The No.1 Website for Pro Audio
 All  This Thread  Reviews  Gear Database  Gear for sale     Latest  Trending
Ryzen 3000 series
Old 4 weeks ago
  #241
Gear Nut
 
Elektronix's Avatar
 

They will come with the latest bios, I doubt it will be long before they're available as the Tomahawk Max is already available to preorder on CCL for £101.32
Old 4 weeks ago
  #242
Here for the gear
 

Quote:
Originally Posted by Pete Kaine View Post
I've been retesting today with some 3600MHz kits being run at 3725 largely because I don't have any suitable 3800MHz packs in stock and because I can't quite hit the 3733 points they suggested with the slower memory.

Even so, I've seen noticeable improvements with the faster memory this time in some of the testing, in some scenarios although now that I've also got the newer build of Vins test it's confusing things even further in the results.
Do you happen to know the exact CL value for those 3600MHz kits? If they're run at CL16 or higher, that still gives over 8,8ns in memory latency, which is apparently above the sweet spot and not low enough.
It would be interesting to see if the hypothesis is true that lower latency memory will in turn make the CPU perform better. This could become much more important for audio work (on the AMD platform at least) than previously thought, assuming it is the case.

Quote:
Originally Posted by Pete Kaine View Post
In fact, the more testing that I'm doing here, the more questions are coming up at each stage. I'm probably going to end up having to rewrite a chunk of that once I'm finished too.
I think I speak for everyone here in saying that we're eager to see the new results and further benchmarks. You're definitely doing the community a service by meticulously testing and sharing the results, it is much appreciated.
Old 4 weeks ago
  #243
Here for the gear
 

external expansion case for pci cards (to pcie)

I've been lurking in this thread for a while, thanks to everybody contributing insights and testing things!

It is somewhat off-topic, but since many may be looking at this because they'll build a new PC, and like me they may be looking for ways to reuse their PCI audio interface...I found this:

external PCI expansion box that connects to PCIe

I know somebody in this thread had luck using a pcie to pci adapter inside their case, but it seems like it could be pretty tricky to fit it all in there, at least without case modifications. This is pretty expensive, but still a lot cheaper than a new audio interface. Since I have a RME 9632 PCI card, I think I'll give this a try when I buy a new system later this year.
Old 4 weeks ago
  #244
Lives for gear
 
Pictus's Avatar
 

Quote:
Originally Posted by Elektronix View Post
I've read that there are Bios issues with the 3000 series on the MSI B450's, related to them having a 16MB ROM. MSI are going to be releasing "MAX" versions of their B450 boards with a 32MB ROM.

So you might want to wait a while if you want to go the B450 route.
With the new bigger BIOS CHIP they can add back support for the
older CPUs, not relevant for Ryzen CPUs.
https://www.msi.com/Motherboard/support/B450-TOMAHAWK
Old 4 weeks ago
  #245
So it seems that the 3700x doesn't really work with b450 boards. At least there is no guarantee right now as a lot of people are having big problems with them.

Anyone know how dpc latency is on any x570 boards?
Old 4 weeks ago
  #246
Lives for gear
 
Lesha's Avatar
Quote:
Originally Posted by imhungry View Post
So it seems that the 3700x doesn't really work with b450 boards. At least there is no guarantee right now as a lot of people are having big problems with them.
I would say that the majority works fine, it all depends on your POV if you listen to the loud minority.

I am certain that the BIOS updates will fix all problems soon.
Old 4 weeks ago
  #247
Lives for gear
 
Pictus's Avatar
 

Quote:
Originally Posted by imhungry View Post
So it seems that the 3700x doesn't really work with b450 boards. At least there is no guarantee right now as a lot of people are having big problems with them.
Looks like the BIOS still crude, this is probably the case for many motherboards...
https://www.reddit.com/r/MSI_Gaming/...date/?sort=new

Quote:
Anyone know how dpc latency is on any x570 boards?
One here
Old 4 weeks ago
  #248
Lives for gear
 
TAFKAT's Avatar
 

Quote:
Originally Posted by Pete Kaine View Post

If I'm reading that right (and I'm not sure that I am) smaller plug-ins make it easier to load balance cleanly without any massive spikes appearing in the metering. So yes, completely agree. It's also why we use smaller plugin's like SGA to ensure that's the case.

I'll stick my head in here and give my 2c.

