[dougms3]

Has anyone kept up with the tech involved with DDR5 ram?

It seems they developed it with noise and jitter reduction capabilities.

https://resources.altium.com/p/ddr5-pcb-design-and-signal-integrity-what-designers-need-to-know

I'm wondering how much of an improvement it might offer for computer audio.

 

[bemeurer]

None.

The signal integrity improvement is to allow higher transfer rates to the DRAM IC. One could make the claim that due to DDR5's inherent error-correction mechanism, it would be possible that a corruption to an audio sample loaded into DRAM could be prevented, but any effects there are statistically insignificant.

There is a good paper on the occurrence of DRAM, but unfortunately, it's from back in DDR2 times. There is, however, a more recent paper with recently similar findings. In that last one, they observed 55,000 faults over 12,000 terabyte-hours of experimentation. A normal system isn't using DRAM so proactively, the experiment was designed to stress-test DRAM, so the error rates would be _far_lower.

In other words, the likelihood that a stochastic DRAM error would happen to hit some audio data, and that the 1-bit flip would be in a place significant enough to meaningfully alter the sound, is essentially null. And that's just thinking about one 24-bit sample, the idea that it'd happen repeatedly and contiguously to cause an audible effect is even more inconceivable.

You'd have better luck preventing cosmic rays from causing bit corruption.

 

[dougms3]

None.

The signal integrity improvement is to allow higher transfer rates to the DRAM IC. One could make the claim that due to DDR5's inherent error-correction mechanism, it would be possible that a corruption to an audio sample loaded into DRAM could be prevented, but any effects there are statistically insignificant.

There is a good paper on the occurrence of DRAM, but unfortunately, it's from back in DDR2 times. There is, however, a more recent paper with recently similar findings. In that last one, they observed 55,000 faults over 12,000 terabyte-hours of experimentation. A normal system isn't using DRAM so proactively, the experiment was designed to stress-test DRAM, so the error rates would be _far_lower.

In other words, the likelihood that a stochastic DRAM error would happen to hit some audio data, and that the 1-bit flip would be in a place significant enough to meaningfully alter the sound, is essentially null. And that's just thinking about one 24-bit sample, the idea that it'd happen repeatedly and contiguously to cause an audible effect is even more inconceivable.

You'd have better luck preventing cosmic rays from causing bit corruption.

I'm referring to the noise and jitter reduction tech not error correction.

The PC is an EMI storm generator, it injects noise from all the components and spreads around everywhere. A reduction in noise and jitter from the ram may be beneficial to sound quality.

 

[bfreedma]

I'm referring to the noise and jitter reduction tech not error correction.

The PC is an EMI storm generator, it injects noise from all the components and spreads around everywhere. A reduction in noise and jitter from the ram may be beneficial to sound quality.

As the previous poster explained in accurate detail, it will have no effect. Audio playback takes an such incredibly small percentage of DDR5’s potential throughput that jitter is irrelevant. Even if that wasn’t the case, jitter is not targeting a specific data type and certainly isn’t acting as a repeatable audible filter.

Read the linked material - it goes into great detail and explains, conclusively, why DRAM jitter is not and can not impact audio playback.

BTW, prior to DDR5, the data rates in RAM were below the transfer rate threshold where jitter was a potential internal issue, so DDR5 didn’t improve the jitter performance when compared to previous memory architectures. And in DDR5, these techniques are only applied when the chip nears its max transfer rate, certainly not something any audio stream is going to approach. https://ieeexplore.ieee.org/document/8329722

TL/DR: DDR5 can’t improve audio compared to previous generations of DRAM because the transfer rates on those older technologies didn’t introduce jitter.

 

[dougms3]

As the previous poster explained in accurate detail, it will have no effect. Audio playback takes an such incredibly small percentage of DDR5’s potential throughput that jitter is irrelevant. Even if that wasn’t the case, jitter is not targeting a specific data type and certainly isn’t acting as a repeatable audible filter.

Read the linked material - it goes into great detail and explains, conclusively, why RAM jitter is not and can not impact audio playback.

Again, the linked material goes into great detail and explains alot about error correction. I'm not interested with bandwidth, throughput or error correction.

I'm interested in the noise and jitter reduction.

Jitter affects all data, some are negligibly affected but with audio, it makes a difference.

I searched the articles and they mention absolutely nothing about jitter or audio. Unless I'm missing it, would you be so kind as to provide an excerpt where it states that?

 

[bfreedma]

Again, the linked material goes into great detail and explains alot about error correction. I'm not interested with bandwidth, throughput or error correction.

I'm interested in the noise and jitter reduction.

