ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
I’ve been trying to benchmark memory on my new build that I am hoping somebody will help me with. I have EVGA X299 Dark with i9-10900X and I am trying to populate 4 memory slots with Corsair 3000 MHz CL15 32GB DIMMs ( https://www.corsair.com/u.../p/CMK64GX4M2C3000C15) in each for a total of 128GB. I am using Corsair’s XMP Profile and memory is stable, passing all memory tests I have thrown at it. When I run PassMark PerformanceTest with just one slot populated I get Read Uncached 13965 and 8192 Write. With two slots populated I get Read Uncached 15248 and Write 13143. With three I get Read Uncached 15311 and Write 13748. With four I get Read Uncached 10153 and Write 15347. In other words going from single to dual gave me ~ +9% reads and +60% writes. Going from dual to triple gave me practically -no- difference. And going to quad resulted in ~ 12% improvement of writes over triple (+87% over single) _BUT_ HUGE DROP in reads, not just when compared to triple and double but even when compared to single. According to PassMark my quad channel reads are ~ 33% slower over single channel ones! I’ve checked am I having stick that is performing worse than others and all of them when put in same slot benchmark practically identically. I’ve checked am I having a slot that is performing worse than others and same stick when put in different slots benchmarks practically identically across all of them. So what is going on with my reads?! I would appreciate help, please, in figuring it out. I would also appreciate pointers in direction of other memory benchmarks so I can rule out PassMark as possible culprit. Hopefully something that is very detailed, thorough, configurable and with result that can be compared to real world specs rather than just single figure that has no common unit of measure.
|
Sajin
EVGA Forum Moderator
- Total Posts : 49227
- Reward points : 0
- Joined: 2010/06/07 21:11:51
- Location: Texas, USA.
- Status: offline
- Ribbons : 199
Re: Low Memory Benchmark Woes
2020/03/25 10:39:28
(permalink)
Try the aida64 memory benchmark.
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/25 10:48:16
(permalink)
Sajin
Try the aida64 memory benchmark.
Thank you, I have already tried Aida64 and unfortunately it seems to use cached reads lumping everything into single figure.
|
Sajin
EVGA Forum Moderator
- Total Posts : 49227
- Reward points : 0
- Joined: 2010/06/07 21:11:51
- Location: Texas, USA.
- Status: offline
- Ribbons : 199
Re: Low Memory Benchmark Woes
2020/03/25 11:15:11
(permalink)
Only other one I know would be maxxmem2.
|
Sajin
EVGA Forum Moderator
- Total Posts : 49227
- Reward points : 0
- Joined: 2010/06/07 21:11:51
- Location: Texas, USA.
- Status: offline
- Ribbons : 199
Re: Low Memory Benchmark Woes
2020/03/25 11:57:06
(permalink)
☄ Helpfulby ZoranC 2020/03/25 12:19:20
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/25 12:18:27
(permalink)
Sajin
Only other one I know would be maxxmem2.
Thank you! Already tried that one too and it too seems to use cached figures aggregated into single value :(
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/25 12:20:38
(permalink)
|
chinobino.
New Member
- Total Posts : 42
- Reward points : 0
- Joined: 2008/09/19 23:57:25
- Status: offline
- Ribbons : 6
Re: Low Memory Benchmark Woes
2020/03/25 13:31:48
(permalink)
Can you post some screenshots of your memory timings using ASRock Timing Configurator in single, dual and quad channel? It sounds like there is incorrect auto configuration when more DIMMs are added, which could be from poor RAM training or maybe BIOS is not well 'tuned' for 32GB DIMMs. Also, if you have some 8GB or 16GB DIMMs, maybe take some screenshots with those installed for comparison and post them here too.
post edited by chinobino. - 2020/03/25 13:35:19
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/25 13:45:02
(permalink)
chinobino.
Can you post some screenshots of your memory timings using ASRock Timing Configurator in single, dual and quad channel?
Thank you! I will do that as soon as I can. chinobino.
It sounds like there is incorrect auto configuration when more DIMMs are added, which could be from poor RAM training or maybe BIOS is not well 'tuned' for 32GB DIMMs.
That's exactly what I was thinking about last night, that XMP timings once I pass two could be resulting in it. chinobino.
Also, if you have some 8GB or 16GB DIMMs, maybe take some screenshots with those installed for comparison and post them here too.
Unfortunately I don't have any 8GB or 16GB DIMMs. If I was to purchase some which ones you would be recommending?
|
chinobino.
New Member
- Total Posts : 42
- Reward points : 0
- Joined: 2008/09/19 23:57:25
- Status: offline
- Ribbons : 6
Re: Low Memory Benchmark Woes
2020/03/25 13:51:13
(permalink)
It would be ideal to compare a 16GB x 4 DDR4-3000 CL15 kit with what you have but I don't expect you to buy it just for troubleshooting.
