DIY Project High Performance Audio PC with high quality wiring

[StreamFidelity]

Yesterday I continued with the Solarflare NIC (Network Interface Card) X2522.

 

It is important to have good contact with the solarflare chip, because the high-performance chip gets very hot.

 

The spring pins must be pushed in until they lock.

 

 

The HDPLEX back plate is of better quality than the old version, but the fis Audio 4mm aluminum plate is much better and more accurate.

 

 

Below is a top view in the fully assembled state. At the top right is the low-noise Taiko Audio 500W DC DC-ATX converter. The Motherboard MSI MPG-Z690-EDGE-WIFI-DDR4 follows in the middle. It receives clean current via high-purity copper strands with gold-plated Molex pins and pure silver solder.

 

This is followed by the G.Skill DIMM 16 GB DDR4-4000 kit memory, which is located in the RAM banks in a certain order. Below is the massive passive cooling device from the HDPLEX H5 Version 3 chassis for the Intel® Core ™ i9-12900K processor. One of the most effective cooling devices I know. On the right is the passive cooling device of the Solarflare X2522 NIC. At the bottom right the CPU is supplied with power by two fis Audio Molex 8 pin - 8 pin ATX from the Keces P28.

 

 

I've tried a lot on ethernet cards, but none of them come close to Solarflare. It's not just the galvanic isolation via fiber optics, but above all the very low latencies and jitter combined with high power reserves. While I read elsewhere that open remote connections impair the sound, this is no longer the case here. Copying data back and forth or loading the complex metadata from Roon are carried out very quickly. Differences between playing from the memory or from the Internet are no longer perceptible to me with this card. The firmware also contributes to this. The Solarflare X2522 NIC can be configured for ultra low latencies in the BIOS.

 

 

The lid was replaced by a 10mm acrylic glass with optimized ventilation. Some also use the ventilation openings as cup holders. 😂

 

 

My experience installing and setting up the new components

 

The new HDPLEX Chassis impresses with its very good and massive passive cooling. Only a few adjustments had to be made, such as the CPU holder. In terms of processing quality, Streacom is still better, but the cooling performance is comparatively poor. A Streacom F10 only allows PCIe cards with riser cables, which for me is a bad design.

 

The MSI motherboard was a challenge. It initially booted into the BIOS with an error message. As it turned out later, the cause was the XMP (Extreme Memory Profile). I approach the RAM with a slower rate. Otherwise the BIOS menu is responsive, easy to use and very extensive.

 

In the beginning I used the HQPlayer operating system. You burn it on a USB stick and boot with it. Immediately the internet connection was up and the music played. The installation of Windows 11 Pro was a bit more cumbersome. Not all drivers were present, of course. Despite the usual struggling, Windows is currently performing better. But here I am still waiting for an update from the HQPlayer because the cores were not all controlled correctly.

 

A sound report will follow.

[StreamFidelity]

6 hours ago, luisma said:

Beautiful and thank you for reporting back, the only conflict I have is why and how could you made it passive?

 

Thank you very much first.

 

Fans or no fans are a concern for a lot of people. Here is my point of view.

 

I basically don't want any moving parts in my electronic devices. What doesn't move can't vibrate either. Lots of people spend a lot of money on anti-vibration platforms and device feet for good reason. That shouldn't apply to the audio computer of all things?

 

I had installed a 140mm case fan in an audio PC in the Streacom FC9, because the passive cooling for the Intel i9-9900K was not sufficient. Inaudible at 200rpm. Vibrations were mitigated by a rubber suspension.

 

 

Later I installed a SOtM fan filter because the PWM control caused electronic interference. I wasn't really happy with it. 

 

 

Some place fans outside the device and cool from the outside. This avoids electronic interference, but the cooling is very inefficient. In addition, there is the development of dust. Where nothing moves, no dust can be raised. In the extreme case it looks like this. 😂

 

Source: https://www.igorslab.de/15-dinge-die-man-besser-lassen-sollte/15/

 

With all activated cores (8 P-Cores 4.2GHz + 8 E-Cores 3.6GHz) the Intel i9-12900K CPU did not get warmer than 80 °C. In current operation with 8 P-Cores 4.0GHz and AVX512, the temperature is approx. 60 °C. Passive cooling can work very well.

