Lowest latency at high CPU PCIe soundcard recommendation <$300

[Pokeyoats]

Good Day all,

I was wondering if someone could advise me of a PCIe soundcard that functions at the lowest latency at high CPU loads. An ideal goal, would be a card that could function at 256-512 byte buffers and maintain a crackle free latency under 10ms.

As I have a Virus TI in the fray, the ASIO drivers would also need to use ^2 buffer sizes (such as 128, 256, 512, 1024).

It also needs to have good drivers for Windows and OSX (dual booting).

My budget is under $300 (therfore RME unlikely) and if anyone could advise me of this I would be VERY grateful.

Following are the relevant hardware specs for the system:
CPU: i7 3770K
Motherboard: GA-Z77X-UP5 TH
RAM: 32GB (1600 MHz)

I asked this same question on the Access Virus forums and was advised to get a Native Instruments Komplete 6. I did, and I am quite disappointed that I followed this advice.

The card struggles at high CPU loads and even though Native Instruments have pointed me to beta drivers that do improve it quite considerably I am still just not very happy with it.

It is certainly useful for MIC input but it did not meet my low latecy/high CPU goals.

I really appreciate any advice that anyone could be kind enough to give.

[Kaine]

Your not going to get any better at that price point I'm afraid. The Lynx AES 16 has the lowest RTL at 256 and even then it can only hit 12.2µs at that buffer setting and is going to cost you 3.5X your budget. If you want the lower RTL setting your going to have to raise your buffer, the NI should hit 10µs at a 64 buffer setup.

[elcabong48]

Your motherboard has a single PCI (aside from PCIe) slot. If it's not already being used, the PCI Maudio audiophile 192 would work well at 128 and maybe even 64. It's an older card and is pretty cheap now, ~$100 or so I think. I had a much slower dual core system, and that card had very low latency and great sound quality. I used to get great performance at 128.

I don't know why more manufacturers haven't jumped on PCIe. It might be that with cards on the internal bus, you have to be careful about how you power up your keyboard or speakers to prevent spikes through the system. With USB you only have that single USB cable so voltages to speakers, keyboards, etc. can be more easily buffered from the PC.

[tenshin111]

I don't know why more manufacturers haven't jumped on PCIe.

I have wondered the same thing. There is very little to choose from when it comes to PCIe interfaces.

[digidennis]

http://www.creative.com/emu/products/pr ... &pid=19169

A really great card for a very reasonable price, with quality converters and drivers. Unfortunately windows only

[Pokeyoats]

Well this is becoming quite the dilemma for me.

The problem is not what the absolute lowest latency is for a sound card. The truth is that I can get my Komplete 6 running at 32 buffers and around 3ms latency. Which is fine for one Ableton and one VST. The problem I am having is that such a low buffer is not sustainable when any form of processing is going on.

The Komplete 6 has actually been quite a nightmare. At 50% CPU and above it was crackling and carrying on, even at 1024 buffers and >50ms latency. NI have provided some new beta drivers that have allowed the card to run a bit better at 512 and 80% CPU, but there are still some clicks.

What I find rather troublesome is that I can use my Virus TI at >120% CPU (which is what Ableton reports + Win CPU - about 93% system total real CPU). At 512 it offers a total ms of 38, and I guess I could continue to use it as the primary sound card. However when being used as a sound card you lose the ability to have any form of input to it for its vocoder/Atmozier.

Likewise, I have gotten my Creative Audigy ZS Platinum Pro to run at 3ms without a crackle using Massive/Rapture/Dimension Pro as a standalone device outside of Ableton. But in the case of the Audigy it crackles and severely misbehaves when the Ableton CPU meter hits 60%.

So with all of the sound cards I have, I certainly have a situation where a Virus TI where can maintain a pure, steady sound at 100% system CPU over USB and thought, SURELY there is a PCIe card that could perform similarly but with lower latency.

And I guess I say PCIe because the Audigy 2 ZS is a PCI card and has it's problems at high cpu, and although I understand that the Sound Blaster cards are not considered to be very good for audio production I haven't had the opportunity to have a good experience with another PCI card and thought I would hope on the improved performance of the PCIe bus.

