The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
Intel vs AMD - DawBench (Pro-Q/Pro-MB/Kramer) on Reaper & Ableton Condenser Microphones
Old 8th June 2017
  #1
Gear Maniac
 

Thread Starter
Intel vs AMD - DawBench (Pro-Q/Pro-MB/Kramer) on Reaper & Ableton

Hi guys!

I decided to bench test the new 1800x Ryzen CPU against Intel's 6900K offering for audio related tasks. I took the benchmark files you can find on DawBench.com and customized them to test more popular plugins:

Fabfilter Pro-MB, Fabfilter Pro-Q, and Kramer Tape.

DawBench also lacked Ableton test files so I replicated their testing environment and have included my modified benchmark files in the video description.



Would love to hear what you guys think!
Old 8th June 2017
  #2
Lives for gear
 

Thanks for doing of that test!

Michal
Old 8th June 2017
  #3
Lives for gear
 
TAFKAT's Avatar
 

Very interesting to see some DAWbench testing outside of the established box so to speak, I never got around to porting to Ableton , so great to see you have managed to port the session across.

The sessions can be used as templates for any plugin yoiu prefer to test with. I am confused how you manged to get 5 thousand instances activated in the session tho ?

Might have been an idea to test at other latencies lower than 1024 to get a better overview.

Out of interest, what interface were you using, but at 1024 it wouldn't really have much if any sway to be honest.

Old 8th June 2017
  #4
Gear Maniac
 

Thread Starter
Quote:
Originally Posted by TAFKAT View Post
Very interesting to see some DAWbench testing outside of the established box so to speak, I never got around to porting to Ableton , so great to see you have managed to port the session across.

The sessions can used as templates for any plugin yoiu prefer to test with, I am confused how you manged to get 5 thousand instances activated in the session tho ?

Might have been an idea to test at other latencies lower than 1024 to get a better overview.

Sure, I had to expand the testing environment to way more than 40 tracks to achieve this. Instead of 40 tracks, I made the environment 70 tracks, each loaded with 8, and then called that grouping a "block", and then duplicated that grouping a bunch.

It was a very long test, took me nearly an hour to complete! In the end I had 10 groups of 70 tracks in each all fully loaded before capping out in Reaper testing with Pro-Q.

Also, an interesting aside which I will make another video on. fabfilter pro-q was actually more resource efficient than Ableton's stock EQ plugin. I will test it against other stock eq plugins as well (such as Reaper's stock EQ plugin).

EDIT: Just saw you edited your post to include a question about interface. This was testing with my Lynx Hilo Mastering DAC
Old 8th June 2017
  #5
Lives for gear
 
TAFKAT's Avatar
 

Quote:
Originally Posted by PrismaPhonic View Post
Sure, I had to expand the testing environment to way more than 40 tracks to achieve this. Instead of 40 tracks, I made the environment 70 tracks, each loaded with 8, and then called that grouping a "block", and then duplicated that grouping a bunch.

It was a very long test, took me nearly an hour to complete! In the end I had 10 groups of 70 tracks in each all fully loaded before capping out in Reaper testing with Pro-Q.
O.K, interesting.

I was about to ask how long it took to activate 5K plugins one at a time.

As a rule I selected heavier resource intensive plugins to keep the total number within the 320 max ( 8 inserts per 40 x sine track). That keeps the testing time down , especially when testing across a range of latency settings.

The current crop of Ryzen/i7's are definitely a challenge due to the huge resources available.


Quote:
EDIT: Just saw you edited your post to include a question about interface. This was testing with my Lynx Hilo Mastering DAC
Cool, if you do get some time , it would be good to see some results at lower latencies , maybe just stick to the more resource intensive plugins to keep testing time down.

Aside from the DAW comparative performance which is a whole other area, the Ableton results are raising a few questions as to why there are even larger performance variables with the Ryzen against the i7. It could come down to compilers , which is opening another can or worms.