I was just thinking of ways to see the change in performance with different capacity DIMMs and thought you may have had access to another RAM kit.
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/25 14:39:09
(permalink)
chinobino.
It would be ideal to compare a 16GB x 4 DDR4-3000 CL15 kit with what you have but I don't expect you to buy it just for troubleshooting.
I was just thinking of ways to see the change in performance with different capacity DIMMs and thought you may have had access to another RAM kit.
I don't mind getting a second kit to have as a spare when needed. I was thinking it should be 4 sticks, otherwise it won't let me troubleshoot when I do need to troubleshoot. I was also thinking it should be 16 GB per stick because that would still give me 64 GB total which would be plenty enough to work with if/when I have to fallback onto spare from main kit. Rest of specs don't have to match what I have right now but I don't want to push my IMC too high so I don't introduce that variable into equation so I figure either 3000 or 32000, not more. With that said I was looking at this one: https://www.gskill.com/product/165/184/1536137188/F4-3200C14Q-64GVK-Overview but I can't find it in stock by a reputable seller anywhere.
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/25 15:49:13
(permalink)
OK, I have downloaded ASRock Timing Configurator but how do you use this thing???
|
chinobino.
New Member
- Total Posts : 42
- Reward points : 0
- Joined: 2008/09/19 23:57:25
- Status: offline
- Ribbons : 6
Re: Low Memory Benchmark Woes
2020/03/25 15:53:21
(permalink)
When troubleshooting you always want to minimize the number of variables that could be potentially causing a problem. That's why I recommend testing with a smaller capacity 3000 MHz CL15 kit, so you can compare how the board auto configures with XMP while only changing one main variable (i.e. DIMM density). For every possible memory speed the motherboard changes many timings and will have thresholds for specific memory clock ranges. In the old days you would sometimes hear about FSB or memory strap 'holes', where there would be a frequency range that was either not usable or extremely unstable due to the BIOS not having a workable configuration or being limited by the frequency intervals of the clockgen ICs. It is not as common these days as we have much more granular clockgen ICs and mostly run synchronous clocks on Intel chipsets but there can still be problem frequencies where the BIOS has not been programmed to deal with a specific combination of clock speed and hardware configuration. In these cases you need to test by trial and error and it helps to have a properly functioning configuration to compare which timing(s) may be effecting performance negatively. TLDR; Troubleshooting is time consuming so you want to keep as many variables the same as possible, look for a smaller capacity 3000 MHz CL15 kit with the same RAM IC's if possible. [Edit] If you have ASRock Timing Configurator 4.0.4 it should support the X299 chipset so that you can create some screenshots. You don't need to do anything else as we are just using it to read the current RAM timings.
post edited by chinobino. - 2020/03/25 15:56:26
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/25 16:45:03
(permalink)
chinobino.
When troubleshooting you always want to minimize the number of variables that could be potentially causing a problem.
Got it. [Edit] If you have ASRock Timing Configurator 4.0.4 it should support the X299 chipset so that you can create some screenshots. You don't need to do anything else as we are just using it to read the current RAM timings. I've downloaded that timing configurator and when I run it it is showing values that seem to be "garbage". It says I am in single channel mode, that my DRAM frequency is 6300, that my tCL, tRCD, tRP, tRAS are 0 ... it shows practically every single column as 0.
|
chinobino.
New Member
- Total Posts : 42
- Reward points : 0
- Joined: 2008/09/19 23:57:25
- Status: offline
- Ribbons : 6
Re: Low Memory Benchmark Woes
2020/03/26 04:05:30
(permalink)
Ok, so that's interesting - did you try running as administrator? Can you please try using Asus MemTweakit 2.02.44 to see if you get the same? Failing that you may have to take screenshots in the BIOS using F12 and a USB stick.
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 09:48:23
(permalink)
chinobino.
Ok, so that's interesting - did you try running as administrator?
Yeah, I tried that too and it too was returning "garbage". chinobino.
Can you please try using Asus MemTweakit 2.02.44 to see if you get the same?
Failing that you may have to take screenshots in the BIOS using F12 and a USB stick.
I'll do that next. Thank you again!
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 15:49:55
(permalink)
Today I have received different 4x32GB kit. This one is G.Skill https://www.gskill.com/product/165/326/1571734065/F4-3200C16Q-128GVKRipjaws-VDDR4-3200MHz-CL16-18-18-38-1.35V128GB-(4x32GB) G.Skill lists it as QVLd for X299 Dark. Results have shown, unfortunately, same pattern. PassMark uncached reads for single channel was 14942. Dual channel 15660. Triple channel 15879. Quad channel HUGE drop to 10454! I'm having same pattern whether I use Corsair CL15 3000 kit or G.Skill CL16 3200, whether I use auto profile or XMP, no matter what I do so I am left with impression it is not due to memory settings or memory.
