Jump to content


Photo

RAM adventures - Extremely interesting results


  • Please log in to reply
28 replies to this topic

#1 pavgeris

pavgeris

    Beginner

  • Members
  • 44 posts

Posted 07 February 2019 - 06:08 PM

Hi guys.

 

Today I decided to play a little with my RAM configuration. I have always been reading here and there about the do's, the don'ts, etc. So, I decided to reach to the final results by myself and take decisions for my final setup.

 

To make the long story short, the results were shocking. It seems that the configuration of RAM plays a major role in the system. If you have the patience and willingness to play around, I am sure that you will discover equally impressive results.

 

My system consists of the following:

  • Dual PC configuration
  • 1st PC (Control PC):
    Motherboard: Supermicro X10SAE
    Memory: DDR3 ECC unregistered
    CPU: Intel Xeon E3-1240L v3
    Onboard Ethernet controllers: Intel I210 & I217
    Operating system: Windows 2019 core with Audiophile Optimizer 3 beta 30
    Music server: ROON server
    Power supply: Seasonic Platinum 400 W fanless
  • 2nd PC (Audio PC) not something extraordinary
    Motherboard Asrock X370 Taichi
    RAM: 16Gbytes
    CPU: AMD Ryzen x1700
    Operating system: Windows 10 1809
    Music rendered: HQPlayer
    Power supply: Corsair 450W (nothing special)
  • NAS music storage:
    Motherboard: Supermicro X9SRH-7TF
    CPU: 10-core Xeon series 2600 (2680 if I am right)
    Memory: Hynix 8 x 16Gb ECC registered, 128Gb in total
    Operating system: FreeNAS ver 11.1
    HDDs: 12 x 8TBytes Western Digital Red 5400 rpm
    Ethernet adapter: Intel X540 10GbE

My adventures in Ram were performed on the control PC. I tried 1 & 2 Gb RAM chips in various configurations.

  1. Config A: 2 x 1 Gb ECC UDIMMs (dual memory channel), 2 Gbytes in total
  2. Config B: 1 x 2 Gb ECC UDIMM (single memory channel), 2 Gbytes in total
  3. Config C: 2 x 2 Gb ECC UDIMMs (dual memory channel), 4 Gbytes in total
  4. Config D: 4 x 2 Gb ECC UDIMMs (dual memory channel), 8 Gbytes in total

I was listening in A for almost 2 days. Today I switched to B. The total amount of memory was the same (2Gb), however from dual memory channel I went to single memory channel. The end result was pretty close to that of Config A but lacked the magic. Depth of field, distinctive layers of depth were less defined, music lacked just a bit of the intimacy and thrill.

 

Then it was time to go to Config C. It is 4 Gbytes consisting of 2 RAM chips. Dual memory channel. From those who are playing with DDR3 memory, I think that the vast majority are playing with this RAM configuration.
Guys, I was not prepared for this shock! Music lacked almost every aspect that makes us to want to listen more and more. Layers of depth were diminished, music became totally uninvolving, voices were messed up, high freq extension was drowned in mud! I would never expect that this combo of 2x2Gb would have such a devastating effect in music! I really pressurized myself to sit down and listen critically. I could not! The result was absolutely unbearable. Intolerable! By far the worse sound ever came out of this fantastic board!

 

Something into my mind told me that I should try to cover all the 4 memory banks. Of course same type of chips. I have read around that increasing the total quantity of RAM, from 4Gb to 8Gb, leads to worse results. ON THE CONTRARY! Music now became much better in all aspects. Flow, energy, dynamics, depth portrayal, everything was back again! The difference was dramatic. The end result was not up to the level of Config A but very slightly recessed. Music did not have the perfection of Config A but it was magnitudes of order better in everything if compared to the other Configurations. The result was not only good and tolerable but very good. It was probably the best result among others, if you have not of course experienced Config A. A has only 2 Gb where D has 8 Gb! However, even the total amount of RAM is 4x, the final result is fantastic and slightly recessed in everything, like dynamics and theatrical performance, but in total very good!

