You say yourself that your LatencyMon values (btw kernel timer is measured in µs not ns) are not great.
TBH .. I would bring this into shape 1st before complaining about the RayDAT.

With my last 2 mainboards I get kernel timer between 2-10 µs with a few peaks around 20µs.
Make sure you use LatencyMon 4.02 and that you measure on an idle system.

You have not only to disable EIST (in fact EIST and SpeedStep I have enabled) but you should definiticely disable
that the CPU is able to enter energy saving states below C0/C1. So disable any form of C- and T- States.
EIST (P-States) you don't need to worry about, but of course you can disable it.

Even if you are not capable of speaking german this article could help a little otherwise try google translations.
The screenshots speak for themselves.

https://www.tonstudio-forum.de/blog/ind … -X10SRi-F/

RayDAT will do better, but the spikes of 500 µs are too high for my taste. Around 50µs with peaks under 100 I would regard as acceptable.

I have suspicion that it has to do with internal Intel graphics. On a Laptop this caused clicks and pops when browing the Web with Firefox. Simply scrolling down a loaded internet page by 1-3 lines created a click. I had to compensate this with a ridiculous high ASIO buffer size of 512 for pure stereo playback of FLAC file in Musicbee (using RME UFX ASIO driver).

It turned out that click and pops went away when using the internal nVidia Graphics for Firefox, then I could use the lowest ASIO buffersize which is possible again (48 on UFX). Since then I am not a fan of Intel Graphics inside of CPU.

Get i.e. this card with zero fan feature, you wont hear anything under office / recording operation:
https://www.heise.de/preisvergleich/msi … 14302.html

Get a second hand Radeon HD3450 or HD5450. Cheap silent low power and work 100% here.
You don't mention at cpu...So difficult to say. But remember that other audiocards often use triple buffers so latency is in reality much higher (cubese reports the latency correct) I can work a little at 32 but not to much 64 is fine 128 is optimal. And since there are no hidden buffers with RME (which might explain some of the loss in performance) 128 is re3al low latency with this card.
Good luck!

I actually tested the round trip latency using a loop from line out to line in, then zoomed in and measured samples. At 64 samples buffer size the M-Audio USB is running at 6.8ms total round trip, the RME is 4.2ms, which is better - but the stability is basically the same, thats the real issue.

I would take 6.8ms with significantly better stability at high plugin counts over 4.2ms with the same stability at the same plugin counts as a shitty old USB2 interface. Thats what I bought the RME for, thats what I was expecting/hoping for.

I currently have it plugged into a PCIe x16 slot, should it be in something different?

Theres currently no othe PCIe cards installed, though ramses suggested I should get a dedicated gfx card to unburden the CPU. Would you agree with that? I dont use the computer for anything but audio, no internet or gaming, so I was thinking of a fanless cheap card like this http://www.gigabyte.com.au/products/pro … x?pid=5819

Is that a good idea?

True, I will try this. Do you have your RayDAT in a x1 slot?

This is my motherboard: http://www.gigabyte.com.au/products/pro … id=4978#ov

It has PCIe x16, x8, x4 and x1 slots.

Also, ramses, Im curious to know, at 64 samples buffer size with your RayDAT, how many tracks with an instance of Guitar Rig 5 running, say, the Styles > Blues > Billy Dual Grange preset can you run simultaneously? Are you able to test that?

Because currently, with both my RayDAT and my old USB2 M-audio, I can get to around 8-10 tracks before it starts crackling

That was an interesting read. One big takeaway from it was to uninstall my Intel SATA drivers and use Windows default drivers.

Now LatencyMon is reporting a highest DPC and ISR of around 70, and generally hovers around 10-20! Even the pagefaults number has gone down a lot. The highest pagefault so far only fills half the bar, instead of going off the charts into the red like it used to.

I will do some audio tests again, but hopefully this improves things a lot.

Id still like to hear back from MC about which port is ideal to plug the RayDAT into on my motherboard. And also, getting a dedicated gfx card does seem like a good idea, to take load off the CPU for gfx processing. Using the CPU for gfx is probably why I get pops when doing GUI updates like opening/closing dialogues, moving windows around etc.

There's nothing I can look for in the specs to indicate which one is ideal?

start with the x1 slots!

Oh OK! Always use RME so... never had to check!

Agreed! But there are a few plugins that won't run nicely at low latencys. A few Kontakt instruments crackle because of their scripts. Maybe there are more. But I agree, for testing I loaded my cpu up with a heavy cpu wise multiband compressor and I could load it up to 99% at 128. At 32 a little lower but still up to 80% (from memory).

Just looked at the card .. you go for passive cooled cards.

I wouldn't do that. They produce unnecessary heat inside of the case
which at the end requires more work for blowers -> noise.

When I started building a silent recording workstation nVidia came out with Maxwell series.
If you buy a card, which is not overclocked, then these cards are very energy efficient.

And the best of it is, if the card supports the so called "Zero Fan Feature", then the graphic cards blowers won't turn on if it stays under 60°C. Then it depends on the implementation of the vendor in its GPU BIOS, what happens once the blowers started with temperatures over 61°C. In my case (980) the blowers turn off, once the card reaches again 50°C.

To put an example. I bought the GTX 970 and the GTX 980. After some experiments, how to organize blowers in the case and which blowers to use, I am now in the very fortune situation, that  the GTX 980 stays cool under 60°C for all use cases of my system: Office, Recording, Video Editing, Gaming.

I can recommend the MSI series of GTX9xx cards, as their blowers (shall they run) generate the least noise. I saw a video of around 8 or 10 different GTX cards, a mic in front of them and the noise that came out of the MSI card was lowest and didn't generate a nerving tone others were not so nice.