[StreamFidelity]

6 hours ago, chipvn said:

From what I saw, he is running HQPlayer with the filter and modulator combination doesn't required more than 40W CPU Total package power on the 12900k (enable only 8 power cores and max CPU clock max at 4GHz).

Thats correct 👍

+ enable AVX512!

 

On 12/1/2021 at 6:49 PM, StreamFidelity said:

44.1kHz to DSD256x48 / ASDM7ECv2

 

 

96kHz to DSD256x48 / ASDM7ECv2

 

 

My final assessment of the sound quality (SQ)

 

I notice a more precise, punchier bass and "springy" transients in percussion and guitar sounds. But maybe it's just me, because my body moves to the rhythm of the music the whole time. 😂 I am very enthusiastic.

[StreamFidelity]

The manual is one of the most accurate I've ever read. 👍 It says on page 43:

Quote

The primary 24-pin power supply connector (JPWR1) meets the ATX SSI EPS 12V specification. An 8-pin (JPWR2) processor power connector must also be connected to your power supply.

 

I am assuming that the Taiko DC-ATX converter connected both 24 pin ATX (JPWR1) and 8 pin to the CPU (JPWR2)?

 

29 minutes ago, Patatorz said:

I'm testing a Taiko DC/ATX module and during the test i'm trying to power the CPU of the Motherboard with an external Power Supply in 12V.

 

You have now tried to supply the 8 PIN CPU connection with another LPS. That does not work. It has to come from the same LPS that is used to supply the Taiko DC-ATX converter.

 

In my case I am using Keces P28. 24V / 4A go to the Taiko DC-ATX converter and 2 x 8 PIN CPU are supplied directly from the same Keces P28 via 12V / 10A and 12V / 6A.

[StreamFidelity]

1 hour ago, Patatorz said:

What you mean is that to have the 12v/8pin running, it should come from the same PSU and not from 2 separate : right ?

Yes

 

1 hour ago, Patatorz said:

How is it possible as i think i read that some people are feeding the 12V/8pins with another PSU than for the Taiko DC/ATX ?

Unlikely. It never worked for me with ASUS boards.

[StreamFidelity]

17 hours ago, Zauurx said:

Intel Core i9-11900F

 

Intel's 11th generation processor has received a few bad reviews. Essentially, it was about the poor energy efficiency. Not good for a passive cooling system.

 

When it comes to cost, go for an i9-10900 or i9-10850. When it comes to maximum performance, choose a 12th-core.

[StreamFidelity]

2 hours ago, Louie said:

Thus, I am looking for upscale to DSD512, which CPU that you will suggest?

 

In my opinion, it is always better to have more power reserves. The i9-12900 enables DSD512 / ASDM7ECv2. I'm not sure if a weaker CPU can be used.

 

2 hours ago, Louie said:

Also, do you have idea on which Z690 mainboard that have better SQ? It seem you are using MSI, how is your feeling on SQ?

 

msi was a positive surprise to ASUS on the part of BIOS. It runs very smoothly and enables, for example, the fan monitoring to be switched off at the push of a button. The direct phase design is also impressive. ASUS has a hybrid model.

 

At the moment I prefer DDR4 RAM due to the low latency and the selection of Z690 boards is limited. With DDR5 I prefer to wait and see how it develops.

[StreamFidelity]

1 hour ago, Patatorz said:

Today i use 3 rollerblocks from symposium but your feedbacks/recommendation  could be very useful before i start to put in place the Motherboard.

 

I prefer fis Magic Spacers. Unfortunately they are currently sold out. Decoupling sensitive electronic components is always a good thing.

 

[StreamFidelity]

First I would like to point out that I am a distributor of JCAT products. However, I do not recommend products that I am not convinced of. This to the classification of the following report!

 

As announced by Marcin, JCAT has developed a new linear ATX power supply: JCAT OPTIMO S ATX.

The JCAT OPTIMO S ATX LPS arrived yesterday. It was packed safe for transport in a cardboard box and shockproof.