As above, the point needs to stressed that the test needs to have an incremental and progressive load to be able to quantify performance variables more accurately, and its a balancing act re how small/large the respective individual overhead is per instance. RXC original was 8 Bands per instance , EXT is double that at 16, I am now building an RXC-EXT 32, literally doubling/quadrupling the overhead per instance as we have scaled over the course of the project. The reason that the RXC is a perfect reference plugin for the benchmark is it has unlimited bands and we can scale the overhead as required.

Quote:
The other problem for me is that nobody ever agrees on the plugins. We used to only do the DSP test but then everyone was asking for something with an instrument, so Vin knocked up the Vi test. In the time since I received requests for retesting with X/Y/Z plugin so that people have a better idea of what they should expect.
One thing I have learned over the years, that no matter what we develop or deliver, we can never appease everyone, so I don't even attempt to.

Having said that....

Quote:
The problem is everyone's asking for different things. Acustica, U-He both come up often, UVI, Native, Serum, Diva, Avenger.

The thing is, I'm perfectly well aware that some of those engines are far less efficient than the others, so some of them will load heavier than others.
I have decided to develop a DAWbench VI - EDM ( for want of a better name ), using u-he products to deliver a DSP VI testing environment , so we can have a wider scope. I am planning on using Zebra2, RePro and Diva in varying degrees as elements.

Current DAWbench VI - K5 Beta is being expanded to 8000 notes , which should give us a big enough boat till at least the end of the year.

Old 4 weeks ago
  #249
Lives for gear
 

I appreciate your work Vin, I really do... but "EDM"?...

....every time I read that I die a little on the insited...
Old 4 weeks ago
  #250
Lives for gear
 
TAFKAT's Avatar
 

Quote:
Originally Posted by mattiasnyc View Post
I appreciate your work Vin, I really do... but "EDM"?...

....every time I read that I die a little on the insited...
Ha, its just a working title for now, until I come up with something better :-)

Old 4 weeks ago
  #251
Quote:
Originally Posted by TAFKAT View Post
Ha, its just a working title for now, until I come up with something better :-)

DAWBENCH VI-SynthyMcSynthface...no?
Old 4 weeks ago
  #252
Lives for gear
 

Quote:
Originally Posted by TAFKAT View Post
Ha, its just a working title for now, until I come up with something better :-)

I say wait. Fire up the benchmark, run it on Ryzen and an i9, and if the i9 wins then keep the name to doubly "annoy" me!



jk

I really do appreciate your work.

Aren't we in a great place today? Regardless of what we choose; even competition elsewhere - gaming, data center, corporations etc - should bring either prices down or performance up. Good times again...
Old 4 weeks ago
  #253
Gear Head
 

Any news on when the 'creator' motherboards will be coming out?
Old 4 weeks ago
  #254
Great discussions guys! Interested in VIN's new VI bench (VINVI ).

Here's an unanswered question from earlier in this thread.

I found a "delta" between my current i7 4790K and the 3900X by going back through Petes benches. It ended with the 3900X giving roughly 265% more (64 samples, averaged between VI:s and comps).

Interestingly, CPU Benchmark delta is very similar (285%). Seems like a decent indicator now that AMD have got ok latency.

The question:
According to the above link, single threaded rating is actually not much improved. Does this mean that upgrading wouldn't really change the number of heavy plugins one can have on a *single* track, for example the master track?
Old 4 weeks ago
  #255
Lives for gear
 
Lesha's Avatar
These are memory kits with Hynix CJR chips than can be easily ran at 3600 C16, and are relatively cheap.
https://geizhals.de/g-skill-sniperx-...-a1815726.html
https://geizhals.de/g-skill-sniperx-...-a1815722.html

I bought a cheap 3000 CL15 Micron E-Die chip memory and I am running it @ 3533 CL16 stable using a Ryzen 1700 and MSI B450 Gaming Pro Carbon AC
https://geizhals.de/crucial-ballisti...-a1971747.html
Old 4 weeks ago
  #256
Lives for gear
 
Lesha's Avatar
Quote:
Originally Posted by thedberg View Post
The question:
According to the above link, single threaded rating is actually not much improved. Does this mean that upgrading wouldn't really change the number of heavy plugins one can have on a *single* track, for example the master track?
You will have a 16% improvement.
Old 4 weeks ago
  #257
Lives for gear
 
Sleazy_Rider's Avatar
 

I flashed my B450 MSI Pro Gaming Carbon AC with the latest BIOS without issues.