 

Going back to Config A sent music to the initial levels. Everything was there but more evident. Depth of field better defined, air, HF extension, thrill, excitement, everything was back. More defined. Better.

 

What I found out in my setup was rather straightforward. It is not that devastating to go from 2 Gb to 8 Gb as devastating is not to use all the memory banks of your motherboard! It seems that covering 4 banks instead of 2 leads to a SHOCKING and DRAMATIC result. Far better! Basically, music is almost totally vanished in Config C. Unfortunately I did not have 4 chips of 1Gb each to fill the remaining banks. From what I have experienced, going from 4 to 8 Gb by filling all the four banks, had the most dramatic effect, although going higher in RAM usually deteriorates sound quality. It seems that the CPU would prefer to 'see' all the banks covered, in dual memory channel set up.

 

If you have a DDR3 MoBo, I would strongly suggest to buy 4 pieces of 1Gb each and cover all the RAM banks. At least this seemed to be the best configuration i  my system. I assume that, what I would lose from going from 2Gb to 4Gb, would give me back the benefit of filling all the memory banks and maybe in the end this would be the best configuration.

 

Stay tuned!


  • Adrian, motberg, FelipeRolim and 3 others like this

#2 Marcin_gps

Marcin_gps

    JPLAY & JCAT Founder

  • Administrators
  • 4,395 posts

Posted 08 February 2019 - 12:54 PM

Thanks for the report! Too bad you didn't have 2x 4GB or 4x 4GB modules :) 


  • FelipeRolim likes this

JPLAY for iOS: the new reference hi-fi app. 


JCAT: streaming specific products for high-end audio
 
Join Computer Audio Enthusiasts group on facebook

#3 pavgeris

pavgeris

    Beginner

  • Members
  • 44 posts

Posted 08 February 2019 - 03:57 PM

Thanks for the report! Too bad you didn't have 2x 4GB or 4x 4GB modules :)

I am expecting quite a lot of RAM chips. I will be having another 2x1Gb so as to be able to use 4x1Gb in total.

I am also waiting to receive 4x4Gb in order to be used in my RAMdisk configuration.

I will report everything of course!

 

What I have found is that we should use ALL the available banks on our MoBos. For ordinary installations with SSD, I assume that 4x 1Gb will be the best combo, while for RAMDISK the best would be 4x4.

 

So in the end I will have tested:

  • 2x 1Gb
  • 4x 1Gb
  • 1x 2Gb
  • 2x 2Gb
  • 4x 2Gb
  • 1x 4Gb
  • 2x 4Gb
  • 4x 4Gb

I am expecting 4x 1Gb to surpass them all. Memory in all cases will be ECC unregistered.


  • Adrian and FelipeRolim like this

#4 Jogar

Jogar

    Advanced

  • Members
  • PipPip
  • 60 posts
  • LocationFrance

Posted 08 February 2019 - 04:58 PM

Thank you for this very interesting report. The improved SQ with 2Gb dual channel compared to 2Gb single channel or occupying more memory slots is unexpected. Any thoughts what might be going on here? Have you compared industrial with standard unregistered ECC RAM? Your scientific approach to this is pretty cool.



#5 FelipeRolim

FelipeRolim

    Die Hard

  • Members
  • PipPipPip
  • 323 posts
  • LocationQuatiguá/Paraná/Brazil

Posted 09 February 2019 - 12:38 PM

Hello Pavgeris. Very interesting your impressions. Thank you for sharing them. I commented about this subject on this post, but I didn't extend on the subject. In my experience, 16GB of RAM deliver better audio quality than using 8GB of the same RAM, but I've never tested compare dual-dimm vs. single-dimm. I also don't know the reasons why, even because, with or without RAMdisk, Windows doesn't require more than 600 or 700MB of RAM and JPLAY also only takes up minimal space. Even so, 16GB of RAM always play better than 8GB of RAM in my system. In the topic I started about the requirements of a good computer for audio, I always made clear that, in my sense, the better the hardware, the more powerful it is, the more it will be able to deliver playback quality, even if the use is minimum. I believe this also applies to RAM.

