Intel 14th/13th gen, latency/stability issues?
-
- KVRian
- 782 posts since 9 May, 2005
Just ran Cinebench 2024 on M2 Ultra, 13900k, and 14900k.
M2 Ultra multi-core = 1928
13900k multi-core = 2120
14900k multi-core = 2195
14700k multi-core = 2041
Obviously Cinebench doesn't represent all real-world facets/scenarios.
When you're trying to have a conversation about performance, a custom DAW test isn't something that can be conveniently quantified across multiple users. Both parties would need 100% identical plugins/project.
Some users would then argue that they don't use said tested plugins... or a project of said construction.
Cinebench is a useful tool to conveniently compare performance.
Everyone can access, it's consistent, and it takes a mere 10 minutes
14700k outperforms M.2 Ultra (let alone 13900k, 14900k, and 14900ks).
With Mac Studio, you're currently leaving a lot of drive performance on the table.
A nice TB/USB4 external M.2 Enclosure sustains ~3659MB/Sec... while the PCIe 4.0 M.2 drive itself can sustain ~7300MB/Sec.
The machine next to me runs half a dozen PCIe 4.0 M.2 drives at full speed.
Threadripper has always been a bad choice for DAW purposes.
350w TDP (7980x)
Active Cooled chipsets
Poor ultra-low-latency performance
A well-spec'd 7980x machine clocks-in the better part of $10k. CPU alone is $5k.
Ryzen 7950x3D is a good low-latency performer... but you're paying slightly more for slightly lower performance (vs Intel).
Intel's Arrow Lake (15th Gen) and Ryzen 9xxx are also in the pipe (released later this year).
Arrow lake will bring native PCIe 5.0 support for M.2 drives... so you'll be able to connect PCIe 5.0 M.2 SSDs for 12,0000+MB/Sec.
M2 Ultra multi-core = 1928
13900k multi-core = 2120
14900k multi-core = 2195
14700k multi-core = 2041
Obviously Cinebench doesn't represent all real-world facets/scenarios.
When you're trying to have a conversation about performance, a custom DAW test isn't something that can be conveniently quantified across multiple users. Both parties would need 100% identical plugins/project.
Some users would then argue that they don't use said tested plugins... or a project of said construction.
Cinebench is a useful tool to conveniently compare performance.
Everyone can access, it's consistent, and it takes a mere 10 minutes
14700k outperforms M.2 Ultra (let alone 13900k, 14900k, and 14900ks).
With Mac Studio, you're currently leaving a lot of drive performance on the table.
A nice TB/USB4 external M.2 Enclosure sustains ~3659MB/Sec... while the PCIe 4.0 M.2 drive itself can sustain ~7300MB/Sec.
The machine next to me runs half a dozen PCIe 4.0 M.2 drives at full speed.
Threadripper has always been a bad choice for DAW purposes.
350w TDP (7980x)
Active Cooled chipsets
Poor ultra-low-latency performance
A well-spec'd 7980x machine clocks-in the better part of $10k. CPU alone is $5k.
Ryzen 7950x3D is a good low-latency performer... but you're paying slightly more for slightly lower performance (vs Intel).
Intel's Arrow Lake (15th Gen) and Ryzen 9xxx are also in the pipe (released later this year).
Arrow lake will bring native PCIe 5.0 support for M.2 drives... so you'll be able to connect PCIe 5.0 M.2 SSDs for 12,0000+MB/Sec.
-
- KVRAF
- 1606 posts since 20 Feb, 2003
Yes, no benchmark can represent all real-world scenarios. But that’s particularly true for DAW usage due to the variables involved. It’s very unwise, if building a DAW today, to base decisions on anything outside of benchmarks directly involving the intended host software you’ll use.Jim Roseberry wrote: ↑Mon May 06, 2024 7:16 pmObviously Cinebench doesn't represent all real-world facets/scenarios.