The Bios is the GSE lite version. You loose no functionality just GFX and bloat.

In fact I prefer it

The MAX B450s are identicle with just a bigger Bios chip..

Just waiting 3900X to hit stock...I bought a 3700x but cancelled it as I want 12/24 threads

I certainly won’t be swapping out my mobo...it’s ready and waiting to plonk in a 3900X
Old 4 weeks ago
  #258
Lives for gear
 
Pictus's Avatar
 

Quote:
Originally Posted by Pete Kaine View Post
Right, so I've had two Asus boards out and that Aorus and they are all hitting the same 800 DPC at about 15 mins... I'm starting to wonder if this is the much-acclaimed Windows 1903 spike I keep hearing about.
It can be...
Try some of the latest insider build.
Old 4 weeks ago
  #259
Quote:
Originally Posted by Lesha View Post
You will have a 16% improvement.
Argh. Does this benchmark really map so directly to our use case? Are improvements actually that meager after five years of processor development!?

Lot's of us will have one or a few tracks, often the master output, that are extra heavily burdened. Buying a new computer won't make much difference if this is true.

Well. At least I might put off a purchase and that saves money
Old 4 weeks ago
  #260
Lives for gear
 
Lesha's Avatar
Quote:
Originally Posted by thedberg View Post
Argh. Does this benchmark really map so directly to our use case? Are improvements actually that meager after five years of processor development!?

Lot's of us will have one or a few tracks, often the master output, that are extra heavily burdened. Buying a new computer won't make much difference if this is true.

Well. At least I might put off a purchase and that saves money
That is single thread performance improvement compared to Core i7-4790K.
The multithread improvement is 2.85x, so not too shabby
Old 4 weeks ago
  #261
Lives for gear
 
TAFKAT's Avatar
 

Quote:
Originally Posted by DAW PLUS View Post
DAWBENCH VI-SynthyMcSynthface...no?
VI-SMcSF-EXT !!

Quote:
Originally Posted by mattiasnyc View Post
I say wait. Fire up the benchmark, run it on Ryzen and an i9, and if the i9 wins then keep the name to doubly "annoy" me!

Now why would I ever want to do that ...

Old 4 weeks ago
  #262
Here for the gear
 

Quote:
Originally Posted by Pete Kaine View Post
Right, so I've had two Asus boards out and that Aorus and they are all hitting the same 800 DPC at about 15 mins... I'm starting to wonder if this is the much-acclaimed Windows 1903 spike I keep hearing about.
Is there any reason why you're using Asus boards in your company's builds and, specifically, not ASRock? I ask since the ASRock X570 boards support Thunderbolt via an Add-In Card. I'm not sure any other motherboard maker is doing that on AMD side thus far.
Old 4 weeks ago
  #263
Lives for gear
Quote:
Originally Posted by nonstatic View Post
The best low latency performance is currently found in (some) PCIe and Thunderbolt soundcards. PCIe is fine if you work in one location, but many of us don't. For those people, Thunderbolt interfaces are the best choice if you use lots of DSP/VI's and need excellent performance at very low latencies.

Note that Thunderbolt itself does not guarantee great low latency performance, you can find plenty of info on that in TAFKAT's sticky thread.

RME has proven it's possible to make class-leading performance over USB3, for example (only 25% the bandwidth of TB2). IMO, more manufacturers should have developed in that direction due to the nightmare of compatibility/licensing issues with Thunderbolt. Not to mention that the new USB 3.2 Gen 2x2 has the same theoretical max bandwidth as TB2 anyway. But that's another conversation.
As you properly note, Thunderbolt does not in itself guarantee great low-latency performance. Equally, our reader community must be careful to also not bind together concept of the bandwidth of a link as a predictor of its latency.

For example, good old PCI (not PCIe) was only about 1-gigabit, yet offered terrific low-latency because its low-overhead hardware-based point-to-point protocol data path to the RAM and CPU. PCIe does much more with the point-to-point (no need for link arbitration) connectivity and ramps up the bandwidth using lanes.

Thunderbolt is a packet-based implementation that blends streams of PCIe and DisplayPort, having some link arbitration to enable daisy-chaining while sacrificing little in terms of additional latency.