Jitter affects all data, some are negligibly affected but with audio, it makes a difference.

I searched the articles and they mention absolutely nothing about jitter or audio. Unless I'm missing it, would you be so kind as to provide an excerpt where it states that?

You clearly didn’t read the link I posted as this is the first paragraph:

“Abstract:
Intel's signal integrity (SI) analysis for memory in the server segment has neither considered correlated jitter nor handled jitter amplification over channel when performing fast analytical signaling analyses. This inaccuracy is no longer feasible with the intended data rates of DDR5. Here, we propose a DDR5 flow that starts from jitter sequences or histograms and ends with signaling analysis via FastBER that can comprehend jitter amplification over I/O channels regardless of whether Tx jitter is correlated or uncorrelated. The paper contributes results of large scale testing of FastBER signaling in such a scenario and also offers solutions to practical issues like jitter extrapolation and time-margin skew due to setup and hold asymmetry.”

I’m getting the sense that you don’t understand how trivial an audio stream is when compared to DDR5’s max throughput. You’re looking for a problem that simply doesn’t exist in this use case. Keep in mind that we’re discussing jitter within the chip and it’s supporting PCB. None of this is (or can) impact audio output.

If you have peer reviewed documents that support your case (not marketing material from a vendor), please post it.

Edit - Micron‘s DDR5 SDRAM data sheet provides great detail on throughput and jitter. https://media-www.micron.com/-/medi...?rev=f76eb9631b674e66a2026c324a95cb67#page329 I’ve directly linked you to the section on jitter, but it’s best to read the entire document to understand the use cases where one should or should not be concerned with DDR5 and jitter.

 

[bemeurer]

I'm referring to the noise and jitter reduction tech not error correction.

The PC is an EMI storm generator, it injects noise from all the components and spreads around everywhere. A reduction in noise and jitter from the ram may be beneficial to sound quality.

If that's what you're concerned about, I imagine DDR5 will, if anything, be worse due to the higher transfer rates. Moreover, the article you linked talks about the DRAM signal integrity, but that has no bearing on how much noise (i.e. interference) it's causing to other things. I imagine future DDR iterations, which will likely involve techniques like PAM and QAM signaling, will only make things worse from this perspective.

I do want to point out that a reduction in DRAM jitter will have no impact on sound. It has nothing to do with how jitter in USB behaves because here it's related to the DRAM IC's latency characteristics. The Micron documentation helps clarify this. (c.f. Unit Interval and Jitter Definitions)

Again, the linked material goes into great detail and explains alot about error correction. I'm not interested with bandwidth, throughput or error correction.

I'm interested in the noise and jitter reduction.

Jitter affects all data, some are negligibly affected but with audio, it makes a difference.

I searched the articles and they mention absolutely nothing about jitter or audio. Unless I'm missing it, would you be so kind as to provide an excerpt where it states that?

Bandwidth, noise, and jitter are all correlated factors here.

It's not true that all jitter affects all data. For example, network is extremely jittery, in the order of miliseconds usually, but due to how TCP works it does not affect the data under normal conditions. Extreme jitter, obviously, can lead to packet loss and signal degradation, but my point is plenty of channels are inherently jittery and work without loss due to that being accounted for in elsewhere in the system (such as by employing resending with TCP).

And yes, the articles make no mention of audio because it is so far removed from the question of DRAM jitter as to be irrelevant. It's akin to saying, "I read the article on NAND healing via annealing, but it made no mention of the sonic impact!"

 

[bemeurer]

A point I forgot in my previous message: You can just build a Faraday cage around your computer if you think EMI is causing issues. Then use a USB interface with a powerful PLL and you can also forget about jitter.

Alternatively, leave the PC in a separate room and use a network bridge.

Although, to be honest, I find the premise that EMI from a PC built with FCC-compliant parts is causing a ruckus on other devices dubious.

 

[dougms3]

You clearly didn’t read the link I posted as this is the first paragraph:

“Abstract:
Intel's signal integrity (SI) analysis for memory in the server segment has neither considered correlated jitter nor handled jitter amplification over channel when performing fast analytical signaling analyses. This inaccuracy is no longer feasible with the intended data rates of DDR5. Here, we propose a DDR5 flow that starts from jitter sequences or histograms and ends with signaling analysis via FastBER that can comprehend jitter amplification over I/O channels regardless of whether Tx jitter is correlated or uncorrelated. The paper contributes results of large scale testing of FastBER signaling in such a scenario and also offers solutions to practical issues like jitter extrapolation and time-margin skew due to setup and hold asymmetry.”

You clearly didn't read that paragraph.