Old 9th June 2017
  #6
Gear Maniac
 

Thread Starter
Quote:
Originally Posted by TAFKAT View Post
Cool, if you do get some time , it would be good to see some results at lower latencies , maybe just stick to the more resource intensive plugins to keep testing time down.
Yeah, I can run those tests at lower latencies, at least for Kramer and Pro-MB. Won't bother making it into a video but I'll post my results here :-)
Old 9th June 2017
  #7
Lives for gear
 
Finnish's Avatar
 

Potential Ryzen user (in the future) so I'd like to see some results @64samples in Reaper
Old 9th June 2017
  #8
Gear Head
 

Nice one
Old 9th June 2017
  #9
Lives for gear
 

With upcoming Threadripper we will not compare the audio performance of an 8 core AMD vs an Intel 8 core. But the best performance you get from a system for your money .

5K for an iMac 8 core to run a FL project on a 16 core Ryzen doesn´t sound like a smart investment to me at the moment.

Nevertheless interesting months to follow the developments.
Old 9th June 2017
  #10
Lives for gear
 

Quote:
Originally Posted by lllubi View Post
With upcoming Threadripper we will not compare the audio performance of an 8 core AMD vs an Intel 8 core. But the best performance you get from a system for your money.
This is not really related to some upcoming products, you can already do that.. and I believe, for the majority of people here (eg. non-professionals) the price/value ratio is major attribute for their decision. So pricewise R7 1800X is comparable rather to 6800K or 6850K.

Anyway there is always something behind the corner.. for someone it is Threadripper, for others it can be i9. It will never stops.. whenever you look for hardware, you'll always find some recommendation to wait for some forthcoming product.

Quote:
5K for an iMac 8 core to run a FL project on a 16 core Ryzen doesn't sound like a smart investment to me at the moment.
I don't really get that.. Well if someone relies on Fruity Loops, he likely works at PC.. If someone has complete Mac based environment, then he will likely consider lot of another factors than just sheer CPU power.
Other thing is that core count mantra 8>6, 6>4, 18 kill-em-all.. doesn't really applies to all situations.

Michal
Old 9th June 2017
  #11
Gear Maniac
I take it these Kramers count as stereo instances? According to the stats on Waves' Soundgrid Servers, the Kramer in mono gets 128 at 44k and 44 at 96k. They told me you would halve those values for stereo instances.

This is another test from this forum:

https://www.youtube.com/watch?v=FQ5DGWXShhI

In Pro Tools he's getting 88 Kramers in stereo in a 48k session. Does this mean on a Ryzen 1800x in Pro Tools you'd only be getting 40 some Kramers compared to 184 in Reaper? My assumption is that the 1800x wouldn't enable one to run a ton more Kramers in Pro Tools when compared to the 1700. Seems amazing to me that Reaper could have such a drastic performance improvement over PT. That's a difference of almost 140 instances. I guess the sessions could be setup differently and that could contribute. Definitely interested in hearing everyone's thoughts on these.
Old 9th June 2017
  #12
Lives for gear
 

Quote:
Originally Posted by PrismaPhonic View Post
Hi guys!

I decided to bench test the new 1800x Ryzen CPU against Intel's 6900K offering for audio related tasks. I took the benchmark files you can find on DawBench.com and customized them to test more popular plugins:

Fabfilter Pro-MB, Fabfilter Pro-Q, and Kramer Tape.

DawBench also lacked Ableton test files so I replicated their testing environment and have included my modified benchmark files in the video description.

Would love to hear what you guys think!
I don't think the wording in your conclusion is appropriate. When you say that one thing (X) has an advantage over another (Y) and then quantify that advantage, I'm guessing most people will perceive that as

X-Y = advantage as an absolute number, and
(X-Y)/Y = advantage as a percentage of change.

Imagine that we're discussing you and I competing in a marathon, and you complain that I had an unfair advantage. You could for example say "Well, Mattias got to start 15 minutes before me". So what you're doing is describing the delta, the advantage itself. In other words, reading only your conclusion you make it seem that going from an 1800x to a 6900k I can expect an advantage, i.e. a positive change or increase, of 152.79% over the 1800x. If that was true then on average if the 1800x could run 100 plugins the advantage itself would 152 more plugins, for a total of 252 plugins.

Looking at your data though it seems to be completely in line with what one would expect with a competitive chip. The difference itself isn't 152%, it's 52%, and that roughly mirrors its price.