[Kaine]

I don't know why more manufacturers haven't jumped on PCIe.

Because people started moving over to laptops making the production of internal cards more of a niche product. You'll notice that most of the cards based around PCIe under the £300 mark are all based around older designs that have been mearly updated for for the newer slot rather than pump money in to new product designs.

Well this is becoming quite the dilemma for me.

The problem is not what the absolute lowest latency is for a sound card. The truth is that I can get my Komplete 6 running at 32 buffers and around 3ms latency. Which is fine for one Ableton and one VST. The problem I am having is that such a low buffer is not sustainable when any form of processing is going on.

You can ignore claims of 3ms in the drivers unless your discussing it only going either in or out. Have a look here : http://dawbench.com/images/dawbench-llp-09-12-2.jpg for some independent testing. Only the lynx and the RME AES can hit a total RTL time of under 3µs. If you look at the column to the left labeled I/O you can see a few more can get in the 3µs region for either input or output but the total RTL is far higher than claimed.

Of course depending on what your doing defines how important it is. If your writting in the box then only the output time is going to be all that important, but if your recording and monitoring in real time the RTL total is going to be a key factor as anything over 10µs starts to become awkward for monitoring drummers/guitarists.

The Komplete 6 has actually been quite a nightmare. At 50% CPU and above it was crackling and carrying on, even at 1024 buffers and >50ms latency. NI have provided some new beta drivers that have allowed the card to run a bit better at 512 and 80% CPU, but there are still some clicks.

TBH it sounds like you have another issue in there somewhere. On a 3770k you should be seeing far better performance out of that unit. We hit 80%/90% cpu load before anything breaks up running at a 32 buffer. Try researching to see if the are any more tweaks you can do to that rig, my first guess would be turning the power options to "performance" in windows and disabling as much as you can in bios (including C states, eist etc....) if you haven't already. I can't give you any specific advise as I moved away from GB a few years back, but I did so because they could be overly fussy at times, which is perhaps what is manifesting here.

What I find rather troublesome is that I can use my Virus TI at >120% CPU (which is what Ableton reports + Win CPU - about 93% system total real CPU).

Ableton's load figure takes into account cpu/drive/memory loading part of which is the CPU load itself, so you can't really add them together to get a figure.

[Pokeyoats]

Ah, I feel I should explain, the 3770k system isn't built yet. I actually just got the loan approved yesterday for the system I am building and equipment, thus the inquiry as to a new sound card. Currently I am just running on a Q6600.

I have actually been wondering if the improved performance of the system will help the Komplete 6 in any way, but I was being confused, so to speak, by the Virus being able to perform flawlessly at high CPU and really was wondering if it was poor hardware in the Komplete 6 or the nature of its USB.

I understand that the Virus is actually USB 1.1 which I found even more curious given that it was performing without issue with such low requirements but the Komplete 6 couldn't.

The link you provided is fantastic, and I am going to read through it, perhaps though it is worth me re-testing the Komplete 6 once my new production system is all built.

[aMUSEd]

I don't know about you but I get incredibly good low latency ASIO using my Virus TI as a sound card. Better in fact than with my NI interface which is a Kore v 1 (same drivers and chip as the K6).

[Pokeyoats]

Absolutely, the Virus performs incredibly using ASIO. I can hit 80% CPU at 256 buffers without a crackle or a pop.

It's just that using it as a sound card disables input so that you can't use it as a Vocoder, or it's Atomizer. Which really isn't the worst thing that could happen.

The Virus is awesome at HIGH CPU loads, but it's not particularly fast, I believe the output latency at 256 is about 30ms but this isn't terrible by any stretch.

My original desire to get the Komplete 6 was in the hopes that I could use it as a sound card that coped at high CPU with lower latency allowing me to use every function on the Virus. It seemed like a simple solution, I just didn't realize how variable performance was between sound cards/devices.

It's starting to look like I should save some money (I'll buy more plug-ins, lol) and keep using the Virus as the primary sound card and use the Komplete 6 for MIC input.

