Jump to content

Jim Roseberry

  • Content Count

  • Joined

  • Last visited

  • Days Won


Posts posted by Jim Roseberry

  1. Here's the same Helix Native test running on a 5950x.


    Still a substantial load... but significantly less so



    This is comparing an $800 CPU (5950x)... with an $800 computer (M1 Mac Mini with 16GB RAM).

    It's not a fair comparison by any means.

    The M1 Mac Mini is a slick, small, quiet machine.


    Note that the 5950x (using Antelope's ASIO driver) can actually go down to a 24-sample ASIO buffer size (glitch-free).

    The Mac Mini wouldn't allow buffer size smaller than 32-samples.  Though CPU load would have been thru the roof, I suspect it could have sustained the load (at 24-samples) glitch-free.


    More to come as time allows



    • Thanks 2

  2. Here's a stress-test I like to run.  Indicative of single-core performance and if the machine can sustain heavy load (glitch-free) at super small buffer size.

    • Orion Studio Synergy Core audio interface set to 96k using a 32-sample buffer size (1ms total round-trip latency)
    • Presonus Studio One Pro v5 running Helix Native with a substantial patch


    Here's the performance of the Mac M1 Mini



    There's been a lot of hype regarding the M1's single-core performance.

    I've read posts claiming single-core performance bested any current CPU.


    As you can see, this test is a substantial load for the M1.

    Though CPU load was high, audio playback was completely glitch-free.


    I'll post the same test running on a 5950x.


    • Like 1
    • Thanks 2

  3. 38 minutes ago, InstrEd said:

    Heck competition is great. Now Intel, AMD on the x86 hardware front have work to do and MS on the software end. Just think when Ryzen is on 5nm for laptop chips😊

    With the Vermeer 5xxx series, AMD finally got their low-latency performance together.

    One thing with the 5950x; the cores heat up quickly under load... causing fan-speed ramp-ups (which can be annoying).

    Otherwise, a fantastic performer...

    • Thanks 1

  4. On 6/10/2021 at 3:57 PM, batsbrew said:

    the REVV Generator 7-40 is very cool.

    I bought the Revv Generator 120 mkIII (based on liking the D20 and G20 and the G2 and G3 pedals).

    Didn't like it at all...

    Probably the most disappointed I've been in any amp purchase.

    I'm wondering now if the amp was defective.


  5. 9 hours ago, Hugh Mann said:

    This is incorrect with an m1. You really should test one yourself before you go spreading false information.  M1 can easily handle the grandure and way more.  

    I'm not spreading any misinformation whatsoever.

    You might want to look closer at the picture in my above post.

    That is from a M1 mini that has 16GB RAM.


    Bottom line is that you're not going to be doing anything serious with 8GB RAM.

    The OS and one standalone Kontakt instance with The Grandeur and you're already above 8GB of RAM used.

    • Like 1

  6. On 6/11/2021 at 4:41 PM, sarine said:

    We're left with 16 VSTi's (synths) consuming 3.9GB of RAM.

    FWIW, None of those virtual-instruments are using deep-sampled libraries.

    Single instance of The Grandeur (Kontakt standalone - no DAW application open) puts Mac Mini well over 8GB RAM.

    • Like 1

  7. M1 Mac Mini running OSX Big Sur v11.3

    This is with most extraneous things disabled.

    Running idle, it's using well over 7GB RAM.

    Open standalone Kontakt 6 and load The Grandeur (piano library)... and memory usage is 8.37GB.

    That's no DAW application... just a single large piano sample library running standalone.



    More details to follow

  8. Anybody else notice that performance at ultra low latency (96k using a 32-sample ASIO buffer size - 1ms total round-trip latency) is significantly better than before?

    You can now run Helix Native at 1ms round-trip latency (above settings)... without ASIO peak overloads.

    That wasn't the case with Cubase 10/10.5.


    I'd been putting off upgrading to Cubase 11... because ultra low latency performance wasn't particularly great.

