Erik Lindahl
Forum Replies Created
-
Yeah I saw that clip the other day. I have however read numerous reports of very poor speeds, or rather unreliable speeds over TB networking.
-
[Gustavo Bermudas] “No, it’s the same aggregate as TB1, that graph can be misleading.
ATTO released an extensive document detailing the differences”
Yes, TB1 and 2 have the same total bandwidth. Even the ATTO document states this.
“Thunderbolt exists as (4) fully independent 10Gbps channels ‐ two upstream and two downstream (see left side of diagram below).”
“Thunderbolt 2 enables (2) 20Gbps bi‐direction channels (see right side of above diagram) instead of (4) of 10Gbps channels.”
Intels diagram and what ATTO writes sounds quite crystal clear to me. Even the second link states exactly the same thing.
“Thunderbolt 1 has 4 independent 10 Gb/s channels, Thunderbolt 2 combines these to give 2 20 Gb/s bidirectional channels“
-
[Gustavo Bermudas] “For what I understand TB2 is the same as TB1, with the only difference that TB1 has two 10 gig channels, one for each direction, and TB2 combines those 2 channels into one stream if it’s unidirectional.”
That’s incorrect I think.
[Anandtech] “Thunderbolt 2 provides that solution. By combining the channels together, Thunderbolt 2 enables two 20Gbps bi-direction channels instead of two sets of 10Gbps channels. There’s no overall increase in bandwidth, but the solution is now more capable.”
Aanandtech.com: Thunderbolt 2 Everything you need to know
Even looking at Intels info we get the same data.
Thunderbolt 1 has 4×10 Gbps – two upstream, two downstream
Thunderbolt 2 has 2×20 Gbps – one upstream, one downstream -
I think Thunderbolt is bi-directional so a total of 40 gbit/s. In reality with overhead taken into account I think this translates to a peak of about 1.5GB/s (1500 MB/s). So, uncompressed 6K would handle roughly 13 fps fully uncompressed, 4:4:4 16bpps if the previous figures are correct.
Not sure that’s the case with RED RAW. Downstream is much lower as I presume the RAW-file is sent at its 1/5 the size or similar (i.e up to 60 fps). Back to local storage, if they actually do send it 100% uncompressed you’d be stuck at about 50% of real time. I’d image even lossless compression could bring the file size down by 50% almost giving realtime.
All in theory of course.
If compression and scaling is handled by the card, going to anything below uncompressed 4:4:4 16bpp RGB 6K (which to be fair is what most people do) should give realtime performance. At sort of depends where things happen in the pipeline. Most people would probably go from 6K to 1080p where I’d imagine scaling is done on the card hence only a 1080p uncompressed stream comes back over TB.
Playback of R3D media even at 6K should be realtime as downstream we’re talking 1/5 of the bandwidth.
I also think Thunderbolt 2 supports port aggregation. Linking two ports could potentially handle 6K uncompressed, in theory.
-
Over at the Lift Gamma Gain forum they did test the new MacPro against a 2010 modell with various GPU’s. Sadly the nVidia GTX 780 wasn’t able to perform the Red Cine X export due to some OpenCL error. The new MacPro however performed very well, especially the D700 machine.
6K debayer to ProRes 4444 / 23.98 fps:
2013 12c 2xD700: 29 seconds
2010 12c 2x280X: 39 seconds
2013 6c 2xD500: 52 secondsIt would of course be interesting to see how an iMac would fair in the above. It does however have a much slower GPU than the machines in the test. It’s clear Red Cine X today only uses ONE of the two GPU’s however as the old 2010 12c machine got the same result with one 280X card.
-
What I have to agree with in the arstechnica review is the lack of dual CPU options. This would open two lacking feature:
A) more peak multithreaded performance.
B) cheaper multithreaded performance.
C) cheaper upgrade options and more peak RAM.As they note a dual 6-core system would be cheaper than a single 12-core system and have supperior single and multithreaded performance. Dual 8-, 10- or 12-core systems would push performance way off the current machines charts. This really is a design flaw from Apples point of view. Seeing technically how well the single CPU design works even at full load with both GPU’s Apple surely could have solved this with out to much headache.
It would be interesting with a system that has four boards rather than three where the end-user has more options.
– choice of 1 or 2 CPU’s
– choice of 1 or 2 GPU’s
– choice of using unused slots for internal disk expansion (seeing general PCIe cards would be virtually impossible).This would open up the current hardware to a wider audience for sure.
I also find the GPU performance issue very alarming. Apple and / AMD really have to step up there game here. Clearly apps like FCPX utilize the machines potential but third party devs need this available to them as well. If the case is as bad as A mentions – up to 40% less perfromance in OSX on top of at best 50% of “top CPU choices” – Apples workstation has a bit of a problem.
The MacPro 2013 is a pretty interesting development. I’d love to see this grow into other product la given Apple tends to draw very clean lines between product categories. For example the high-end 4- and 6-core chips in the consumer market with consumer grade GPU(s) would make TERRIFIC lower end work or gaming systems. The Thunderbolt design would however be a problem here.
All that said, the MacPro 2013 holds a place for many. People just need to understand what the system is and what it is not. Apple also needs to “get with the workstation program” and sort out some soft and hardware features.
-
I think they are identified as Radeon 7970’s or R9280X under Windows which is normal consumer grade cards, given they do have 6GB VRAM. It’s a marketing gimmick I’d say. And it’s a bit sad the MacPro got stuck with a 2 year old card even if they boosted it with 6GB VRAM. The R2 290X would be quite a bit more powerful but they might be running on the hot side for the small case.
Not bad cards but lower-high end consumer cards with boosted VRAM.
-
Depends on what you do. Realtime FCPX-work it might not be optimal or work at all. However, imagine being to send off a comp in AE or a cut in FCPX where the secondary station does all the work. The only thing that would affect the main station is i/o.
For me having that type of setup would be optimal IF it actually worked and wasn’t a setup nightmare.
-
Per dollar the MacPro 2013 is competetivly priced but you simply can’t get a dual CPU (24 core) machine with 512 GB of RAM and four GPU’s. Heck, you can’t even get a MacPro with nVidia GPU’s.
Then again you can’t get a PC with the specific size and spec the MacPro holds. In general these are huge systems.
The theory of connecting multiple MacPro’s via TB2 would be interesting but as far as I know TCP/IP over thunderbolt is very unreliable. But the idea of a main machine and offload machine could be interesting for a certain audience.
-
A base 12-core machine costs $6500 yes. It should also handily beat any previous Mac. However, Arstechnica is very correct in the fact that the 2013 MacPro can’t compete with the higher end PCs.