My only other gripe against the Komplete is that the Windows drivers are so bad, it even crackles if you're playing a game (and I do like to play the occasional game on my system - OS's are isolated, I have one OS on one partition purely for Music and another general purpose boot on another partition for everything else.)

[keel]

I have focusrite saffire and the asio buffer size is 10.0 ms, then there is FireWire buffer 2.0 ms. Input latency is 11.6 ms and output is 15.5 ms. I'm just wonder what these are in those numbers? Like asio4all, it has 128,256,512 and etc. Is these my settings like 128 or 256? Or what? I'm just wondering because I run out of CPU power quite fast, even I have fast pc... Thanks!

[Pokeyoats]

I would not recommend ASIO4All at all, which I think is your problem.

In all of my dealings with it it consumed way more CPU then it should.

And I believe that the devices you're using have native ASIO drivers, another reason to ditch ASIO4ALL; which is primarily of use for people that have sound cards without ASIO drivers.

[keel]

I didn't say that I use asio4all, and no, I don't use it. I'm using the saffire's own asio. I'm just wondering how those ms and numbers are compared together. Nothing else.

[Pokeyoats]

Ah, I think I see. If I reiterate what you already know, my apologies.

The buffer is a, well, buffer. The size determines how many times the CPU needs to read from memory and move data across the bus. The absolute BEST description I have seen of how an Audio Buffer works is this from:

http://thestereobus.com/2007/12/13/deal ... io-buffer/

"Audio buffers exist in order to compensate for moments when the computer's CPU is too distracted by other things going on to keep up with the task of processing audio. Think of the audio buffer as a bucket of water that fills up from a stream. There's a hole in the bottom of the bucket out of which you get your water. In a situation where the stream is diverted and the water stops flowing into the bucket, it will be a while before the water in the bucket has collected is used up. If the stream can be redirected back into the bucket in time, you will experience an uninterrupted flow of water. If the bucket didn't exist and the stream were diverted, the water would stop immediately. This is the purpose of the audio buffer… it's the bucket and the audio stream is the water.

The larger the bucket you have, the longer it will take for the water to be depleted if the stream were to stop flowing. This means that the stream can be diverted for longer before you notice it has stopped flowing. This is what's called a large buffer size.

If the bucket is small, it will take less time for the water to stop flowing if the stream is diverted, but you'll get the water faster.

Consider if someone were to pour food coloring into the stream of water. With a larger bucket, you won't be aware of the change for some time until the colored water makes its way to the hole in the bottom. With a smaller bucket, you'll become aware of the change faster. This is what it's like for audio that flows in and out of your audio buffer.

Your computer will respond faster to changes you make to your music if you have a small buffer size and slower if you have a large buffer size."

From the chart that was linked, it is pretty much an output latency graph (not combined, output only) that charts how long it takes for the sound cards buffer to receive and fill the data stream and then how long it takes for the sound card to process this binary data into a sound wave."

As can be seen from the listing, some devices have extremely fast chips/circuitry/busses on board that allow them to receive data and then convert that data at blistering speeds, i.e. the RME device that rates at 10. Though for $2900 dollars, it would want to.

What I don't fully understand is how some cards are able to be more stable at high CPU loads then other cards. The best answer I can theorize is that lower quality cards generate longer wait states, causing your CPU to wait longer to complete the task before it can move on to its next processing thread. But this is just my theory.

If my theory is correct, it could sort of explain how differing cards at a lower latency perform differently. The smaller the buffer, the more times the card interrupts the CPU to ask it for data, this generates wait states more frequently and the longer the wait state/longer the delay to complete the op, the greater the latency and interruption to your CPU's processing which would also result in a higher (perceived) CPU utilization (it has to do smaller operations many times, vs a large operation fewer times).

[ZenPunkHippy]

What I don't fully understand is how some cards are able to be more stable at high CPU loads then other cards.

The magic is in the drivers. The reason high end cards rate so well in terms of latency and system load is quality drivers. The efficiency of the code, basically. RME drivers are rock solid. Lynx drivers also.

Like many things, you get what you pay for ...

Peace,
Andy.