Looking under the hood: PTU, No. of Tracks & Computer Data Transfer Rates

[antonis]

After Avid announcing that they will be increasing the number or available voices it dawned me that perhaps I need to brush up my computing knowledge.

This of course has to do with the eternal question “do I need to upgrade my computer?”, the forever-anticipated arrival of a new Mac Pro, the need for higher track counts in immersive media and so on.

Other reason why going in to this detail is to gauge what could be the true power of a native system and make myself a more savvy audio equipment buyer. I am not going to imply that Avid had "handicapped" PTU with a lower track count from the get go. It is nevertheless, good to know what happens under the hood. I've been driving the car long enough. Now, I would like to know a bit more how the engine runs. ;D

Note: I might be off track here as computing is not really my strong suite. Also, I understand that some of the numbers below are theoretical and in performance varies.

Summing up:

• Calculating the data rate of a mono 24-bit/ 48Khz track is: 144 KB/s (Bits Per Second (bps) = Sample Rate (Hz) * Word Length (bits) * Channel Count)
• The current max voice count in Pro Tools Ultimate is 256 (subject for an increase).
• 256 voices x 144 KB/s = 36,864 KB/s = 36.844 MB/s. Hence, about 37 MB/s is the data rate requirements to stream 256 unprocessed mono tracks.
• I am using a HDN card, which apparently is x1 PCIe 2.0 (If I am not mistaken, think the same goes for the HDX card).
• According to Wikipedia, a PCIe (x1) slot has the transfer capability of 500 MB/s.

From all the above, it appears from a first glance that there is loads of bandwidth to spare! Enough to run over a thousand tracks.

Of course this must be an oversimplified scenario but would anyone with stronger computing skills than mine be able to answer some questions or put in more details to all the above?

1. If I am not mistaken each PCIe lane is bidirectional. If so, that means that a PCIe x1 lane a max 500MB/s to the CPU and from the CPU (?)

2. Would anybody know, how is the insert plugin processing affecting the bandwidth of the CPU to PCIe communication? For example, if an audio track has 5 inserts does that mean that the computing sequence is:

Audio track streaming from hard drive/ RAM Cache to the CPU -> Native plugin 1 processing -> Native plugin 2 processing -> …. -> Native plugin 5 processing -> PCIe DA? (perhaps somewhere around this process is where PT takes the delay compensation figures accrued from all the tracks).

I am sure that the way CPU instructions and execution are taking place is more sophisticated and I feel that I am putting this down with words somehow naively! Anyway...

3. If so, is the data transfer rate speed governed by the CPU clock or is there another intermediary component that affects it?

4. According to Avid, the FPGA does nothing on the HDN card when not recording. Surely, it must be doing something when outputting hence the lower latency numbers between HDN and a completely Native systems. Would anyone have anything on that?

[Darryl Ramm]

As you have shown PCIe bandwith here is, even a single PCIe 2.0 x 1 lane that the HD Native card uses, is not going to be a practical bottleneck. There may be more going on with overhead, such as some data padding, and there will certainly be practical limits introduced by the software architecture, the drivers will not be able to slam the data through at full bus speeds. But your numbers should be a good enough flavor.

Needing a more powerful computer or not to run Pro Tools all about native CPU plugin processing... doing lots and lots of floating point calculations, for many concurrent threads without too much latency to cause glitches. Fast CPU, lots of cores, and lots of memory (properly matched to the CPU/motherboard) and fast IO so the CPU never has to wait for that. Knowing you are unlikely to hit a bandwith limit is one thing, but the rest of this stuff is more black magic, better dealt with by finding out what people with similar workloads are running on.

The HD Native card FPGA is used for low-latency monitoring, so yes only used when "recording" stuff. Lower latencies have nothing to do with that FPGA and all to do with avoiding high-latency interface mechanisms (like USB) and lots of care over the software stack. Other modern interfaces exist that have lower latencies than Avid HD Native.

Darryl

[reichman]

Antonis:

I admire your effort to calculate theoretical maximums, but your are forgetting the most important variable in the equation: HDX is the flagship platform, and Avid can't let HD Native have better specs without confusing the product lineup.