 

Once again, thank you for sharing. With your permission, I'd like to cite the link to your topic in the one I created. And forgive me the disappearance. I'm running out of time, but I'm still exploring the possibilities for RAMdisk.


  • pavgeris likes this

- Speakers KEF Reference 203/2 with spikes;

- Schiit Freya preamp. (with Electro-Harmonix 6SN7 Gold-Pin tubes and Synergistic Research Orange Fuse) and Audio Research VS115 (with Tung-Sol KT120 and Svetlana 6H30 tubes and AMR Gold Fuse);

- DAC Ayre QB-9 192/24 + Furutech FI-03 ( R) + Synergistic Research Orange Fuse;

- Digital transport: the Control-PC is a dedicated computer with a Gigabyte H310N motherboard, a Intel Core i3-8100T processor, an Arctic Alpine 12 Passive Cooler, 16GB Corsair Vengeance DDR4 RAM 2400MHz, a JCAT Net Card FEMTO and 1TB Samsung 850 Pro SSD; the Audio-PC is a dedicated computer with a Gigabyte H310M M.2 2.0 motherboard, a Intel Core i3-9100T processor, an Arctic Alpine 12 Passive Cooler, 16GB Corsair Vengeance DDR4 RAM 2400MHz, a JCAT Net Card FEMTO and a JCAT USB Card FEMTO. Both computers use Windows Server 2019 Datacenter Core Mode (17763.737) with RAMdisk as Operational System, JPLAY FEMTO Alternative version, Minority Clean and Audiophile Optimizer v. 3.00 (1A in Control-PC and 4D in Audio-PC). Both also use hand made linear power supplies with a Furutech FI-06 ( R) NCF.

- Cables: Nordost Tyr (speaker), Nordost Valhalla XLR and Siltech 770i XLR (interconnects), Curious Evolved USB Cable, Nordost Valhalla Power Cable (DAC), Transparent PowerLink MM2 (power amp.), 2 x Purist Audio Design Canorus Praesto Revision Power Cable (power distributor and computers), Purist Audio Design Limited Edition Praesto Revision Power Cable (preamp.) and Jupiter Pure Silver Cotton Insulated (all internal wires from the computers).

- Acessories: Furutech GTX-D (G); Furutech GTX-D ( R) (x2); Sunrise Deep Line (x3); ebony spikes, carbon fiber bases; Nordost Qv2 (x3); acoustical and electrical reinforcements.


#6 pavgeris

pavgeris

    Beginner

  • Members
  • 44 posts

Posted 13 February 2019 - 06:25 PM

Thank you for this very interesting report. The improved SQ with 2Gb dual channel compared to 2Gb single channel or occupying more memory slots is unexpected. Any thoughts what might be going on here? Have you compared industrial with standard unregistered ECC RAM? Your scientific approach to this is pretty cool.

 

I have no non-ECC memory. Only ECC memory! So I can't verify!

A friend has listened to the same variant and found that ECC plays definitely better.


  • Marcin_gps likes this

#7 pavgeris

pavgeris

    Beginner

  • Members
  • 44 posts

Posted 13 February 2019 - 06:27 PM

Once again, thank you for sharing. With your permission, I'd like to cite the link to your topic in the one I created. And forgive me the disappearance. I'm running out of time, but I'm still exploring the possibilities for RAMdisk.

 

You should not ask. You should have done it already! :-)

I have also been disappeared due to lack of time!



#8 Jogar

Jogar

    Advanced

  • Members
  • PipPip
  • 60 posts
  • LocationFrance

Posted 13 February 2019 - 11:26 PM

"I have no non-ECC memory. Only ECC memory! So I can't verify!

A friend has listened to the same variant and found that ECC plays definitely better."

 

Hi Pavgeris, perhaps my question was not clear: I find a slight but significant improvement in SQ with more expensive industrial ECC RAM (e.g. APACER DDR3 SODIMM unregistered ECC) compared to standard outlet (e.g. Crucial, Kingston) SODIMM unregistered  ECC. The improvement in SQ is such that I cannot go back to using non-industrial ECC memory. Because of your really good approach to finding answers about SQ and RAM slots, I just wondered whether you had given consideration to the intrinsic quality of the ECC RAM itself. I'd be interested to know whether Felipe's looked at this too. Regards.


