to expect 3GB/s+ out of a TB2 port.

The real trick to all this is what devices will actually come to the table. The unfortunate side of this is that very few users, and even fewer devices, are going to need any of those multiplexing abilities or huge bandwidth. So I fear that we may not see some of these faster expansion chassis or mega-speed storage options that are possible. If we do, they will likely be niche products and rather expensive. Time will tell.
Click to expand...

I don't think you are correct here. First, Thunderbolt 2 is not PCIe 3.0. Its still PCIe 2.0. Secondly, there is only one 20 Gb/s channel in the Thunderbolt 2.0 specification, it is simply bi-directional (this is why there are two channels shown in the Intel diagram, one is upstream and one is downstream). So max theoretical bandwidth is nowhere closer 3.9 GB/s per port. With overhead from 8b/10b, you're looking at closer to 16 Gb/s peak performance, which is 2 GB/s.

It follows, then that two TB ports are lucky to get half the real world performance of a x16 PCIe 2.0 slot.
 
Last edited:
Looks like Apple isn't the only company releasing "canned products". D-Link just announced a high-end router that looks EXACTLY like the new Mac Pro. As Jarred would say..."Wowzers!". So if any of you are looking for an advanced home router to match your spiffy new Mac Pro, take a look at the DGL-5500.
 
Playing cards

Playing cards

Austin Wismer said:
I don't think you are correct here. First, Thunderbolt 2 is not PCIe 3.0. Its still PCIe 2.0. Secondly, there is only one 20 Gb/s channel in the Thunderbolt 2.0 specification, it is simply bi-directional (this is why there are two channels shown in the Intel diagram, one is upstream and one is downstream). So max theoretical bandwidth is nowhere closer 3.9 GB/s per port. With overhead from 8b/10b, you're looking at closer to 16 Gb/s peak performance, which is 2 GB/s.
It follows, then that two TB ports are lucky to get half the real world performance of a x16 PCIe 2.0 slot.
Click to expand...

In your analysis you cite overhead and note the "real world" performance of a bus vs the theoretical spec when evaluating TB2, but not for PCIe? I certainly don't have all the answers, shipping TB2 devices, both host and peripherals, are at the very least 3 months away. Not to mention we lack metrics on efficiency ratios when multiplexing the ports. If your goal is to defend the value of PCI card architecture, then I don't doubt its maturity vs TB gives it the short term advantage.

My counter would be that in a world of 22nm process chips, the form factor of the PCI bus requires more power, cooling and physical space than makes sense going forward. Only time will tell if Apple and TB2 will provide the necessary muscle for the "big data" future, but with nano tech on the doorstep I see PCIe as a dead man walking.

Cheers - #19
 
arguably all technology is a dead man walking, but we like standards and backward compatibility since we spend so much damn money on the ecosystems we invested in... one day the usb form factor won't make sense too I am sure, but for now I like that I can use all my old stuff with a new machine... As a freelancer its really hard to buy into anything these days unless we can make our money back, but with the never ending faster update cycles of cameras, computers, ports, and cords unless you buy wisely its a nightmare to try to make a ROI before parts are obsolete and you can't even move them on ebay at less then half price... its pretty vicious these days, and it just seems to get faster the more I spend...

I personally miss the days of buying a tool that lasted for quite some time. I have been trying to justify purchases more these days. But for me the whole reason of the Mac Pro being useful was the PCI expandability. Without it, I might as well buy a more mobile solution...
 
Blair S. Paulsen said:
In your analysis you cite overhead and note the "real world" performance of a bus vs the theoretical spec when evaluating TB2, but not for PCIe? I certainly don't have all the answers, shipping TB2 devices, both host and peripherals, are at the very least 3 months away. Not to mention we lack metrics on efficiency ratios when multiplexing the ports. If your goal is to defend the value of PCI card architecture, then I don't doubt its maturity vs TB gives it the short term advantage.

My counter would be that in a world of 22nm process chips, the form factor of the PCI bus requires more power, cooling and physical space than makes sense going forward. Only time will tell if Apple and TB2 will provide the necessary muscle for the "big data" future, but with nano tech on the doorstep I see PCIe as a dead man walking.

Cheers - #19
Click to expand...

As I understand it, the 500 MB/s figure for PCIe 2.0 includes encoding overhead. Which is why when you calculate the bandwidth of TB (4 lanes of PCIe 2.0), you get 2GB/s, which is 16 Gb/s, not 20 Gb/s.

I'm not trying to claim superiority of one standard over the other, just clarifying the facts. They are both different, and have different places as far as I'm concerned (although TB will probably replace PCIe in the long run). TB is hot swappable, carries display signal, can be connected via longer distances, but PCIe simply has higher bandwidth at this point. It is just a trade-off.
 
Brian Merlen said:
I personally miss the days of buying a tool that lasted for quite some time. I have been trying to justify purchases more these days. But for me the whole reason of the Mac Pro being useful was the PCI expandability. Without it, I might as well buy a more mobile solution...
Click to expand...