Also - a possibly dumb question: Are you entirely sure that processing is actually taking place when those channels are muted/have volume turned down? I'm just curious because I know that some DAWs won't have instances take up resources fully if there's no audio flowing through them, and the difference between the DAWs was so tremendous. Could you maybe test it by simply turning down the master and turning up the volume on those channels?
Old 9th June 2017
  #13
Lives for gear
 
TAFKAT's Avatar
 

Quote:
Originally Posted by PrismaPhonic View Post
Yeah, I can run those tests at lower latencies, at least for Kramer and . Won't bother making it into a video but I'll post my results here :-)
Cool, yeh just stick to the higher resource plugs , that will give a better snapshot and stick within the standard DAWbench testing environment.

I wouldn't bother with Ableton , seems to be some other variables at play there

Old 10th June 2017
  #14
Lives for gear
 
fireberd's Avatar
Old 10th June 2017
  #15
Lives for gear
 

Quote:
Originally Posted by fireberd View Post
Jim Roseberry, who builds DAW systems (his day job), did some comparison and posted on the Cakewalk Sonar Forum

i7-6850k vs. i7-7700k vs. Ryzen 1800x | Cakewalk Forums
Curiously, he's saying "with socket 2011-3... you won't see appreciable performance gain using DDR4 above 2400MHz.
By comparison, Ryzen sees a significant performance boost (especially at small ASIO buffer sizes)."

in contrast to that fella that posts here...
Old 10th June 2017
  #16
Lives for gear
 
fireberd's Avatar
I haven't done any testing myself, just thought I'd throw in that post since Jim (studiocatPurrrfect Audio) is into PC DAW building and seems to be a straight guy (he helps quite a bit with technical stuff on that forum).
Old 10th June 2017
  #17
Lives for gear
 

Quote:
Originally Posted by mattiasnyc View Post
Curiously, he's saying "with socket 2011-3... you won't see appreciable performance gain using DDR4 above 2400MHz.
By comparison, Ryzen sees a significant performance boost (especially at small ASIO buffer sizes)."
Have you tried direct comparison of different memory clock speeds at Ryzen? I'd be curious about some close to real-world figures. Say with DAWBench, because its opposite to what Peter Kaine and one other user mentioned in the big Ryzen thread - eg. it's the same as with Intel and memory speed doesn't really affect performance with DAW workloads.

Michal
Old 10th June 2017
  #18
Lives for gear
 

I don't have a Ryzen chip. If I had one I would have run the benchmark suite and shared the results.

It's interesting though because if the lower latency performance on Ryzen is not as good as people want because of the cross-ccx communication then increasing the memory speed would increase performance, since the memory speed and infinity fabric speed over which the ccx' communicate are tied together. Higher memory speed = lower latency when the ccx complexes share data.

So, I'm wondering how one tester can see an improvement and the other not. It just doesn't quite make sense to me.
Old 10th June 2017
  #19
Lives for gear
 

Quote:
Originally Posted by mattiasnyc View Post
I don't have a Ryzen chip. If I had one I would have run the benchmark suite and shared the results.

It's interesting though because if the lower latency performance on Ryzen is not as good as people want because of the cross-ccx communication then increasing the memory speed would increase performance, since the memory speed and infinity fabric speed over which the ccx' communicate are tied together. Higher memory speed = lower latency when the ccx complexes share data.
...
Hmm.. so in this case, I probably wouldn't talk about increased memory clock like important attribute for DAW performance, until someone really proves that.

I directly asked Pete about that, because he did Ryzen tried it with different clock speeds and Dawbench.
And yes I agree with you, Ryzen's architecture with separate 4 core CCXs brings some challenges (as tradeoff for cheaper price and mfg. yield rate with smaller die) with regards to scheduling and sharing of cache data via Infinity Fabric, which is clocked together with memory. So while higher clock might be important for some workload with more intensive data sharing between CCXs, it definitely won't apply to everything. I can imagine, heavily threaded DAW workload with many individual track workers doesn't necessarily benefit so much from smaller latency at infinity fabric.
Which might be bit different than some games or compression algorithms with more intensive cache sharing architecture, where that increased memory clock speed might give you couple of percent.
Although different architecture.. with Intel setups, speed of RAM doesn't really matter and when I tested it with various DAW workloads, difference was at level of statistical error.