The USB family of connections is far less direct in its access to the RAM and CPU, and in fact requires quite a lot of CPU/driver work that drives up the latency. RME has done a stellar job in overcoming these time-consuming impediments.

Okay, the pedantic itch has been scratched today. See ya!
Old 4 weeks ago
  #264
Here for the gear
 

Quote:
Originally Posted by Pete Kaine View Post
I don't think I have access to any yet. We tend to favour Corsair here and I'm not entirely sure if they have 3600MHz with C16 timing as I can't see any in stock here and we normally hold it all. I'll have to follow up with my rep about it.

The current RAM tested is all C18.
To clarify - 3600MHz at C16 timing is, according to the thread here, not fast enough. It needs to be at least clocked at 3600Mhz, CL14, resulting in a total CAS latency of 7.77ns.

The total CAS latency needs to be lower than the sweet spot of 8.3ns.

The RAM you tested, likely at speeds between 3200Mhz (given Corsair LPX) has a latency of more than 11ns. This could very well be what's causing the much lower benchmarks compared to Intel chips, assuming the other guy is right. It's already a well-known fact that Ryzen memory speeds matter a lot more compared to Intel chips.

Again, the question we all want to have answered is if Ryzen performs better for DAW work with very low latency memory. If this is the case, then we can better select for what kind of RAM we should use for our DAW workstations, and it might have an effect on the type of benchmarks we should compare both with.

See this quote from the thread that I linked on why memory latency is apparently very important on Ryzen in regards to audio workloads:

Quote:
I teach you something most people don't know about Ryzen. Ryzen's memory controller has a sweet spot of CAS latency of 8.3ns. When the CAS is higher than that, you might notice CPU under-utilization in some cases. Now I'll teach you how to figure out CAS latency:

For example, for 3200mhz CL14 ram:

14/1600*1000 = 8.75ns

So, 3200mhz CL14 ram has a CAS latency of 8.75ns. That's pretty close to Ryzen's sweet spot of 8.3ns, but not quite there yet.

Let's try 3400mhz CL14 now:

14/1700*1000 = 8.23ns

That's much better. Slightly better than Ryzen's sweet spot 8.3ns.

Let's see 3466mhz CL14:

14/1733*1000 = 8,07ns

That's extremely good. Much better than the sweet spot of 8.3ns.

Let's see 3600 CL14:

14/1800*1000 = 7.77ns

This is some god like CAS latency.

Why is this important? Because audio production uses a lot of tracks and lots of plugins which uses a lot of memory. The lower the CAS latency, the faster the memory can addresses requests from the CPU, which in turn makes the CPU work better. All this is ideal for latency sensitive workload like audio production.

Forget about "core latency." Memory latency is the secret sauce. And lower memory latency equals higher memory bandwidth too, as everything is related.
And here's a quick chart for DDR4 memory and their latency calculations. The goal is to stay below 8.3ns, and the lower we get the better the performance should get (apparently). Disregard the green coloring.

Old 4 weeks ago
  #265
Lives for gear
 

Quote:
Originally Posted by Deleted User View Post
The total CAS latency needs to be lower than the sweet spot of 8.3ns.
Just fyi: "Sweet spot" means an ideal spot. There can't be anything better at a different point than the sweet spot, because that would make that other spot the sweet spot.

And it actually makes sense to consider this semantic difference as valuable because AMD itself has indicated that generally speaking - because of the infinity fabric's connection with the clock - there is indeed a sweet spot for memory speeds and below AND above that performance drops.

So, "sweet spot" is the ideal spot, not a "threshold" or anything.
Old 4 weeks ago
  #266
Lives for gear
 
CasimirsBlake's Avatar
 

Quote:
Originally Posted by narfs View Post
One can overclock 3200 CL16 E-die memory to 3600 CL16 pretty easily. And it goes for around 150 USD per 2x16GB (32 GB total). That seems fine to me.
No guarantee that will happen every time though.

Some would prefer to pay extra for guaranteed specs, knowing it will work.
Old 4 weeks ago
  #267
Here for the gear
 

Quote:
Originally Posted by CasimirsBlake View Post
No guarantee that will happen every time though.

Some would prefer to pay extra for guaranteed specs, knowing it will work.
No guarantee you will wake up every morning either.