Yep, that presents issues if someone wants an easy number they can look at. But performance variability, between hosts, is simply too large to do otherwise. Soundcards and plugin choice further impact both the performance and stability. Cinebench communicates nothing about any of this. Instead it suggests a 64 core Threadripper is the best choice. Which, as you acknowledge, may not quite be the case with audio
Benchmarks themselves can also be at issue. Cinebench 23 would’ve caused incorrect assumptions regarding any performance delta between Intel and Apple Silicon. Your v24 results range between between a 5.5% (14700) and 12.1% (14900) difference VS M2 Ultra. How easily might such a percentage be swallowed up by the many variables involved in a host running plugins?
Along with their ability to mislead someone assuming how much differences translate elsewhere, the over-focus on benchmarks has arguably resulted in other negative consequences. EG Many motherboards have recently used defaults which ran Intel’s latest chips out of spec. Better benchmark scores were, apparently, more important to these mobo companies than their customers system stability. Many 13th /14th gen tests will now benchmark lower - if users want a stable system.
Outperforms for what though? I use Cubase. Tests of plugin counts, sample rates and buffers etc, are something I might relate to. Software, which utilises resources as differently as Cinebench? Perhaps not so much.14700k outperforms M.2 Ultra (let alone 13900k, 14900k, and 14900ks).
On this point I'd mostly agree. Thunderbolt 4 (40Gb/s) reserves 8Gb for video so displays always work when connected. So you’re left with the equivalent of 6 PCIE3 X4 (24) lanes, spread across the 6 ports, not counting the 2 USB ports. That’d mean 6 SSD’s topping out ~2850MB/s each. A limitation? For raw video.. Maybe. But not exactly the biggest limitation for most peoples audio useWith Mac Studio, you're currently leaving a lot of drive performance on the table.
About the only feasible audio scenario might be loading 100GB+ orchestral templates into ram quickly, though other overheads would likely reduce the differences. But it’d be an interesting test for anyone in that boat, perhaps.. and maybe one of the few reasons (for now) why someone might choose a Mac Pro, over a Studio, if inclined to go in that direction.
Thunderbolt appeared to cause an awful lot of issues for Windows systems though. Hopefully USB 4 will fair better. All I can say is Cubase 13, on Apple Silicon / Sonoma, has been the most stable version I’ve used in 20+ years, Windows or Mac.
-
- KVRian
- 782 posts since 9 May, 2005
I agree with much of what you're saying.PAK wrote: ↑Tue May 07, 2024 5:21 am Along with their ability to mislead someone assuming how much differences translate elsewhere, the over-focus on benchmarks has arguably resulted in other negative consequences. EG Many motherboards have recently used defaults which ran Intel’s latest chips out of spec. Better benchmark scores were, apparently, more important to these mobo companies than their customers system stability. Many 13th /14th gen tests will now benchmark lower - if users want a stable system.
Gaming benchmarks have always been a bit skewed.
ie: In some motherboards, you'll see parameters for running specific benchmarks.
Gaming drives hardware development and sales.
Witness video card packaging/marketing (obviously targeting young guys).
I've built a LOT of 13th/14th Gen machines (for both myself and many clients).
None have stability issues... and their benchmark performance hasn't decreased.
There's been a lot of YouTube hype about this subject (Asus)... but the user has always been in control. All necessary settings are there to prevent instability (and or thermal-throttle).
I do agree that Default settings should not be "cooking" the CPUs.
What I've seen that's a bit peculiar.
In 30 years (pre Covid), I experienced 5 defective CPUs.
Post Covid (less than 5 years), I've experienced 5 defective CPUs.
If you deal with enough quantity, there's always a certain percentage of defective parts.
QC post Covid is worse than pre
Since Win10 officially supported "PCIe via Thunderbolt" (with Thunderbolt-3 controllers... and then Thunderbolt-4)... it's always been rock-solid.
Those who have issues don't know what they're doing.
Myself and many clients use UA Apollo, RME Fireface UFX+, Presonus Quantum, and Antelope.
All have been rock-solid (never disconnect or act flaky).
Online, you'll see folks complaining about Antelope.
I wouldn't swap my Orion Studio Synergy Core for anything (short of a newer version).
TB is down to the hardware and configuration.
Tend the details... and it works beautifully.