We're running a few Mac Pros here far longer than 5 and some 6 years.
Our XServe is older than that.
Im pretty sure our Mac centric gear has outlasted both Dell and BOXX units by over 4 years.
For us however, we will need to see the updates on the TB2 and peripherals before moving to the new Mac Pro.
Im so far impressed with what Ive learned to date.
 
Austin Wismer said:
As I understand it, the 500 MB/s figure for PCIe 2.0 includes encoding overhead. Which is why when you calculate the bandwidth of TB (4 lanes of PCIe 2.0), you get 2GB/s, which is 16 Gb/s, not 20 Gb/s.

I'm not trying to claim superiority of one standard over the other, just clarifying the facts. They are both different, and have different places as far as I'm concerned (although TB will probably replace PCIe in the long run). TB is hot swappable, carries display signal, can be connected via longer distances, but PCIe simply has higher bandwidth at this point. It is just a trade-off.
Click to expand...

That's pretty much correct in terms of PCIe 2.0 bandwidth. It's 500MB/s per lane, or 2000MB/s for an X4 host. Which is what drives Thunderbolt 1.0. With the 10bit/byte overhead, we get 1600MB/s or 16Gbps backing each Thunderbolt 1.0 port.

Austin Wismer said:
I don't think you are correct here. First, Thunderbolt 2 is not PCIe 3.0. Its still PCIe 2.0. Secondly, there is only one 20 Gb/s channel in the Thunderbolt 2.0 specification, it is simply bi-directional (this is why there are two channels shown in the Intel diagram, one is upstream and one is downstream). So max theoretical bandwidth is nowhere closer 3.9 GB/s per port. With overhead from 8b/10b, you're looking at closer to 16 Gb/s peak performance, which is 2 GB/s.

It follows, then that two TB ports are lucky to get half the real world performance of a x16 PCIe 2.0 slot.
Click to expand...

Thunderbolt 2 PCIe topology is still indeed based on PCIe 2.0. However, the Falcon Ridge controller is backed by PCIe v3. So I guess you could say that some semantics are getting in the way. I only said it was backed by PCIe 3.0, which is indeed the case.

If you want to get to the nitty gritty of Thunderbolt channels, Thunderbolt 1 has two 10Gbps channels. Thunderbolt 2 has 4 of them. By default, when the port runs in Thunderbolt 2 mode, the channels are bonded as 2 x 20Gbps channels per port. As for the upstream and downstream orientation, that is the intended operation, but in fact both channels are bi-directional capable, just as they are in Thunderbolt 1. When using a port for Displayport or video signaling, the secondary 20Gbps allocation is a one-way street, just as the secondary 10Gbps channel is when connecting a display to a Thunderbolt 1.0 host. When a display is connected to Thunderbolt 1, the secondary channel becomes video only and the primary channel has to operate in bi-directional mode. I have yet to see any complete diagram online that fully represents the operation of Thunderbolt 2, or Thunderbolt 1 for that matter.

There are several unanswered questions and some compromises along the way as well. Using Thunderbolt 1 devices on a Thunderbolt 2 port seems to cause the primary 20Gbps allocation to split into dual 10Gbps streams to accommodate the v1 topology. While all channels in a Thunderbolt port can be bi-directional, there is a significant penalty on the switching, so therefore the intended use is an upstream and downstream orientation.

Intel has released a lot of preliminary specs through their usual dev channels. I'm assimilating my info by what I'm being fed from a couple sources, trying to verify as much as I can. Full release doesn't happen for another week or three. Only their key integration partners have the info so far -- Apple, HP, ASUS... And Intel hasn't lifted the gag order just yet. Still waiting on confirmation of asynchronous matching and dynamic bandwidth allocation or load balancing of ports and other features.

Each TB2 port is backed by 4 lanes of PCIe v3, which is 985MB/s per lane for an aggregate of 3940MB/s or 3.84GB/s. Translate this with the 10bit/byte overhead and that works out to 33Gbps. When it's a "40Gbps port". So as with Thunderbolt 1, we are already seeing a port with insufficient backing to reach its advertised potential. If it holds true to the performance we've come to expect from TB1, we get about 93% nominal efficiency, so based on that, I expect a nominal throughput per port on TB2 of about 2.9GB/s... I was saying 3GB/s in my post above, but after I put some real numbers to it, it's not so rosy... Most devices are only going to see a little better than half of that as the ports will be primarily allocated as split up/ down streams with some efficiency loss.
 
Jeff Kilgroe said:
That's pretty much correct in terms of PCIe 2.0 bandwidth. It's 500MB/s per lane, or 2000MB/s for an X4 host. Which is what drives Thunderbolt 1.0. With the 10bit/byte overhead, we get 1600MB/s or 16Gbps backing each Thunderbolt 1.0 port.