Michal
Old 10th June 2017
  #20
Lives for gear
 

Quote:
Originally Posted by fireberd View Post
Jim Roseberry, who builds DAW systems (his day job), did some comparison and posted on the Cakewalk Sonar Forum

i7-6850k vs. i7-7700k vs. Ryzen 1800x | Cakewalk Forums
He redid the tests with faster memory and - importantly apparently - using 64 bit plugins instead of 32 bit. It doesn't look like AMD is a fan of running 32 bit software for some reason.

Amd ryzen is here

@PrismaPhonic: just to check, are you all 64 bit OS/DAW/plugins?
Old 10th June 2017
  #21
Lives for gear
 

Quote:
Originally Posted by msmucr View Post
Hmm.. so in this case, I probably wouldn't talk about increased memory clock like important attribute for DAW performance, until someone really proves that.

I directly asked Pete about that, because he did Ryzen tried it with different clock speeds and Dawbench.
Yes, he did. And I said "if" for a reason. Someone else linked to another person doing tests saying that increasing memory speed makes a big difference. So they can't both be right.

Quote:
Originally Posted by msmucr View Post
And yes I agree with you, Ryzen's architecture with separate 4 core CCXs brings some challenges (as tradeoff for cheaper price and mfg. yield rate with smaller die) with regards to scheduling and sharing of cache data via Infinity Fabric, which is clocked together with memory. So while higher clock might be important for some workload with more intensive data sharing between CCXs, it definitely won't apply to everything. I can imagine, heavily threaded DAW workload with many individual track workers doesn't necessarily benefit so much from smaller latency at infinity fabric.
Which might be bit different than some games or compression algorithms with more intensive cache sharing architecture, where that increased memory clock speed might give you couple of percent.
Although different architecture.. with Intel setups, speed of RAM doesn't really matter and when I tested it with various DAW workloads, difference was at level of statistical error.

Michal
Intel not benefiting from faster RAM is completely irrelevant if the issue is the cross-ccx data sharing. Again, if.....
Old 10th June 2017
  #22
Lives for gear
 

Quote:
Originally Posted by mattiasnyc View Post
Yes, he did. And I said "if" for a reason. Someone else linked to another person doing tests saying that increasing memory speed makes a big difference. So they can't both be right.
If you're talking about the previous linked post by J. Roseberry, then at the top he just assumes, it will be faster due to CCXs communication.. there are no comparable DAWBench results say at 2133 vs 2933, because he switched to 64bit DAWBench from his previous test at lower RAM clock.

So far, the most comprehensive memory tweak review for Ryzen 7 is there
AMD Ryzen 7 Memory And Tweaking Analysis review - An introduction at a double data-rate
Where you can see results for multiple games and applications. Results vary quite a bit depending on workload.. If our considered DAW load is closer to say Cinebench in MT mode, which is already very parallelized for start, then it's easily bring no improvement.

One remark, that their top clocked 3200 memory sticks has also lower CL (14 vs 16). Running such memories at this high clock is generally know to be very problematic with regards to use with all 4 RAM banks populated according to multiple other report also by my friends.. so it's questionable, if it will be even possible to run for example 4x 8GB setup like that.

Quote:
Intel not benefiting from faster RAM is completely irrelevant if the issue is the cross-ccx data sharing. Again, if.....
I'm aware of that and mentioned that..
Frankly, I'm glad, so far it looks like, it's the same with AMD for DAW workloads, because majority of people buy or consider Ryzen setup either for power/price ratio (or sentiment ).. and such fast RAM sticks, just adds to total price of their builds possible and increase possible compatibility problems at fully populated setups. So IF someone finally proves, it's irrelevant for this massively threaded workflow.. it's good news IMO.

Michal
Old 10th June 2017
  #23
Lives for gear
 

Quote:
Originally Posted by msmucr View Post
If you're talking about the previous linked post by J. Roseberry, then at the top he just assumes, it will be faster due to CCXs communication.. there are no comparable DAWBench results say at 2133 vs 2933, because he switched to 64bit DAWBench from his previous test at lower RAM clock.
Yes, I saw that he did that, and it's interesting that we haven't seen any actual DAW data from testing at different memory speeds (as far as I can recall).