    Didn't see anything listed about performance improvements for v11.

    I might just go for Nuendo 11.  Somewhat affordable (relatively speaking) at $240 for the crossgrade

  9. 14 minutes ago, Hugh Mann said:

    there are  lots of videos that prove that 8 gb is plenty enough on an M1 mac.  They use the very fast ssd as a sort of ram. 


    • PCIe 4.0 NVMe currently sustains ~7GB/Sec
    • DDR4 sustains ~20-25GB/Sec

    Response Time:

    • NVMe is ~0.05ms
    • DDR4 is ~0.00005ms

    NVMe is sequential (not Random Access like RAM)

    While NVMe drives are fast... they're not a good substitute for RAM.

    There is no magic/voodoo with machines.

    Slower hardware means slower performance.


    The M1 MacBook Air is passive-cooled.



    • Like 1

  10. Just now, Fleer said:

    What happens with virtual-instruments that don’t stream samples from disk (Superior Drummer 3, EZ Drummer 2, Addictive Drums, etc)?

    They're loaded into (and stream from) RAM.

    • Thanks 1

  11. 1 hour ago, sarine said:


    Unless you're talking about orchestration or other heavy sample libraries, I have no idea what you're talking about.

    Uhhh... yes.   

    First, not all virtual-instruments stream samples from disk (Superior Drummer 3, EZ Drummer 2, Addictive Drums, etc).

    Those Vi's that do stream from disk... buffer the transient of each sample using a small bit of RAM.


    I'll use Reaper as an example (MacBook Pro in this case)... as it's cross-platform and extremely small-footprint.

    Open an empty project, add a single instance of Kontakt... and load up The Grandeur (acoustic piano).

    With just Reaper and The Grandeur loaded, OSX shows 4.51GB of RAM being used.

    Now, add a second track with Super Drummer 3... and load  the "clean" Ayotte kit.

    OSX now reports 6.61GB memory being used.

    On a MacBook Pro with 8GB, you've got less than 1.5GB of free RAM.

    I wouldn't consider a two track piano/drum project to be "heavy orchestration".  😁


    If a machine runs out of physical RAM, it'll use the VM swapfile (in lieu of physical RAM).

    That kills performance.


    Professional composers working on TV and Film are running a minimum of 64GB RAM (most are now running 128GB).

    Many of these folks are clients...



    • Like 4
    • Thanks 1

  12. 10 hours ago, Bapu said:

    We Need @Jim Roseberry to chime in on 16gb system which is what I would choose along with 1tb or 2tb OS drive.

    I think this would blow away my MacBook Pro Quadcore 8Gb 1tb.

    It would blow away your older MacBook Pro.


    I'm not much of an Apple guy... and not a fan of ultra small form-factor machines for "workstation" purposes.

    A laptop that's passive-cooled with 8GB of RAM is suitable for it's design purpose (Surfing the Web, office duty, etc).

    With 8GB RAM, a small handful of virtual-instruments would have the machine RAM-starved.

    Open up Chrome with half a dozen tabs active/open.  You can chew thru RAM quickly.

    Even using compression to stretch that 8GB further... it's running lean.

    If small-form-factor also means small-cooling, that's going to limit performance.


    Some folks will try to tell you the M1 (mobile CPU) will out-perform something like the 5950x (desktop CPU).

    If expecting that level of performance, the laws of thermal-dynamics (tight enclosure) are going to disappoint you.  😁

    If you're expecting a great performing mobile CPU, you'll be pleasantly surprised.

    It will be interesting to see what Apple does in a desktop version of the M1.


    My idea of a great small-form-factor machine:

    Lian-Li TU-150 mini-ITX case - allows full-sized Noctua 140mm cooler

    Full-sized cooler means you can run a high-end desktop CPU... at full speed.

