AMD Ryzen - Our first look and thoughts for audio.

[Steve Bolivar]

Here's a link to a thread over at the Cakewalk forum on a Ryzen 1800x build.
http://forum.cakewalk.com/m/tm.aspx?m=3569908&p=1
After some initial problems with the mobo's on-board audio driver causing problems he's reporting some promising results with load sharing between the cores.

[camsr]

Compilers don't know about that CPU yet, neither do operating systems, driver & co

Actually, GCC already has a tuning profile for amd 17h (zen core)

[Spyro]

AMD 1700X 3.8 @ GHz

64 = 520 count @ 70% load
128 = 860 count @ 72% load
192 = 1290 count @ 85% load

Intel 6800k 3.8 @ GHz

64 = 540 count @ 85% load
128 = 930 count @ 99% load
192 = 1590 count @ 99% load

Intel 6900k 3.6 @ GHz

64 = 990 count @ 85% load
128 = 1550 count @ 90 % load
192 = exceeds test

Intel 7700k @ 4.5GHz

64 = 560 @ 90% load
128 = 950 @ 98% load
196 = 1270 @ 99% load

This is VERY good for me considering I use 256 or 512 most of the time. Even playing guitar at 256 I cannot feel the latency.

I see the 1700x, 6800k and the 7700k all can achieve (as of now) practically the same voice count at 64 samples. Note that 64 samples is impractical, it puts too much stress on the CPU right away, everything can cause dropouts.

At 128 the gap is greater on the Intel's but not a deal breaker. At 192 samples the 1700x and the 7700k are virtually the same. I've noted the 7700k was running at 196 instead of 192, yes is a minor difference but also reflected in the number of voices so I assume at 192 the 1700x and the 7700k gap will be even greater in favor of Ryzen. And that's from a 1700x running just at 3.8ghz, quite impressive.

Also, the Ryzen seems to have more headroom, either for future optimizations or for other tasks. Basically Ryzen achieved a very competitive voice count at lower CPU usage. And we're talking about rather ridiculous latencies (as I said 256 samples is already an optimal latency)

The benchmarks show some very impressive numbers from the Ryzen 1700x

I'm still waiting for my (now delayed) parts for a 1800x build, but I'm now more and more confident is gonna be a good upgrade from a 3770k at 4.6GHz. And it would not hurt at all to have all 8 cores and 16 threads for other tasks as rendering and transcoding big files.

Also in the high count example in the video the audio dropped slightly, not as bad as I've experienced. And that was running just at 3.7Ghz. :O

[4damind]

Is the difference from the 1700x to the 1800x not as big or would the AMD 1800x outperform the Intel 6900k?

[egbert]

I really appreciate the contributions here - particularly from the pro DAW builders. With new platforms like this we all need all the help we can get.

[Kaine]

We had 3000MHz in there but on the pre-launch hardware it wouldn't run above the CPU rated 2133MHz, we expect it to change with newer builds.

Don't mix the data rate with the frequency.
You use DDR RAM, where the DDR stands for double data rate.
So your DDR4-2133 has a data rate of 2133MT/s (mega transfers per second) at clock frequency of 1066Mhz (half of the rate).

I wasn't mixing it up, I was just making use of what I had to hand at the time and that may appear confusing.

It was 3000MHz RAM I had on the desk for testing. I had some 2600 and 2400 stuff as well, all of it got tried in the board, all of it defaulted back to 2133MHz so I just kept using the initial RAM.

I wasn't making reference to the CPU/RAM controller data rate, if I gave that impression, I can only apologize.

Is the difference from the 1700x to the 1800x not as big or would the AMD 1800x outperform the Intel 6900k?

200MHz is normally neither here or there in performance terms. For the £$80 on the price your getting very little extra. We did run the 1700X at the 1800X stock speeds and it was marginal gains.

However the point there is that both chips will have been sorted from the same silicon. So, in that case the 1800X should be cherry picked as the best in batch and I would hope that they would overclock far better than the 1700X.

[PurpleSunray]

I wasn't making reference to the CPU/RAM clock frequency, if I gave that impression, I can only apologize.

Ok, the I got it wrong =)

me wrote:
The CCX connects to the memory controller - which does run on RAM frequency (intel runs it near core frequency).
DJ Warmonger :
That's interesting, could the RAM frequency affect final performance? What was OP test run at?
You wrote:
We had 3000MHz in there but on the pre-launch hardware it wouldn't run above the CPU rated 2133MHz, we expect it to change with newer builds.

so was thinking that you are reffering to IMC frequency as well (which would be half of ram datarate on DDR) .. but you did answer second question i assume

[kmonkey]

Also lets not forget (when looking at tests where i7 6900k win over ryzen 1800x)

I7 6900k is 10 core cpu
Ryzen 1800x is 8 core cpu

Looking from that - ryzen is pretty darn impressive imo

Edit: my bad i messed up numbers i76900k is 8 core

[egbert101]

<delete>