Quote:
Originally Posted by msmucr View Post
One remark, that their top clocked 3200 memory sticks has also lower CL (14 vs 16). Running such memories at this high clock is generally know to be very problematic with regards to use with all 4 RAM banks populated according to multiple other report also by my friends.. so it's questionable, if it will be even possible to run for example 4x 8GB setup like that.
Well, some users seem to hit low CL and high speeds. Not sure why others have such problems finding the right combinations.

Quote:
Originally Posted by msmucr View Post
Frankly, I'm glad, so far it looks like, it's the same with AMD for DAW workloads, because majority of people buy or consider Ryzen setup either for power/price ratio (or sentiment ).. and such fast RAM sticks, just adds to total price of their builds possible and increase possible compatibility problems at fully populated setups. So IF someone finally proves, it's irrelevant for this massively threaded workflow.. it's good news IMO.

Michal
I have to disagree. The Ryzen works very very well for what it is, and if it would work better with faster/lowerCL RAM then that would be better. One would then have the option of speeding up one's system.
Old 10th June 2017
  #24
Lives for gear
 

Awesome video! A for effort and info!
Old 12th June 2017
  #25
Gear Maniac
 

Thread Starter
Quote:
Originally Posted by msmucr View Post
Have you tried direct comparison of different memory clock speeds at Ryzen? I'd be curious about some close to real-world figures. Say with DAWBench, because its opposite to what Peter Kaine and one other user mentioned in the big Ryzen thread - eg. it's the same as with Intel and memory speed doesn't really affect performance with DAW workloads.

Michal
I would be happy to run my tests at various clock speeds to see if that has an impact because I did have a lot of people calling "foul" on my video due to not matching the 3200 Mhz overclock I was able to achieve on the Intel platform.

Let's not forget though that I did clock the memory as high as I could possibly get it on AMD, but that platform is currently experiencing well known limitations in memory speeds. I did see that a new BIOS update just came out for the x370 board I'm using though so I might be able to clock higher now on the Ryzen machine and can run some tests there.

Personally I don't think it would have an impact on max plugin load handling. There have been plenty of reports that on Ryzen going from 2400 Mhz up to an extreme overclock of 4000 Mhz (how they got it that high is beyond me) showed a 9% performance difference in certain benchmarking. As far as I know this has not been tested in audio workloads but if we assume it's true, then it wouldn't even get close to explaining the 150% performance advantage I saw on the 6900K over the 1800x.

This all being said, something I could have stressed more in my video was the price/performance of these chips. Even with a 150% performance lead by Intel, we are talking about a $1050 chip vs a $450 chip, so the AMD still pulls ahead in price/performance. So why did I even bench test these against each other? Because AMD made the explicit claim that the 6900K is their direct competition with the 1800x. And as we saw towards the end of my video, in Cinebench and Blender rendering tests, they really are nearly identical in performance with Intel taking a very small lead. The area we see larger discrepancy is in audio workloads, which could be related to audio plugin developers optimizing for Intel and not AMD since Intel has been the market lead for nearly a decade now. My video is not attempting to state outright that AMD is a poor performer, but that whatever the causative factors are, whether they be hardware related (poor chip performance) or software related (poor optimizations), we do see a large performance discrepancy between these two platforms right now for audio related workloads.
Old 12th June 2017
  #26
Gear Maniac
 

Thread Starter
Quote:
Originally Posted by captain caveman View Post
He redid the tests with faster memory and - importantly apparently - using 64 bit plugins instead of 32 bit. It doesn't look like AMD is a fan of running 32 bit software for some reason.