Based on this I would recommend to get a MSI GTX 950. As the new cards are out you can get the older ones a little bit cheaper with a little luck. I heard at a certain day in the week computer parts are less expensive in shops.
https://www.heise.de/preisvergleich/msi … 14302.html

With this card you should have absolutely no problems in terms of noise and heat.
It offers you to connect up to 3 monitors in parallel using Display Port.
HDMI and DVI port are also there if required.
The TDP of this card is 90W.

Matthew, what CPU do you have ? You didn't tell.

Intel i7 4790 3.6GHz

Its live playing, using Guitar Rig 5 with the Styles > Blues > Billy Dual Grange preset. I create 1 track with an instance of Guitar Rig 5 with that preset loaded, and set the input to ADAT input 1, with my guitar plugged into the pre, arm it for live monitoring, then duplicate the track until I get pops/crackles. I get to about 8-10 tracks before it crackles on both the RayDAT and the M-Audio USB. I set the volume very low on each track so the summed audio doesnt clip the master.

Oh and I have "High Quality Mode" disabled in Guitar Rig.

The older cards get hotter, nVidia was starting with Maxwell chips to introduce powerful energy efficient chips.
And in that area they are better than ATI as they are able to produce the thinner structures (28 nm) to enable for less TDP.
I only recommend you the separate Graphic card, because in a Laptop the GPU inside CPU turned out to be crap, see my comments which I wrote to this topic. Maybe on a Notebook this effect turns out stronger, but I have my doubts that this is furtune for the Desktop.

A GTX 950 is fully sufficient, I only want that you get a real GPU as the internal GPU inside of your CPU has no memory of its own, so for graphic access it uses part of your DRAM. This design is maybe nice from cost savings perspective.

But not for high performance computing with handling real-time audio loads on an Operating System (Windows) which is not designed as realtime OS .. go figure.

If you are casual gamer, feel free to spend more bucks.

Current temperature of CPU / GPU on my system: with Xeon E5-1650v3 / MSI GTX 980 4G Gaming
Windows High Performance settings:
CPU 38°C / GPU 42°C, on since hours all blowers set to low on fractal design case

When doing Video Rendering with Vegas:
CPU load 64-73% Mem 7/32GB, GPU Load 0-14% and 700MB/4096MB memory usage
CPU 52°C / GPU 50°C

When doing Gaming additionally (on top of the rendering):
CPU load 70-85% Mem 9.5/32GB, GPU Load 0-14% and 1500MB/4096MB memory usage
CPU 58°C / GPU 50°C
Game setup for full quality, Vertical Sync to 60fps, same ping time, no packet loss.

Again why not a HD5450 AMD? Zero issues here and not hot.

I will try that tomorrow!

Thanks, I look forward to hearing how you go. Could you let me know your PC specs when you reply?

Wow that is hot! I dont want that either. I tend to agree with your preference for nVidia, I will look into some other options, cheers

You started talking about a fanless/passive cooled gt 710 to which I referred to with the link and the specs.

Now you come and argue with a hardware review of actively cooled GTX 750 Ti cards which are different.

Here a quick categorization of nVidia product names to "sharpen" your eyes where nVidia graphic cards differ:

G or GS - The standard video cards with core clocks that are not high and are made to keep a minimum performance.
GT - Slightly higher core clocks and possibly more onboard memory than the GT, It is mainly a budget performance card.
GTS - Includes possibly more onboard ram and higher than average core clocks possibly more pixel shaders or pipelines to increase performance.
GTX - Has the most memory of the group and higher clocks than all previous also includes more pixel pipelines and usually outperforms the GTS by atleast 10 - 15 percent.

Ti at the end means more performant model, i.e. because of more shader units.

Here an overview about the GeForce 700 series from 2013:
https://en.wikipedia.org/wiki/List_of_N … 700_series
Here an overview about the GeForce 900 series from 2014:
https://en.wikipedia.org/wiki/List_of_N … 900_series

I wouldn't take any passively cooled card anymore. If you want that, its your decision.

Then I misunderstood you, sorry. What is your budget for a graphic card ?

AFAIR the zero fan feature has been introduced with the more energy efficient 900-series.

So taking a 7xx card won't give you a silent card. But this should still be the goal I think.

Get a used MIS GTX 950 then ...

Oh btw, thank you for all the advice so far. I do appreciate it, and I am taking what you say on board.

Ok thanks, I will see if I can afford the 950 or find a used one in good condition for a good price.

Im hoping once I sort the gfx card, switch the RayDAT to the correct PCI slot (probably one of the x1 slots tbh), and go through that long pdf of pc tuning tips, I will start to see the fabled RME performance I have been hoping for

Im still very curious to see how vinark goes with those Guitar Rig 5 plugin count tests

Wow. That is SO much better than my 8-10 tracks at 64 buffer size. Especially since I have 16GB DDR3 1600 RAM and a non-overclocked CPU that is running natively at the speed of your overclocked CPU.

That is what I was hoping for when I got the RME, lets hope putting the RayDAT in the right PCIe slot, combined with switching to a dedicated card, and doing those PC tuning tips means I can finally see that kind of performance.

Thanks so much for doing that test

Well with ram it is very easy. If you have enough and enough bandwidth for a certain task, more won't do anything. In this case running 40 or so tracks maxes out the cpu before Ram bandwidth and usage.  The fact that our CPU's both run at 3.6 makes them quit equal I think. Generation improvement have been a few % only, yours is 3 or 4 later I believe.