- KVRAF
- Topic Starter
- 4736 posts since 1 Aug, 2005 from Warszawa, Poland
That's a kind of a pat on the back that I was looking for. I assume you didn't experience any weird latency issues either.Jim Roseberry wrote: ↑Tue May 07, 2024 12:35 pm I've built a LOT of 13th/14th Gen machines (for both myself and many clients).
None have stability issues...
Anyway I'm into Intel again, jumping from 1st gen to 14th. I'll be getting ASUS z790 Hero for mobo this week and work from there. I'll see if I can get firewire working, I hope not to have to update from Fireface 400...
Now I'm pondering, would it be worth getting separate small SSD (in today's world that would be 500GB) for system, or carve a 256GB partition from 4TB SSD I'll be getting anyway. Meaning would they share same bandwidth anyway or would they have reserved lanes.
-
- KVRian
- 782 posts since 9 May, 2005
Definitely no DPC Latency issues.Zombie Queen wrote: ↑Tue May 07, 2024 12:54 pmThat's a kind of a pat on the back that I was looking for. I assume you didn't experience any weird latency issues either.Jim Roseberry wrote: ↑Tue May 07, 2024 12:35 pm I've built a LOT of 13th/14th Gen machines (for both myself and many clients).
None have stability issues...
Anyway I'm into Intel again, jumping from 1st gen to 14th. I'll be getting ASUS z790 Hero for mobo this week and work from there. I'll see if I can get firewire working, I hope not to have to update from Fireface 400...
Now I'm pondering, would it be worth getting separate small SSD (in today's world that would be 500GB) for system, or carve a 256GB partition from 4TB SSD I'll be getting anyway. Meaning would they share same bandwidth anyway or would they have reserved lanes.
As I mentioned above, the 14900k allows you to do things like run IK's ToneX (similar to Kemper but in a plugin) at 1ms total round-trip latency.
You couldn't do that (glitch-free) with DPC Latency issues.
Firewire shouldn't be an issue.
We have several clients still running Fireface 400 audio interfaces.
What a tribute to RME that these interfaces are still working nearly 20 years later.
Smaller M.2 SSDs aren't expensive.
I'd want a minimum of three drives.
- OS/Apps
- Audio
- Samples
From a performance perspective, you're going to be best with three smaller M.2 SSDs (vs a single large partitioned drive).
- KVRAF
- Topic Starter
- 4736 posts since 1 Aug, 2005 from Warszawa, Poland
It was my assumption. I've got a Startech card that is supposed to be natively PCIe. I'll see how that works out.Jim Roseberry wrote: ↑Tue May 07, 2024 1:57 pm Firewire shouldn't be an issue.
We have several clients still running Fireface 400 audio interfaces.
Yes. I do not intend to change it unless I'm forced to, which will probably come from software side, I assume. If it starts to act out, I'll swap power caps and try to go on.What a tribute to RME that these interfaces are still working nearly 20 years later.
.From a performance perspective, you're going to be best with three smaller M.2 SSDs (vs a single large partitioned drive)
Thanks, I'll look into it. If I have budget, I'll go with two, system/data.
- KVRAF
- Topic Starter
- 4736 posts since 1 Aug, 2005 from Warszawa, Poland
Actually, my current PC, board/cpu/mem is now 14 years old. It is also working without a glitch.Jim Roseberry wrote: ↑Tue May 07, 2024 1:57 pm What a tribute to RME that these interfaces are still working nearly 20 years later.
- KVRian
- 987 posts since 21 Aug, 2017 from Brasil
The PCIe cards with the TI XIO2213b chip are know to workZombie Queen wrote: ↑Tue May 07, 2024 12:54 pm Anyway I'm into Intel again, jumping from 1st gen to 14th. I'll be getting ASUS z790 Hero for mobo this week and work from there. I'll see if I can get firewire working, I hope not to have to update from Fireface 400...
https://www.amazon.com/Profile-PCI-Expr ... 002S53IG8/
https://gearspace.com/board/music-compu ... s-daw.html
I would wait for the new AMD.
https://www.hardwaretimes.com/amd-ryzen ... ase-zen-5/
but if really wants Intel, ASRock Z790 NOVA WiFi, Z790 Riptide WiFi looks very interesting.
https://www.overclock.net/threads/asroc ... d.1808124/
- KVRAF
- Topic Starter
- 4736 posts since 1 Aug, 2005 from Warszawa, Poland
As I said, I have that part already. And yea, it has that chip.
