Tír Na HiFi

rickmcinnis wrote:satshanti,
One wonders if your experience with mqn-ks could be due to the AMD chip?
As crazy as everything can be in audio; the fact that sbgk is using an Intel powered machine could, to my wacky way of thinking, make a difference in what you are hearing versus what he is hearing.

Yes, you may be right. But there were some Wasapi versions especially released for AMD chips, and they did sound the best in my system. I guess there are a number of factors affecting sound quality, some of which are working the same way for all chipsets, but some may have opposite effects.

uploaded 8.00 16/44 normal avx2, most minimal loop possible. expect some drivers won't like it, but works on ciunas.

sound is clean. unique sound again.

8.01 reverses the 7.89 change as 8.00 maybe a bit strident

m.massimo wrote:About normal 1644 versions on my system:

7.98 < 7.86 >> 7.68 > 7.49, all with 3.64 control

Just listened for curiosity 1644 6.17+3.61 512 compared to 1644 7.86+3.64. I prefer ks, no doubt.

Listened to Charlie Mariano - Deep in a dream (Enja)

Did you tried 1644 6.17 with 512 control, or even better 2.97?

sbgk wrote:uploaded 8.00 16/44 normal avx2, most minimal loop possible. expect some drivers won't like it, but works on ciunas.

sound is clean. unique sound again.

8.01 reverses the 7.89 change as 8.00 maybe a bit strident

You are right, does not work here, both 8.00 and 8.01 have distortion.

sbgk wrote:uploaded 8.00 16/44 normal avx2, most minimal loop possible. expect some drivers won't like it, but works on ciunas.

sound is clean. unique sound again.

8.01 reverses the 7.89 change as 8.00 maybe a bit strident

SonicWeld with Thesycon 2.23 is just working fine with both of them, running realtime priority.

8.03 good nasal vocals on tombstone blues

3daudio wrote:

sbgk wrote:uploaded 8.00 16/44 normal avx2, most minimal loop possible. expect some drivers won't like it, but works on ciunas.

sound is clean. unique sound again.

8.01 reverses the 7.89 change as 8.00 maybe a bit strident

You are right, does not work here, both 8.00 and 8.01 have distortion.

The same here for MQn 1644 8.0 and 8.1 avx2 with Audiophilleo (WDM KS drivers),distorted sound ,but Thesycon 2.20/2.23 +M2Tech are fine with 8.0 and 8.1.Hopefully the loop can be fattened a bit so that more drivers are able to cope. My first impression is that 8.1 is actually the most strident of the two with soprano voices,that could be due increased transparency, more listening required.

sima66 wrote:

m.massimo wrote:About normal 1644 versions on my system:

7.98 < 7.86 >> 7.68 > 7.49, all with 3.64 control

Just listened for curiosity 1644 6.17+3.61 512 compared to 1644 7.86+3.64. I prefer ks, no doubt.

Listened to Charlie Mariano - Deep in a dream (Enja)

Did you tried 1644 6.17 with 512 control, or even better 2.97?

Thank you for your tip.
I've just tried 6.17sse2+512 and 6.17sse2+2.97 (I suppose 2.97 is the control inside 2.97.zip)
I listened to baroque music with organ and strings.
I confirm that I prefer latest ks, 7.86normal+3.64 above all. I found ks more dynamic and trasparent, the sound is clean and detailed with no harshness.
However 6.17sse2+512 is very good and better than I remembered.
That's true on my system with specific OS settings, I suspect that every system is a different story.

Both 8.00normal+3.64 and 8.01normal+3.64 give not distorted sound, but with audible clicks.

8.01 and 8.03 are playing fine here. Haven't done any a/b testing on wasapi/ks for awhile now. These latest versions for me anyway make returning to wasapi a necessity only for 24/88 and 192.

Are the guys having clicks/etc using 2012 or W8? Noticed in signatures some are 8 and others R2 so it would seem driver is the issue.

uploaded 8.04 1644 normal/avx2, attempts to solve the issues people were having and some tweaks to the sq


8.04 a bit soft, 8.05 is more realistic

there are still 7 instructions per buffer that affect the sound, their order seems to make a difference

8.06 order is add, add, or or, mov, mov, mov

8.05 order is add, or, add, or, mov, mov, mov

the last 2 movs are dependant on the 2 adds.

why it affects the sound, I don't know. In theory it shouldn't, guess there's still instruction/memory latency and stalls.

the theory of 8.05 was that add and or use different parts of the cpu so shouldn't be waiting on the add to complete, but sometimes the cpu optimises the code and maybe prefers to see the same instructions together, 8.06 maybe sound better.


add r12, 1764
add rbx, r14
or rcx, rdi
or edx, esi
mov QWORD PTR [rsp+32], rbp
mov QWORD PTR [rbp+40], r12
mov QWORD PTR [rbp+8], rbx