Pro Tools Expert talked about this recently. If Avid let HDn have 2,000 tracks, and HDX couldn't be configured for that due to hardware restrictions, which platform would be the flagship? Neither.

[tom_lowe]

Originally Posted by antonis

After Avid announcing that they will be increasing the number or available voices it dawned me that perhaps I need to brush up my computing knowledge.

This of course has to do with the eternal question “do I need to upgrade my computer?”, the forever-anticipated arrival of a new Mac Pro, the need for higher track counts in immersive media and so on.

Other reason why going in to this detail is to gauge what could be the true power of a native system and make myself a more savvy audio equipment buyer. I am not going to imply that Avid had "handicapped" PTU with a lower track count from the get go. It is nevertheless, good to know what happens under the hood. I've been driving the car long enough. Now, I would like to know a bit more how the engine runs. ;D

Note: I might be off track here as computing is not really my strong suite. Also, I understand that some of the numbers below are theoretical and in performance varies.

Summing up:

• Calculating the data rate of a mono 24-bit/ 48Khz track is: 144 KB/s (Bits Per Second (bps) = Sample Rate (Hz) * Word Length (bits) * Channel Count)
• The current max voice count in Pro Tools Ultimate is 256 (subject for an increase).
• 256 voices x 144 KB/s = 36,864 KB/s = 36.844 MB/s. Hence, about 37 MB/s is the data rate requirements to stream 256 unprocessed mono tracks.
• I am using a HDN card, which apparently is x1 PCIe 2.0 (If I am not mistaken, think the same goes for the HDX card).
• According to Wikipedia, a PCIe (x1) slot has the transfer capability of 500 MB/s.

From all the above, it appears from a first glance that there is loads of bandwidth to spare! Enough to run over a thousand tracks.

Of course this must be an oversimplified scenario but would anyone with stronger computing skills than mine be able to answer some questions or put in more details to all the above?

1. If I am not mistaken each PCIe lane is bidirectional. If so, that means that a PCIe x1 lane a max 500MB/s to the CPU and from the CPU (?)

2. Would anybody know, how is the insert plugin processing affecting the bandwidth of the CPU to PCIe communication? For example, if an audio track has 5 inserts does that mean that the computing sequence is:

Audio track streaming from hard drive/ RAM Cache to the CPU -> Native plugin 1 processing -> Native plugin 2 processing -> …. -> Native plugin 5 processing -> PCIe DA? (perhaps somewhere around this process is where PT takes the delay compensation figures accrued from all the tracks).

I am sure that the way CPU instructions and execution are taking place is more sophisticated and I feel that I am putting this down with words somehow naively! Anyway...

3. If so, is the data transfer rate speed governed by the CPU clock or is there another intermediary component that affects it?

4. According to Avid, the FPGA does nothing on the HDN card when not recording. Surely, it must be doing something when outputting hence the lower latency numbers between HDN and a completely Native systems. Would anyone have anything on that?

All 256 voices aren’t being sent to the PCIe card on HD Native. In HDX the mixer is on the card, so all voices are being sent to it. On HD Native that’s done on the CPU so all that’s being sent out of HDN is the output tracks.

[antonis]

Originally Posted by reichman

I admire your effort to calculate theoretical maximums, but your are forgetting the most important variable in the equation: HDX is the flagship platform, and Avid can't let HD Native have better specs without confusing the product lineup.

Hey Reichman,

I can completely understand that. No qualms with the marketing people!

Only to be clear, the intention was not to post this to criticize Avid. Not at all. I've been enjoying several stable work systems with the current hardware lineup (well, despite some video and some network share issues) both on HDX at the work place and with my HDN. Pro Tools has been a terrific work tool for me and I would assume the same everyone around here.

This genuinely a geeky quest!

Originally Posted by tom_lowe

All 256 voices aren’t being sent to the PCIe card on HD Native. In HDX the mixer is on the card, so all voices are being sent to it. On HD Native that’s done on the CPU so all that’s being sent out of HDN is the output tracks.

That's even worse then!