  • Marcin_gps likes this

#9 pavgeris

pavgeris

    Beginner

  • Members
  • 44 posts

Posted 14 February 2019 - 11:37 AM

I have istened to Apacer DDR3 industrial ECC memory. Extensively.

I also have Hynix ECC DDR3 memory. Same config.

The Hynix plays considerably better than Apacer. But it is not possible to judge different brands, different production lots, etc.


  • Jogar likes this

#10 pavgeris

pavgeris

    Beginner

  • Members
  • 44 posts

Posted 16 April 2019 - 05:28 PM

Hello Guys!

 

Since my last post, I have received my 4Gbytes RAM chips in ECC mode and have finished my evaluations.

I attach a picture of the chips I have available.

Hynix 1Gb

Apacer 2Gb

Hynix 4Gb

 

Best of all configurations sounded the Hynix 1x4Gb installed on the audio PC. By far the best results.

The next was Hynix 4x 1Gb, somewhat worse in SQ but fine.

The worst of all was 2x 2Gb Apacer chips. I have received a lot of recommendations for Apacer ECC RAM (specially extended temp range) but the difference to those Hynix chips is really enormous!

 

I have always been a fan of Hynix chips, they had made a tremendous result inside my NAS already.

 

If you want the best of all, go for DDR3 Hynix ECC memory. You won't regret it!

DSC_0316_zpsehtsnhtf.jpg


  • Adrian, FelipeRolim, Jogar and 1 other like this

#11 Nobudy

Nobudy

    Die Hard

  • Members
  • PipPipPip
  • 285 posts

Posted 16 April 2019 - 05:34 PM

agrees that Hynix sounds wonderful
have used them for years with good sound


  • Adrian, FelipeRolim and pavgeris like this

#12 TJHUB

TJHUB

    Die Hard

  • Members
  • PipPipPip
  • 203 posts

Posted 25 April 2019 - 12:42 AM

I decided to try Hynix ECC RAM in my single PC setup. I purchased a single 4MB stick to replace my dual 2MB sticks of Corsair. I agree the Hynix RAM has improved clarity and increased definition. It is a little leaner sounding, but that works for my setup. It's not a huge difference in sound, but very noticeable. I'm happy with the purchase.
  • Marcin_gps likes this
JPLAY Femto Setup: Single PC (Intel S1200KPR/XEON based), Windows Server 2016 GUI, JPLAY Femto, UltraStream engine (DAC Link = 500Hz, KS, and Hibernate), Motherboard USB port to DAC. HD-Plex linear power supply powering PC. Music library stored locally on OS 2.5” HDD (2TB Western Digital Blue, 5,400RPM). UPnP control point: Bubble UPnP or fidata.

#13 DC

DC

    Die Hard

  • Members
  • PipPipPip
  • 601 posts

Posted 25 April 2019 - 09:18 AM

Hi,

this memory seems to be very difficult to find in France, unfortunately.



#14 Nobudy

Nobudy

    Die Hard

  • Members
  • PipPipPip
  • 285 posts

Posted 25 April 2019 - 01:54 PM

ebay /europe only

stay away from asia

best from Germany or GB



#15 DC

DC

    Die Hard

  • Members
  • PipPipPip
  • 601 posts

Posted 25 April 2019 - 01:58 PM

Many thanks for the tip Nobudy.

 

Do you think this one could be equivalent.

 

https://www.amazon.f...sr=1-1-fkmrnull



#16 Nobudy

Nobudy

    Die Hard

  • Members
  • PipPipPip
  • 285 posts

Posted 25 April 2019 - 03:23 PM

try to find pure Hynix,

 

not rebranded



#17 Nobudy

Nobudy

    Die Hard

  • Members
  • PipPipPip
  • 285 posts

Posted 25 April 2019 - 03:31 PM

*



#18 Gordian

Gordian

    Newbie

  • Members
  • Pip
  • 1 posts

Posted 25 April 2019 - 06:06 PM

Hi all, 

 

Tried 2x4Gb vs 4x4Gb, I can hear no difference between the 2 configurations.