 

 

This power supply is heavy (27kg) and very large. But it fits well on my Solidsteel rack.

 

 

The cables Molex 24 pin to Molex 24 pin ATX and Molex 8 pin to Molex 8 pin CPU are included and are really heavy. This is due to the cable material made of shielded silver-plated copper with 18 AWG.

 

 

I connected the cable ends directly to the fis audio PC. I was able to expand the Taiko DC-ATX converter because it is no longer needed.

 

 

My first test. Oversample PCM to DSD1024? No problem.

 

 

Load high-quality filters poly-sinc-ext3, ASDM7ECv2, DSD256 of the HQPlayer? The power is made available without restriction.

 

At higher loads, the LPS doesn't know limits but the temperature development of the CPU. In this setting the temperature fluctuates between 85°C and 95°C.

 

 

My favorite setting for CD format is poly-sinc-lp-2s, ASDM7ECv2, DSD256. With an average load of around 7%, this is also a minor issue for the Intel i9-12900K CPU. Incidentally, I deactivated 6 of the 8 E-Cores (efficient cores) in the BIOS, since a maximum of only 2 E-Cores were lightly loaded in all test scenarios. That saves energy.

 

 

How does that sound?

 

When something is reproduced particularly well, the volume perceives to decrease. This is because the background noise is pushed further down, i.e. the so-called blackness increases. The LPS has a ripple noise of only <25uV RMS (0.025 millivolt [mV]). Added to this is the current strength of 23A on the 12V rail. That is full power reserves.

 

The individual instruments are presented physically to touch. With this LPS, fine structures are revealed even in complex compositions, so that it is a pleasure. The timing of percussion instruments is springy and on point. One looks in vain for any sharpening.

 

The JCAT OPTIMO S ATX LPS is a game changer for me. Others probably suspected that too, because the first batch has already sold out. The next one won't come until March 2022.

 

 

 

 

[StreamFidelity]

58 minutes ago, RickyV said:

How much burn-in time did it get? 

 

At the time of my report, the JCAT OPTIMO S ATX was playing for about 5 hours! I usually burn in devices first, but here the improvement was so obvious.

 

58 minutes ago, RickyV said:

Did you already noticed some differences between the Keces-> Taiko supply vs the Jcat or is that to early to tell?

 

The differences are pretty big. I am trying to make a subjective assessment. I've been through a few iterrations. My summary:

 

HDPLEX 200W + 400W HDEPLEX DC-ATX

 

It was significantly better than a switching power supply. Unfortunately, this LPS went up in smoke because the load became too great. I haven't tried the successor HDPLEX 300W.

 

1 x Keces P8 + 400W HDEPLEX DC-ATX

 

That was a step forward in terms of security.

 

2 x Keces P8 + 400W HDEPLEX DC-ATX

 

The direct CPU supply was an important step towards increasing the SQ.

 

2 x Keces P8 + 500W Taiko Audio DC-ATX

 

I tried the full ATX supply via Taiko Audio. But the direct supply of the CPU with a second Keces P8 was the winner again. There has been some discussion about it here. I think 1 Keces P8 with a maximum of 20V / 8A is simply too weak to drive a Taiko Audio DC-ATX properly. It is designed for an unregulated LPS, which I have never tried for safety reasons. In any case, the combination was another step forward in my system.

 

1 x Keces P28 + 500W Taiko Audio DC-ATX

 

This power supply unit has several separate outputs but only one transformer. It got very hot and in the end I was "only" able to play DSD128 with ASDM7ECv2. Nevertheless, the sound was fantastic and the best in my system to date.

 

500W JCAT OPTIMO S ATX

 

It's amazing how the SQ has increased again. Now it has been playing for 24 hours and is getting better and better! In addition to the excellent quality of the power supply, there are also many arguments in favor of the technical principle.

 

3 independent rails for 3.3V, 5V and 12V supply the audio PC. Each of these rails has its own transformer, which reduces the power loss. This is noticeable in a significantly lower heat development, which is comparable to my mono power amplifiers T + A M10. Anyone who sees the cooling fins of my power amplifiers can imagine what I mean.