I was considering Taichi, but it has no flash from USB, so I would have to put in a 13th gen CPU first or get to a service to flash. Plus, people were really trashing this mobo on forums. Anyway, I already ordered Asus. Also considered Gigabyte Master (no TB4) and MSI Ace (no iGPU support).but if really wants Intel, ASRock Z790 NOVA WiFi, Z790 Riptide WiFi looks very interesting
-
- KVRian
- 782 posts since 9 May, 2005
You won't be disappointed with the Z790 Hero.
That's what's in the machine I'm typing on.
That's what's in the machine I'm typing on.
- KVRAF
- Topic Starter
- 4736 posts since 1 Aug, 2005 from Warszawa, Poland
-
- KVRian
- 548 posts since 1 Jul, 2009
Well, it seems I do now. It happened a couple of times while listening to music in foobar2000. Audio stops for milliseconds and resumes. One time, the audio went out for 1 second or more and USB errors showed up in RME Fireface Settings App, which is rare. I then disabled USB selective suspend in windows power plan, let's see if it still happens. I have to say, these new CPU's are fast but they come with headaches. You need to research days or weeks to know what settings or software to use, to get a glitch free PC for audio work...PITAZombie Queen wrote: Key question, do you experience any audio playback glitches, outside of Bitwig? In everyday listening? I'm trying to figure out if it's a virtual problem or a real life annoyance.
- KVRAF
- Topic Starter
- 4736 posts since 1 Aug, 2005 from Warszawa, Poland
Now you are telling me. I just picked 14700 from mail box...
Do you have "virtualization" enabled in BIOS? I've seen suggestions it may have something to do with it.
Anyway, looks like I'm getting into most exciting/controversial platform, if anything it doesn't look like boring, CPUs cooking, CPUs bending...
So do you guys use third party CPU mounting frames (Thermal Grizzly / Thermaltake)?
It's funny, Intel can design and manufacture processor structures made of billions of transistors and at the same time they are incapable of making the frigging mounting mechanism right. As the matter of fact, as I look at the CPU box now, they can't even carve the window in the paper box properly.
So, I've seen guys claiming they have CPUs bent by factory mount. And I wonder... Now we have stability problems appearing over time with highest end units... Which are ones running really hot while being squeezed unevenly. Isn't it easier to bend hot objects?
Do you have "virtualization" enabled in BIOS? I've seen suggestions it may have something to do with it.
Anyway, looks like I'm getting into most exciting/controversial platform, if anything it doesn't look like boring, CPUs cooking, CPUs bending...
So do you guys use third party CPU mounting frames (Thermal Grizzly / Thermaltake)?
It's funny, Intel can design and manufacture processor structures made of billions of transistors and at the same time they are incapable of making the frigging mounting mechanism right. As the matter of fact, as I look at the CPU box now, they can't even carve the window in the paper box properly.
So, I've seen guys claiming they have CPUs bent by factory mount. And I wonder... Now we have stability problems appearing over time with highest end units... Which are ones running really hot while being squeezed unevenly. Isn't it easier to bend hot objects?
-
- KVRian
- 548 posts since 1 Jul, 2009
Congrats . It's a beast but needs to be tamed.
I have the Thermaltake contact frame. Red about the bending so I bought one for peace of mind.
Yes, virtualization is enabled. Maybe I will disable it to see if it helps.
I have the Thermaltake contact frame. Red about the bending so I bought one for peace of mind.
Yes, virtualization is enabled. Maybe I will disable it to see if it helps.
- KVRAF
- Topic Starter
- 4736 posts since 1 Aug, 2005 from Warszawa, Poland
My understanding is that you should enable it only if you actually need it, like running VM. Let us know, if disabling it has any effect