May try 2Gb or 1Gb if I manage to find Hynix ram


  • motberg likes this

#19 Jogar

Jogar

    Advanced

  • Members
  • PipPip
  • 60 posts
  • LocationFrance

Posted 26 April 2019 - 09:02 PM

Hi Guys,

I find this subject fascinating and many thanks for your contributions. According to Crucial, there are only three DRAM chip manufacturers (industrial wide temperature or commercial): Micron, Samsung and Hynix, see 

www.crucial.com/usa/en/store-truth-about-memory-manufacturers 

Presumably other manufacturers assemble DRAM chips without always specifying which company has actually manufactured the chips in their product.

For example, Advantech uses Hynix DRAM (but perhaps not in all their DRAM?) and Apacer gives no information about this.

If this is true, then what we could be doing when testing different RAM is comparing the respective sound qualities of the DRAM chips of any of the three manufacturers and even DRAM chips of the same manufacturer assembled by other brands. This issue obviously requires clarification to enable us to make meaningful comparisons.

After Pavgeris' and Nobudy's recommendations on Hynix DRAM (which I'd like to test in my system too) I also find it difficult to find Hynix outside Ali Express or Ebay.

RAM obviously has a major impact on sound quality which in my limited experience is influenced by the motherboard power supply and cables.

Finally has anyone tested the Elfidelity AXF-74 DDR3 or AXF-75 DDR4 noise filters?



#20 FelipeRolim

FelipeRolim

    Die Hard

  • Members
  • PipPipPip
  • 323 posts
  • LocationQuatiguá/Paraná/Brazil

Posted 26 April 2019 - 10:42 PM

About Corsair Vengeance DDR3 memories:

 

 

Attached Files


- Speakers KEF Reference 203/2 with spikes;

- Schiit Freya preamp. (with Electro-Harmonix 6SN7 Gold-Pin tubes and Synergistic Research Orange Fuse) and Audio Research VS115 (with Tung-Sol KT120 and Svetlana 6H30 tubes and AMR Gold Fuse);

- DAC Ayre QB-9 192/24 + Furutech FI-03 ( R) + Synergistic Research Orange Fuse;

- Digital transport: the Control-PC is a dedicated computer with a Gigabyte H310N motherboard, a Intel Core i3-8100T processor, an Arctic Alpine 12 Passive Cooler, 16GB Corsair Vengeance DDR4 RAM 2400MHz, a JCAT Net Card FEMTO and 1TB Samsung 850 Pro SSD; the Audio-PC is a dedicated computer with a Gigabyte H310M M.2 2.0 motherboard, a Intel Core i3-9100T processor, an Arctic Alpine 12 Passive Cooler, 16GB Corsair Vengeance DDR4 RAM 2400MHz, a JCAT Net Card FEMTO and a JCAT USB Card FEMTO. Both computers use Windows Server 2019 Datacenter Core Mode (17763.737) with RAMdisk as Operational System, JPLAY FEMTO Alternative version, Minority Clean and Audiophile Optimizer v. 3.00 (1A in Control-PC and 4D in Audio-PC). Both also use hand made linear power supplies with a Furutech FI-06 ( R) NCF.

- Cables: Nordost Tyr (speaker), Nordost Valhalla XLR and Siltech 770i XLR (interconnects), Curious Evolved USB Cable, Nordost Valhalla Power Cable (DAC), Transparent PowerLink MM2 (power amp.), 2 x Purist Audio Design Canorus Praesto Revision Power Cable (power distributor and computers), Purist Audio Design Limited Edition Praesto Revision Power Cable (preamp.) and Jupiter Pure Silver Cotton Insulated (all internal wires from the computers).

- Acessories: Furutech GTX-D (G); Furutech GTX-D ( R) (x2); Sunrise Deep Line (x3); ebony spikes, carbon fiber bases; Nordost Qv2 (x3); acoustical and electrical reinforcements.





0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users