 

There are also fewer cables because there is no DC-ATX converter. The cable from JCAT are pretty good. That puts the price of the JCAT OPTIMO S ATX into perspective.

 

I am really very enthusiastic, because this fabulous transparency and instrument separation is excellent. 👍

[StreamFidelity]

Since it is possible to stream via Qobuz with the HQPlayer, I am of course also concerned with the question of what is better:

 

HQPlayer with or without Roon?

 

Roon on its own does not have a good reputation among many when it comes to SQ (sound quality). In addition to rather simple DSP algorithms, Roon, as a data octopus, is constantly busy feeding the music library with information. Roon seems to put more emphasis on the visual representation than on the rendering quality.

 

I really appreciate the user interface and diversity of Roon. That's why I previously combined Roon for music management and presentation with the high-quality algorithms of the HQPlayer. The separation only became really good when I started using a 2 PC system. The Audio Server with the Roon Core and the Audio PC with the HQPlayer.

 

HQPlayer without Roon

 

The user interface of the HQPlayer has always taken getting used to. You just have to know that the library has to choosen on the left (description diagonally) under "Album view". There the albums or playlists can be selected from the NAS or newly from Qobuz. The "Transport view" is displayed on the right. The usefulness of this representation is still not apparent to me today. I haven't figured out how to turn on random playback until today. Or how I can like or dislike titles. I close the lid of my laptop and the playback end, why? The HQPlayer Core is somewhere else entirely. So much for the negative things.

 

The streaming via Qobuz is going well and a lot (too much?) Of meta information is displayed. Even the album covers of Qobuz are loaded.

 

 

HQPlayer with Roon

 

Here Roon presents a tidy surface. For good reason, DSP is deactivated in Roon and delivers the data natively to the HQPlayer. At Roon, I really appreciate the fact that there is no difference between the albums on the NAS and on Qobuz. I don't have to switch back and forth.

 

I can name my tags any way I want and if I don't like a title in an album, this title will be disliked. This title is no longer selected when playing the album. Very pleasant. Roon Radio just continues to play suitable tracks if I haven't reloaded an album. That always brought me new, interesting albums that I would not otherwise have come across.

 

 

Which sounds better?

 

It could very well have been that HQPlayer sounds better with Roon. Because if you are streaming via the HQPlayer alone, the metadata and album art must be managed by the HQPlayer. So the HQPlayer has more to do.

 

The good news: I don't hear any difference between the two configurations.

 

Anyone who has a Roon lifetime license like me can continue to use the ingenious user interface without impairing the sound. Provided the Roon Core is running on another PC.

 

Anyone who is annoyed about Roon or only has an annual subscription can consider using the HQPlayer alone. The user interface takes some getting used to, but it performs well.

 

DSD512 with ASDM7ECv2

 

Incidentally, I carried out the test with a very high quality, which the JCAT OPTIMO S ATX power supply and the fis Audio PC are very demanding. The CPU temperature averages around 90°C.

 

The right side wall of the HDPLEX H5 v3 Chassis gets very hot, which is positive. Because the heat is dissipated very well.

 

The JCAT OPTIMO S ATX power supply is a little warmer than hand warm. 👍

[StreamFidelity]

55 minutes ago, The Computer Audiophile said:

Newer or faster doesn’t equal better.

 

This is true. Another example: I am currently not using DDR5 but DDR4 for the Z690 boards. DDR4 is more mature and definitely has significantly lower latencies combined with lower power consumption.

 

1 hour ago, Patatorz said:

Example : no ECC ram, quite old chipset,…

 

ECC leads to an increase in latencies.This is technically unavoidable because of the error check. The question is which is better: fewer errors or faster processing? Fewer errors can maybe achieved with high-quality DDR4 RAMs, just my opinion.

[StreamFidelity]

1 hour ago, Patatorz said:

You mean increases in latencies (if you are looking to decrease latencies).

 

But you read quickly. I corrected that in seconds. You have a very quick error correction. 👍

 

1 hour ago, Patatorz said:

how do you recognize high quality DDR4 ?