Yet I've seen many reports of successful overclocks on Ballistix Sport LT (E-die) on AMD 3k platforms with this memory to 3600 CL16 and 0 failures thus far.

If you're not comfortable with overclocking (and by that I mean simply setting an XMP profile in BIOS), then spend the extra cash. But that was an obvious.

Last edited by narfs; 4 weeks ago at 02:59 PM..
Old 4 weeks ago
  #268
Lives for gear
 
Lesha's Avatar
Quote:
Originally Posted by narfs View Post
No guarantee you will wake up every morning either.

Yet I've seen many reports of successful overclocks on Ballistix Sport LT (E-die) on AMD 3k platforms with this memory to 3600 CL16 and 0 failures thus far.

If you're not comfortable with overclocking (and by that I mean simply setting an XMP profile in BIOS), then spend the extra cash. But that was an obvious.
I have the same memory kit and can overclock it to 3600 but it isn't stable yet. 3533 is fully stable, but have in mind I am still using a first gen Ryzen. Ryzen 3000 has much better memory compatibility so I guess 3600 should be stable there. I will be OCing to 3600 soon and see if it is possible to make it stable.
Old 4 weeks ago
  #269
Lives for gear
 

Quote:
Originally Posted by Pictus View Post
It can be...
Try some of the latest insider build.
But I can't sell it with that build in place, thus it's of no interest to me.

Quote:
Originally Posted by narfs View Post
Is there any reason why you're using Asus boards in your company's builds and, specifically, not ASRock? I ask since the ASRock X570 boards support Thunderbolt via an Add-In Card. I'm not sure any other motherboard maker is doing that on AMD side thus far.
Personal experience having used Asrock on and off in the past. For all the talk about VRM's failing on here, the only ones I've ever seen pop on a new board were on Asrock and it's more than once.

Also, is that card confirmed as working on the ASRock without any modding?
The last generation boards with the same header never received the update to permit it.

I work with ASUS first at launch because I know their BIOSes, I can generally overclock on them without thinking about it and I've got both their sales manager and support manager on speed dial. The same can't be said for a lot of other companies, although that's them having proven that I can't trust their support guys to get things solved quickly.

If I can't solve it with ASUS then my second choice tends to be MSI followed by ASRock for certain form factors although rarely overclocked.

Quote:
Originally Posted by Deleted User View Post
It needs to be at least clocked at 3600Mhz, CL14, resulting in a total CAS latency of 7.77ns.
So I need to fit a unicorn to make it work properly?

Quote:
Originally Posted by Deleted User View Post
The RAM you tested, likely at speeds between 3200Mhz (given Corsair LPX) has a latency of more than 11ns. This could very well be what's causing the much lower benchmarks compared to Intel chips, assuming the other guy is right. It's already a well-known fact that Ryzen memory speeds matter a lot more compared to Intel chips.
It's been claimed in the past and whilst it had some marginal gains for offline working, real-time audio never really benefited before. Hence why I just ended up doing the initial testing at the original 3200MHz recommendation from the last generation.

In light of AMD's new claims I've been retesting at 3725 (3600 clocked up as close as I can to 3733MHz) and I'm seeing some differences, I'll publish those later this week once I've done enough comparative testing to figure out how it sits.

Just to note, anything above 3600MHz is going to cost you 2X - 3X the cost of the 3200MHz RAM. I can almost believe at this point that earlier iterations of the chipset could have needed similar speed RAM but it was perhaps a little ahead of its time as it were.

What I have been initially tackling is what happened to Cubase in this round of testing, you can check that over here: http://www.scanproaudio.info/2019/07...er-comparison/
Old 4 weeks ago
  #270
Here for the gear
 

Quote:
Originally Posted by Pete Kaine View Post
So I need to fit a unicorn to make it work properly?
No, you could just try to find better (faster and lower latency RAM) compared to Corsair LPX with latencies of 11ns, and see if that makes a difference in the benchmarks at for Zen2. The point is that right now, we don't know yet if it does. The chart I linked has the CAS latency calculations - anything around 8.3ns should be the "sweet-spot".

Quote:
Originally Posted by Pete Kaine View Post
In light of AMD's new claims I've been retesting at 3725 (3600 clocked up as close as I can to 3733MHz) and I'm seeing some differences, I'll publish those later this week once I've done enough comparative testing to figure out how it sits.
What's the CAS latency on those sticks?
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
Forum Jump