    Up to 64GB RAM

    Multiple internal SSDs (including M.2 NVMe Ultra)

    Small... but zero performance compromise

    Runs near dead-silent 

    • Like 4

  13. Had a MB Badlander 50.

    Crunch was fantastic.

    Onboard Cabclone IR sounded very good

    The internal fan was LOUD... to the point I couldn't deal with it.

    Otherwise, it was a fantastic amp... and (compared to Friedman, DR-Z, ec) reasonably priced.


    Had a DR-Z CAZ-45.

    It's billed as a high-gain amp.

    If this makes any sense, it's extremely well built... but... (to me) just didn't have any umph (insert your favorite adjective for "balls").

    I'm generally into modded-marshall (gain wise). 

    Wanted to love the CAZ-45... but just didn't care for it.

    DR-Z is located here in OH, has a great reputation, etc.

    Wound up with a BE-100 Deluxe, Bogner Shiva, and 20th Bogner Shiva.

    Of course, this was during the height of Covid.  

    What in the world was I thinking buying 50-60 pound heads?

    Nothing at all "practical" about them.  🤪

    • Thanks 1

  14. On 6/1/2021 at 7:02 AM, Gordon Prokap said:

    This interface has caught my attention, what does Cakewalk report the RTL to be on a 32 sample buffer at 48 KHz?

    At 96k with 32-sample ASIO buffer size, the round-trip latency is 1ms.

    It's going to be right about 2ms at half the sample-rate (48k).

    I've measured this with a loopback test... so it's accurate.

  15. If you've tended to all the details, you can achieve lower round-trip latency with Thunderbolt.


    Think of Thunderbolt as "external PCIe"

    Under ideal circumstances, you can achieve PCIe level performance with an external audio interface.


    The best USB audio interfaces achieve round-trip latency of ~4ms.

    The best Thunderbolt audio interfaces can achieve round-trip latency of sub 1ms.


    All USB audio interfaces I've seen/used/tested that connect via USB-C... are actually USB-2 devices.


    I've been running Thunderbolt audio interfaces on PC for a couple years.

    UA Apollo, RME UFX+, Presonus Quantum, and Antelope Orion Studio Synergy Core (can they make the name a little longer? 🤪)

    All have been rock-solid.


    If you're getting a new machine, be aware that there are currently issues with Thunderbolt-4 controllers.

    Ironically, TB4 is not working well with Intel based machines... but works just fine on the Asus ProArt B550 Creator (AMD 5950x Vermeer CPU).

    Ultimately... this will be ironed out.


    • Like 1
    • Thanks 1

  16. On 3/4/2021 at 11:15 PM, Bruno de Souza Lino said:

    Only thing Intel CPUs are still king is single core performance, which is becoming less and less important nowadays

    Single core performance is still extremely important... especially if you're pushing the limits of ultra low latency performance.

    ie: Let's say you have a Thunderbolt audio interface that yields 1ms total round-trip latency at 96k using a 32-sample ASIO buffer size.

    With those settings, you can play/monitor thru plugins like Helix Native with 1ms round-trip latency (equal to or better than the Hardware).

    At those settings (96k/32-sample ASIO buffer), the machine has ~half a millisecond to process the next ASIO buffer and get it in cue for playback.

    This is not something that lends itself to being heavily multi-threaded (spread across multiple cores).

    If anything interrupts this process, you'll hear glitches.


    If you try doing to above with the Threadripper 3970x (~$2000), you'll hear glitches.

    The 10900k (~$500) can do the above with zero glitches.


    The new 11900k shows performance improvement in many benchmarks (vs the 10900k), but there were some changes in architecture that make it more latent.

    If you're pushing the limits at ultra low latency (32-sample ASIO buffer size or smaller), the 11900k is a step backward.


    The newer Ryzen Vermeer series (5xxx series) has greatly improved low latency performance.

    The 5950x (~$800) can run the above example glitch-free.

    Unlike Threadripper (280w TDP), the 5950x is a much more manageable (quiet) 105w TDP.



  • Create New...