 

A good question. In Germany and Austria there is a very good price finder that allows some parameters: Geizhals 

 

Main memory (RAM) "optimal criteria for audio PCs":
DDR4,
Package size up to 16GB,
Single module size: 8GB,
CL latency up to 8.8ns,
Maximum voltage: 1.35V,
Lighting: none,
Passic Cooling: heat spreader

 

Then I like to look for low latencies and a well-known manufacturer. After the selection, an intensive RAM test is important. If this turns out like this, even the best reported latencies are of no use.

 

 

Because then an audiophile reproduction can become impossible.

 

 

If the RAM test looks like this, everything is fine.

 

[StreamFidelity]

On 12/16/2021 at 10:49 AM, StreamFidelity said:

The JCAT OPTIMO S ATX LPS is a game changer for me. 

 

Today I can justifiably say that the JCAT OPTIMO S ATX LPS is burned in. So after about two weeks or about 100 hours.

 

Oops, isn't that exactly what JCAT gave me? I swear I haven't thought about it until now. I only came up with it by doing my own calculations. 😃

1 hour ago, Mellowman said:

Especially since JCAT recommends between 100 and 200 hours of use to really benefit from it.

 

The reason was that the music flatters me today with a purity and clarity that I have never heard before. This is a very good end to the year. 👍

[StreamFidelity]

On 12/31/2021 at 6:38 AM, Asimov said:

Have you compared HQPlayer (upsampled in DSD) to Foobar2K in the native sample rate?

No

 

On 12/31/2021 at 6:38 AM, Asimov said:

Could you please tell me what I am missing here?

Remote diagnoses are difficult. The filters and modulators of the HQPlayer are very diverse. If you like PCM better than DSD - why not? But it is a pity that Denafrips has a very good design for DSD. 

 

https://www.denafrips.com/specs-pontus

DSD	DSD64 (DoP) on Coax/AES/Opt Input
	DSD1024* On USB & I²S Input Only
PCM	24bits / 44.1, 48, 88.2, 96, 176.4, 192KHz On All Input
	1536kHz* On USB & I²S Input

 

I would experiment a little bit. Leave out the Gustard and go with USB into the DAC. Then you can do PCM oversampling up to 1.536kHz. Listen to DSD again with the new ASDM7ECv2 modulators. 

[StreamFidelity]

With this setting, the HQPlayer cannot sound good. The digital volume control is red. You have to set at least -3dB to avoid clipping. I use -4dB.

 

7 minutes ago, Asimov said:

But I am curious why you are not using any DDC!

 

The T+A SDV 3100 HV has Signalyst NAA (Network Audio Adapter) on board. This is the best connection I have had so far.

[StreamFidelity]

42 minutes ago, Asimov said:

BTW, have you tried any DDC with your Terminator Plus DAC?

 

No. That would certainly have been an interesting experience with the GAIA DDC. I know that Denafrips recommends I²S. 

 

Source: https://www.denafrips.com/terminator-plus

 

My sound experience with JCAT USB XE was great, so I never felt the need.

[StreamFidelity]

2 hours ago, Cazzesman said:

@StreamFidelityI have slowly been working my way through the 146 pages.    There is a massive amount of talk about lowering the heat within the HDPlex H5 cases, with the higher powered CPU's  - TDP 105 and above.

 

I love the look of your Perspex lids.   I can't see anywhere if you have done any testing of temps from your personal Server both with the standard H5 lid on and with the 4 Hole Perspex Lid.

 

Any stats available please?

 

Regards Cazzesman

 

I have copied this question to my thread.  

 

With the old H5 V2 chassis, the CPU temperature difference between an open chassis and with acrylic lid was only 2-3°C (depending on operation and installed components). 

 

The new H5 V3 chassis cools the CPU much better. There is also now a solution for the Solarflare NICs via heatpipes. With that in mind, the optics (whoever likes it) outweighs the cooling effect. 😉

 

I think the acrylic lid will still cool a few degrees better because the air is exhausted directly above the CPU and PCIe cards. For hardcore DSD conversion it maybe make the difference. But I haven't done any measurements. 

 

 

[StreamFidelity]

10 hours ago, Cazzesman said:

When you say 'open chassis' I take it you mean with the metal lid completely off the H5.