Amd ryzen is here

@PrismaPhonic: just to check, are you all 64 bit OS/DAW/plugins?
Correct, all of my tests are running 64 bit DAW with 64 bit plugins. I don't even install 32 bit plugins on my machines :-)

It would be silly due to RAM limitations with 32 bit :-P
Old 12th June 2017
  #27
Lives for gear
 

Quote:
Originally Posted by PrismaPhonic View Post
There have been plenty of reports that on Ryzen going from 2400 Mhz up to an extreme overclock of 4000 Mhz (how they got it that high is beyond me) showed a 9% performance improvement in certain benchmarking. As far as I know this has not been tested in audio workloads but if we assume it's true, then it wouldn't even get close to explaining the 150% performance improvement I saw on the 6900K over the 1800x.
Did you miss my post earlier? It's not a 150% improvement.

EDIT: Again for those who missed it;

Frank: "I ran 22 miles today."
Joe: "How long was your last run?"
Frank: "The last one was 18 miles."

Q: How much did Frank improve?
A: He improved 4 miles.
Q: How many percent is that improvement?
A: 22%
Q: How do we know?
A: Because 18 x 0.22 = 4 (3. 96 to be exact).

In contrast: How much is 122% of 18? It is 21.96. And was Frank's improvement 22 miles?

Last edited by mattiasnyc; 12th June 2017 at 07:10 PM..
Old 12th June 2017
  #28
Gear Maniac
 

Thread Starter
Quote:
Originally Posted by mattiasnyc View Post
Did you miss my post earlier? It's not a 150% improvement.

EDIT: Again for those who missed it;

Frank: "I ran 22 miles today."
Joe: "How long was your last run?"
Frank: "The last one was 18 miles."

Q: How much did Frank improve?
A: He improved 4 miles.
Q: How many percent is that improvement?
A: 22%
Q: How do we know?
A: Because 18 x 0.22 = 4 (3. 96 to be exact).

In contrast: How much is 122% of 18? It is 21.96. And was Frank's improvement 22 miles?
I really think you are trying to argue a very small point here. For the record I'm not going to pointlessly engage with you on this matter any more when everyone can see all of my data and clearly understand what I meant. You are simply trying to distract from what really needs to be talked about.
Old 12th June 2017
  #29
Lives for gear
 

Quote:
Originally Posted by PrismaPhonic View Post
Correct, all of my tests are running 64 bit DAW with 64 bit plugins. I don't even install 32 bit plugins on my machines :-)
Cool thanks for clarifying.
Quote:
It would be silly due to RAM limitations with 32 bit :-P
Running 32 bit plugins in a 64 bit OS/host that supports bridging allows a near identical amount of memory to be accessed (give or take a few MB here and there for the bridging). But yeah, I agree that 32 bit must die.
Old 12th June 2017
  #30
Lives for gear
 

Quote:
Originally Posted by PrismaPhonic View Post
I really think you are trying to argue a very small point here.
It's not a "very small point". Just do the math for once.

If a user gets two pieces of information, a) that an 1800x can manage 63 plugins in Ableton, and b) that a 9600k improves on that by 150%, then that user, if he needs at least 150 plugins, will pick the 9600k expecting to run at least 63 + (63 x 1.5) plugins, i.e. 157. Imagine his confusion when he can only run 90.

That's a "small point" to you?

Quote:
Originally Posted by PrismaPhonic View Post
For the record I'm not going to pointlessly engage with you on this matter any more
You said "Would love to hear what you guys think! " I suppose like most people what you wanted was a pat on the back. All you had to do was say "Oh you know what, you're right. I did phrase that a bit incorrectly. I'll change it."

Grow some thicker skin or don't put out inaccurate information.

Quote:
Originally Posted by PrismaPhonic View Post
when everyone can see all of my data and clearly understand what I meant.
First of all, far from everyone will watch all your data. A lot of people will simply go straight to your conclusion. Secondly, people would have to sit down and do the math to see that your conclusion was wrong in order to clearly understand what you meant. I'm not so sure they'll go through that seeing that you can't even be bothered to acknowledge or correct it yourself.

Quote:
Originally Posted by PrismaPhonic View Post
You are simply trying to distract from what really needs to be talked about.
Which is what? What point do we really need to talk about that hasn't been talked about?
New Reply Submit Thread to Facebook Facebook  Submit Thread to Twitter Twitter  Submit Thread to LinkedIn LinkedIn  Submit Thread to Google+ Google+  Submit Thread to Reddit Reddit 
 
Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump