VST2/3 plugins and CPU extensions usage
-
- KVRist
- Topic Starter
- 63 posts since 14 Nov, 2014
Hello everyone,
I've got a rather technical question. Which cpu extension would a VST plugin for example Diva make use of? I know a lot of audio/video applications make use of the AVX or AVX2 cpu extensions.
The core of my questions is how would an Intel CPU with one 256bit AVX2 unit compare to the new AMD Ryzen CPUs using 2 units of 128bit. I have read that, even the AMD has two units of 128bit, it can only calculate one command per tick. So if the plugin makes use of 256bit AVX2 the new AMD cpus should only be at half the speed. Am I right?
Greetings,
smoothny
I've got a rather technical question. Which cpu extension would a VST plugin for example Diva make use of? I know a lot of audio/video applications make use of the AVX or AVX2 cpu extensions.
The core of my questions is how would an Intel CPU with one 256bit AVX2 unit compare to the new AMD Ryzen CPUs using 2 units of 128bit. I have read that, even the AMD has two units of 128bit, it can only calculate one command per tick. So if the plugin makes use of 256bit AVX2 the new AMD cpus should only be at half the speed. Am I right?
Greetings,
smoothny
- KVRAF
- 23101 posts since 7 Jan, 2009 from Croatia
I think Diva is just using SSE2 and AVX. Not sure about AVX2.
-
- KVRAF
- 35410 posts since 11 Apr, 2010 from Germany
As far as i read, those are also the most used CPU extensions in soft synth plugins. Although only few developers really state what is being used.EvilDragon wrote:I think Diva is just using SSE2 and AVX.
-
- KVRist
- Topic Starter
- 63 posts since 14 Nov, 2014
Ok, maybe I was wrong about AVX2, maybe its the same for AVX.
This is what an article says about the Ryzen CPUs:
"The floating point unit is capable of performing two FMAC operations or a single 256-bit AVX operation per cycle."
I'm just wondering if the Ryzen cpus will outperform Intel CPUs not only with Cinebench R15 but with audio plugin performance too. Because 600$ for an 8 core cpu just sounds too good.
This is what an article says about the Ryzen CPUs:
"The floating point unit is capable of performing two FMAC operations or a single 256-bit AVX operation per cycle."
I'm just wondering if the Ryzen cpus will outperform Intel CPUs not only with Cinebench R15 but with audio plugin performance too. Because 600$ for an 8 core cpu just sounds too good.
- KVRAF
- 2275 posts since 4 Dec, 2011 from Brasília, Brazil
the only one that I remember talking about these extensions is FXpansion Strobe2: AVX/AVX2chk071 wrote:As far as i read, those are also the most used CPU extensions in soft synth plugins. Although only few developers really state what is being used.
My soundcloud: https://soundcloud.com/waltercruz
- KVRAF
- 3053 posts since 25 Apr, 2011
Since nobody (practically) has such a system, it is quite difficult to rely on figures and specs only. Practical testing will show what they will bring, in terms of performance per dollar.smoothny wrote:Ok, maybe I was wrong about AVX2, maybe its the same for AVX.
This is what an article says about the Ryzen CPUs:
"The floating point unit is capable of performing two FMAC operations or a single 256-bit AVX operation per cycle."
I'm just wondering if the Ryzen cpus will outperform Intel CPUs not only with Cinebench R15 but with audio plugin performance too. Because 600$ for an 8 core cpu just sounds too good.
- KVRAF
- 23101 posts since 7 Jan, 2009 from Croatia
I would be wary of first party benchmarks, though... let's wait for Anandtech's reviews.
- KVRAF
- 3053 posts since 25 Apr, 2011
+1EvilDragon wrote:I would be wary of first party benchmarks, though... let's wait for Anandtech's reviews.
- u-he
- 28062 posts since 8 Aug, 2002 from Berlin
I've run a few performance tests, and enabling AVX hasn't made much of a difference.
We use a lot of hand written SSE/SSE2 code which still has the best bang-for-bucks ratio (outcome for time and effort spent).
I'm toying with the idea of doing AVX for selected algorithms, specifically stereo versions of our analogue filter models. However, we haven't measured the benefit yet.
We use a lot of hand written SSE/SSE2 code which still has the best bang-for-bucks ratio (outcome for time and effort spent).
I'm toying with the idea of doing AVX for selected algorithms, specifically stereo versions of our analogue filter models. However, we haven't measured the benefit yet.
-
- KVRist
- Topic Starter
- 63 posts since 14 Nov, 2014
Hi Urs,
thanks for the information. So for the moment there is no advantage for Intel over AMD, but if the AVX unit for Ryzen really only does half the speed there would be an advantage for Intel in the future.
Greetings,
Sascha
thanks for the information. So for the moment there is no advantage for Intel over AMD, but if the AVX unit for Ryzen really only does half the speed there would be an advantage for Intel in the future.
Greetings,
Sascha
-
- KVRer
- 3 posts since 10 Apr, 2017
Anyone else got some more info on this?
Do we know what products are utilizing these newer instruction sets?
Is it probable that more and more DAWs and VSTs will support AVX/AVX2/AVX512 as time goes by, or is it perhaps not enough to gain by implementing them?
Thanks!
Do we know what products are utilizing these newer instruction sets?
Is it probable that more and more DAWs and VSTs will support AVX/AVX2/AVX512 as time goes by, or is it perhaps not enough to gain by implementing them?
Thanks!