Yes

 

11 hours ago, Cazzesman said:

If that is the case do you have data on both the approx temps with the metal lid on and with the lid off?

Not yet. Maybe I'll run a series of tests at some point. At the moment, I don't have the time.

[StreamFidelity]

16 hours ago, deanorthk said:

I have found the Asus formula, but no 9900K

Why not go straight to the latest 12th processor generation? These CPUs and Z690 motherboards are much more powerful. 

[StreamFidelity]

1 hour ago, ASRMichael said:

Hi any idea on sound quality difference with new Z690?

 

The difference between the 9th, 10th and 11th generation is not very big. The 11th generation was pretty big crap in terms of energy efficiency. 

 

Only with the 12th generation did some technical innovations have a very positive effect. There are P-cores and E-cores, which allow a better distribution of the workload. 

 

Intel 11th generation CPUs only used one IMC (Integrated Memory Controller). The 12th generation has two IMCs. This enables a higher data throughput due to parallel data processing. Even with DDR4 RAMs.

 

Source: https://www.igorslab.de/ddr5-deep-di...-der-praxis/2/

 

I now use a Motherboard MSI MPG-Z690-EDGE-WIFI-DDR4 with Z690 chip and direct 75W phase design. The BIOS is performant and the MB is very stable. 

 

Maybe not so important for low power systems. For high power systems, for example for the HQPlayer with DSD1024, a blessing. It sounds so organic and airy.

 

[StreamFidelity]

8 minutes ago, deanorthk said:

but just bought an old XVIIIish century farm in France, and I have kinda other priority:) 

No Audio PC in the world can compete with a farm in France. Enjoy it. 👍

[StreamFidelity]

For all those who are a bit annoyed by the subscription model of Euphony or Roon, there is a free alternative: HQPlayer OS. Of course, a valid HQPlayer Embedded license must be present. But it can be tested free of charge before (every 30min. a restart is required).

 

I wrote something about it here: Use HQPlayer as operating system and as standalone player.

[StreamFidelity]

Just in time for the sales launch on October 20, 2022, there are now a lot of reviews. Since I look to high computing performance, the flagship

 

Intel® Core™ i9-13900K(F) processor

 

is of increased interest for me.

 

Test results

 

My three german favorite sites have published interesting tests also and especially in comparison to the new Ryzen-7000 series and of course to the older Intel generations.

 

igorsLAB: Intel Core i9-13900K and Core i5-13600K in test - showdown of the 13th generation and a 3/4 crown for the last monolith

Quote

Now everyone will probably ask how the three-quarter crown in the title is actually meant. From a purely statistical point of view, it is a clear victory of the Core i9-13900K against the Ryzen 9 7950X in gaming, although life doesn't only consist of pure gaming. The Core i9-13900K often wins in the workstation and creation field, but not always. And even if it is even a bit more efficient than AMD's opponent Ryzen 9 7950X during gaming, it is still the favorite of every power supplier under full load during compute or rendering. Yes, it has truly earned the gaming crown, although the lead of up to 4 percentage points is not so huge that you could call it a landslide.

HardwareLUXX: The Intel dinosaur is not yet extinct - Core i9-13900K and Core i5-13600K in review

Quote

Positive aspects of the Intel Core i9-13900K:
- high single and multi-threaded performance
- fastest gaming processor
- good TDP scaling with power saving potential
Negative aspects of the Intel Core i9-13900K:
- high power consumption under full load
- high temperatures under full load

 

ComputerBase: Intel Raptor Lake in review: Core i9-13900K, i7-13700K & i5-13600K chase Ryzen 7000

Quote

AMD had presented with Ryzen 7000 (review) and delivered strong performance with four processors, but Intel's reaction is no less strong this year. In multi-core applications, AMD can just about save the plus at the upper performance end with the strong Zen 4 cores, but the Ryzen 5 7600X doesn't have enough of it compared to the Core i5-13600K. In terms of efficiency, AMD presents itself better positioned in applications with Ryzen 7000 both ex-factory and after manual adjustment, but in the lower price segment with Ryzen 7 and Ryzen 5, this duel is also close. On an extremely high performance level, Ryzen 7000 has "no chance" against Raptor Lake in games. No wonder that AMD is rumored to follow up with the Ryzen 7000X3D as early as in spring 2023.