"DDR5 can’t improve audio compared to previous generations of DRAM because the transfer rates on those older technologies didn’t introduce jitter"

Show me where it says that, don't mix in your own conclusions with what it says in a research article.

I see you went back and edited your post. If you're trying to troll, do a better job.

 

[dougms3]

If that's what you're concerned about, I imagine DDR5 will, if anything, be worse due to the higher transfer rates. Moreover, the article you linked talks about the DRAM signal integrity, but that has no bearing on how much noise (i.e. interference) it's causing to other things. I imagine future DDR iterations, which will likely involve techniques like PAM and QAM signaling, will only make things worse from this perspective.

I do want to point out that a reduction in DRAM jitter will have no impact on sound. It has nothing to do with how jitter in USB behaves because here it's related to the DRAM IC's latency characteristics. The Micron documentation helps clarify this. (c.f. Unit Interval and Jitter Definitions)

Bandwidth, noise, and jitter are all correlated factors here.

It's not true that all jitter affects all data. For example, network is extremely jittery, in the order of miliseconds usually, but due to how TCP works it does not affect the data under normal conditions. Extreme jitter, obviously, can lead to packet loss and signal degradation, but my point is plenty of channels are inherently jittery and work without loss due to that being accounted for in elsewhere in the system (such as by employing resending with TCP).

And yes, the articles make no mention of audio because it is so far removed from the question of DRAM jitter as to be irrelevant. It's akin to saying, "I read the article on NAND healing via annealing, but it made no mention of the sonic impact!"

I agree with the higher speeds, I imagine there will be greater EMI, I think thats why implemented the noise and jitter reduction tech.

Not only the jitter but noise as well. There are no studies on if jitter affects audio or not, its all theories and conjecture, no one cares except us. It is 100% true that jitter affects all data but as I said, its neglible with most things but not audio. If you're using a streaming service for music, the network jitter will absolutely affect the audio.

20 years ago, the effect of jitter on audio was claimed to be a myth, there aren't many in depth studies or research on this because for the most part again, no one cares except us.

If you can imagine that we can hear the difference with something like a reclocker that adjusts timing in femtoseconds (a quadrillionth of a second), do you think its possible that a fraction of a millisecond can make a difference in sound as well?

I'm more interested in the specific way this tech works in DDR5 to reduce noise.

A point I forgot in my previous message: You can just build a Faraday cage around your computer if you think EMI is causing issues. Then use a USB interface with a powerful PLL and you can also forget about jitter.

Alternatively, leave the PC in a separate room and use a network bridge.

This is not how faraday cage works. The faraday cage will prevent the emi going into or escaping it, it doesn't cancel out all the EMI generated internally.

Although, to be honest, I find the premise that EMI from a PC built with FCC-compliant parts is causing a ruckus on other devices dubious.

Dubious as you may find it, FCC compliance is FCC compliance, it doesn't mean its not emitting EMI and interfering with the audio signal.

 

[joe]

Mod note: Guys, let's drop the personal attacks. I know it's summer, but I don't want to hand out vacations. Can we please be civil?

 

[redrol]

RAM either works %100 or doesn't and the computer crashes. You are conflating things that dont do anything for sound.

 

[dougms3]

RAM either works %100 or doesn't and the computer crashes. You are conflating things that dont do anything for sound.

Lol so angry.

Thanks for pointing that out. Theres docp, timing, voltage and speed overclocking that can be done to maximize the capabilities of the ram, depending on various factors including the silicon lottery. Maybe for you, it works %100 or doesn't but its a little more complicated for others.

This particular part is what I'm interested in.

"There are other sources of noise in DDR5 channels that become even more problematic than in previous generations, especially given the higher speeds required to accommodate the higher data/clock rates (e.g., 3.2 GHz.in DDR5-6400). Despite the use of equalization to overcome channel losses and distortion, extremely precise termination and impedance control is required to ensure equalization can provide signal recovery. In such low voltage devices, these impedance mismatch constraints become ever more critical to ensuring a low BER."

As bemeurur pointed out the higher transfer speeds will create more noise and this is probably why they implemented this noise and jitter reduction tech in the ddr5 ram.

I don't know much about this tech and I found it to be interesting so I thought it'd nice to discuss here in this sound "science" section.

 

[redrol]

You are not in Sound science, you are in Computer Audio. And again, RAM either works or it doesn't, and has zero to do with the audio output of a computer system being correct.
Did you know that the word impedance and termination can be used to describe science things that are not related to sound?

 

[dougms3]

Oops, oh well, I guess I posted in the right section then.

Yes I am aware. Point?