Thunderbolt 2 PCIe topology is still indeed based on PCIe 2.0. However, the Falcon Ridge controller is backed by PCIe v3. So I guess you could say that some semantics are getting in the way. I only said it was backed by PCIe 3.0, which is indeed the case.

If you want to get to the nitty gritty of Thunderbolt channels, Thunderbolt 1 has two 10Gbps channels. Thunderbolt 2 has 4 of them. By default, when the port runs in Thunderbolt 2 mode, the channels are bonded as 2 x 20Gbps channels per port. As for the upstream and downstream orientation, that is the intended operation, but in fact both channels are bi-directional capable, just as they are in Thunderbolt 1. When using a port for Displayport or video signaling, the secondary 20Gbps allocation is a one-way street, just as the secondary 10Gbps channel is when connecting a display to a Thunderbolt 1.0 host. When a display is connected to Thunderbolt 1, the secondary channel becomes video only and the primary channel has to operate in bi-directional mode. I have yet to see any complete diagram online that fully represents the operation of Thunderbolt 2, or Thunderbolt 1 for that matter.

There are several unanswered questions and some compromises along the way as well. Using Thunderbolt 1 devices on a Thunderbolt 2 port seems to cause the primary 20Gbps allocation to split into dual 10Gbps streams to accommodate the v1 topology. While all channels in a Thunderbolt port can be bi-directional, there is a significant penalty on the switching, so therefore the intended use is an upstream and downstream orientation.

Intel has released a lot of preliminary specs through their usual dev channels. I'm assimilating my info by what I'm being fed from a couple sources, trying to verify as much as I can. Full release doesn't happen for another week or three. Only their key integration partners have the info so far -- Apple, HP, ASUS... And Intel hasn't lifted the gag order just yet. Still waiting on confirmation of asynchronous matching and dynamic bandwidth allocation or load balancing of ports and other features.

Each TB2 port is backed by 4 lanes of PCIe v3, which is 985MB/s per lane for an aggregate of 3940MB/s or 3.84GB/s. Translate this with the 10bit/byte overhead and that works out to 33Gbps. When it's a "40Gbps port". So as with Thunderbolt 1, we are already seeing a port with insufficient backing to reach its advertised potential. If it holds true to the performance we've come to expect from TB1, we get about 93% nominal efficiency, so based on that, I expect a nominal throughput per port on TB2 of about 2.9GB/s... I was saying 3GB/s in my post above, but after I put some real numbers to it, it's not so rosy... Most devices are only going to see a little better than half of that as the ports will be primarily allocated as split up/ down streams with some efficiency loss.
Click to expand...

I suppose Anand over at Anandtech could be wrong, but he's usually prety spot on. According to him, there's no increase in aggregate bandwidth. It's simply 4 channels (unidirectional as I gather, 2 bidirectional channels -- the wording is a little confusing) combined into 2 channels, for no net gain in bandwidth. Where the advantage comes from, as I understand it, is that either data or displayport can use the entire channel, as opposed to TB 1, in which neither could claim more than a single 10 Gb/s channel. Again, feel free to take a took at the article here, that's what I'm going off of.

http://www.anandtech.com/show/7049/intel-thunderbolt-2-everything-you-need-to-know

I'm not entirely understanding the difference between a port being backed by a version of PCIe, but based on a different version. I could certainly be speaking out of turn here, but since SOMETHING is running on PCIe 2.0, it would seem like performance would go down to the "least common denominator."
 
I wonder if you can hook-up a Mr. Fusion via Lightning, er...Thunderbolt to generate the 1.21 jiggawatts

Matt Gottshalk said:
back-to-the-future-Doc-is-refueling-Mac-Pro.jpg
Click to expand...
 
I would take these benchmarks with a large dose of salt. They may be extrapolated, as in false. If legit, the software support and chipset drivers, etc.. are not yet finalized. That said, a system based on the Ivy Bridge Xeon with 12 cores in contrast to a current 12-core system, showing about a 10% increase in performance would seem about right, if they were clocked the same. This new Mac Pro they are supposedly benchmarking is clocked slower, so not sure what to make of it yet. Not going to read much into this yet as the origin of the benchmark and the state of the hardware are both of dubious nature.

Probably the biggest issue I see facing the new Mac Pro is not the expandability/ connectivity, which most people seem to be questioning. Nor the GPUs. It's the apparent single-CPU design. We should be able to drop a pair of those 12-core CPUs into a current HP Z820 (or similar) workstation with no other requirement other than a firmware update.

OTOH, the new Mac Pro is so compact, we can actually achieve similar CPU density to 1U blades when we rack these things. Most apps can't touch more than 2 or even 4, let alone 12 cores at a time except for when doing broad scale computational tasks or renderings. Apple really may be on to something here, but I think it's too early to tell for sure. Intel has their hand in it as well.
 
Last edited:
You must log in or register to reply here.
Back
Top