 

The tests try to come to a meaningful result via synthetic benchmarks and application scenarios. And so IgorsLAB awards Intel a 3/4 crown. Thus, Intel scores very well in the majority of cases. However, if AMD performs better than Intel in the application scenarios required by the user, then even a 3/4 crown is of no use. For example, AMD can predominantly score with a better computing performance (3D rendering) under full multi-core utilization and with a better energy efficiency under full load.

I therefore like to take a particularly close look at the test sections that are interesting for me.

 

Energy efficiency 

 

For a fanless passively cooled audio PC with high computing power, energy efficiency is of outstanding importance. High power consumption is always accompanied by high temperature development. Here, the 13th Intel Core generation has a slight advantage with a TJUNCTION of 100°C compared to 95°C for Ryzen 7. Of course, nobody wants to see the maximum temperature, but a higher stress resistance is always good.

 

As already written above, AMD basically performs better than Intel under full load, but that is not my use case. My typical playback of a CD at 44.1kHz involves upsampling to DSD512 using the best modulator currently available, ASDM7ECv2 from HQPlayer. As can be seen in the narrow frame at the bottom of the picture, there is just a load of 15% across all cores. An Intel® Core™ i9-12900K processor is used. Hyper-Threading is enabled, so the P-cores double to 16. Heavily loaded are two cores (no. 1 and 9) that go to 4.4GHz clock frequency. These are the cores for the performance-hungry modulators. There are 8 E-cores, but only 2 are activated, because more are not needed in this application. The power consumption is around 60W, which is much less than the stated TDP of >125W.

 

 

The exciting question for me is whether there is the efficiency leap with low consumption announced by Intel? According to Intel, the single-thread performance should increase by up to 15 percent and the multi-core performance by up to 41 percent. This value refers to a power consumption of 253 watts, too much for our fanless system. But at a power consumption of 65 watts, the i9-13900K is supposed to be as performant as the i9-12900K at 241 watts!

 

Source: https://www.pcwelt.de/news/Core-i9-13900K-Intel-stellt-schnellste-Desktop-CPU-der-Welt-vor-11301464.html

 

Of course, that would be sensational, since the 65W is exactly in the range of my use case. And ComputerBase has worked that out nicely. The i9-12900K is used as a reference at 241W with 100% power.  The approximately same performance with 96% can be accessed with the i9-13900K with only 65W power consumption.
 

Core	65 Watt	88 Watt	125 Watt	241 Watt	253 Watt	no Limit
Core i9-12900K (8+8)	71%	81%	93%	100%	-	103%
Core i9-13900K (8+16)	96%	111%	126%	-	151%	153%

 

 

Summary

 

Intel has not lied about energy efficiency. The range around 60W, which is important for the fis Audio PC, can be used with an i9-13900K with a significantly higher performance than before. Now, of course, our own tests have to follow. Keep at it. 

 

[StreamFidelity]

4 hours ago, Gavin1977 said:

It would be good to compare this against 10980/10900 HEDT processors which a lot of people use but are difficult thermally

 

The Intel Cascade Lake-X (Core X 10000) HEDT (high-end desktop processors) was a continuation of the competition with AMD Threadripper 3000, which Intel had lost. Look here.

 

In a benchmark comparison, an Intel Core i9-10980XE performs worse than an Intel Core i9-13900KF with -48.3%. By the way, the K version is -11.2% worse than the KF version (without GPU) according to this comparison. That was not yet clear to me. I have ordered an Intel Core 9-13900KF. 👍

 

2 hours ago, Gavin1977 said:

And DDR5 seems to have poorer latency compared to DDR4.

 

DDR5 indeed had to struggle with very bad latencies in the beginning. Meanwhile, the CAS latency CL is often at a low 10.00ns. E.g. G.Skill Trident Z5 NEO black DIMM Kit 32GB, DDR5-5600, CL28-34-34-89.

 

In addition, there is the on-die ECC (error correction).The bandwidth has always been better (about 3x faster). I only use DDR5 now.