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 16:15:36
(permalink)
These screenshots from PassMark MemTest86 benchmark might better illustrate drop that happens once all 4 sticks are in (there are four files attached to this post but preview shows only 3, you might want to look at all attachments) ...
|
chinobino.
New Member
- Total Posts : 42
- Reward points : 0
- Joined: 2008/09/19 23:57:25
- Status: offline
- Ribbons : 6
Re: Low Memory Benchmark Woes
2020/03/26 16:56:11
(permalink)
The motherboard/IMC seems to be having a hard time with those high density 32GB modules as the passmark scores show.
Do you get the same negative performance with more DIMMs added when using lower frequencies? i.e. 2133, 2400, 2666
It would be helpful to to know what timings the motherboard is setting.
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 17:13:05
(permalink)
chinobino.
The motherboard/IMC seems to be having a hard time with those high density 32GB modules as the passmark scores show.
I don't think IMC should be having a problem with that. Cascade Lake-X IMC specs indicate it should be able to drive 4x32 at full speed of 2933 and drop to 2666 (10% drop) past that. This board has only 4 slots which implies I should not be getting any official drop, and especially not this much. chinobino.
Do you get the same negative performance with more DIMMs added when using lower frequencies? i.e. 2133, 2400, 2666
I believe I already tried that and that I do but let me double-check it. chinobino.
It would be helpful to to know what timings the motherboard is setting.
You are right, I need to do that, thanks for the reminder :)
|
bob16314
CLASSIFIED ULTRA Member
- Total Posts : 7859
- Reward points : 0
- Joined: 2008/11/07 22:33:22
- Location: Planet of the Babes
- Status: offline
- Ribbons : 761
Re: Low Memory Benchmark Woes
2020/03/26 18:57:12
(permalink)
While an i9-10900X supports up to 256GB of DDR4 and BIOS 1.19 (and newer) adds support for 32GB memory modules, an X299 Dark only officially supports up to 64GB of DDR4. So, while it might work, I would not be surprised in the least bit that running more RAM than is officially supported by the mobo (or BIOS update to support more, which none do) has some kind of negative impact on something.
post edited by bob16314 - 2020/03/26 19:02:21
* Corsair Obsidian 450D Mid-Tower - Airflow Edition * ASUS ROG Maximus X Hero (Wi-Fi AC) * Intel i7-8700K @ 5.0 GHz * 16GB G.SKILL Trident Z 4133MHz * Sabrent Rocket 1TB M.2 SSD * WD Black 500 GB HDD * Seasonic M12 II 750W * Corsair H115i Elite Capellix 280mm * EVGA GTX 760 SC * Win7 Home/Win10 Home * "Whatever it takes, as long as it works" - Me
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 19:28:54
(permalink)
bob16314
While an i9-10900X supports up to 256GB of DDR4 and BIOS 1.19 (and newer) adds support for 32GB memory modules, an X299 Dark only officially supports up to 64GB of DDR4.
So, while it might work, I would not be surprised in the least bit that running more RAM than is officially supported by the mobo (or BIOS update to support more, which none do) has some kind of negative impact on something.
Until we know exact reason I will not assume that is the reason nor comment on that assumption. I will just say one thing for the record: While EVGA did not say "we support 32GB modules and quad channel will work with them as one would expect it to work" it also did not say opposite, "... but these things will work differently: ...", either. In the absence of such disclaimer it is fair to expect no hidden surprises.
|
bob16314
CLASSIFIED ULTRA Member
- Total Posts : 7859
- Reward points : 0
- Joined: 2008/11/07 22:33:22
- Location: Planet of the Babes
- Status: offline
- Ribbons : 761
Re: Low Memory Benchmark Woes
2020/03/26 19:39:09
(permalink)
EVGA X299 Dark
System Memory support:
• Supports Quad-Channel DDR4 up to 4000MHz+ Skylake-X (OC), and 4133MHz+ Kaby Lake-X (OC).
• Supports up to 64GB of DDR4 memory.
BIOS 1.19 Update:
• Support next-gen processors
• New OC Robot (Turbo 3.0) - adds PerCore mode, and allows to set target temperature
• Add PerCore voltage control for Vcore
• Add AVX2 for built-in stress test
• Support 32GB memory modules
• Unlock TjMax setting range from 105 to 110
Just sayin'.
* Corsair Obsidian 450D Mid-Tower - Airflow Edition * ASUS ROG Maximus X Hero (Wi-Fi AC) * Intel i7-8700K @ 5.0 GHz * 16GB G.SKILL Trident Z 4133MHz * Sabrent Rocket 1TB M.2 SSD * WD Black 500 GB HDD * Seasonic M12 II 750W * Corsair H115i Elite Capellix 280mm * EVGA GTX 760 SC * Win7 Home/Win10 Home * "Whatever it takes, as long as it works" - Me
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 20:23:36
(permalink)
bob16314
EVGA X299 Dark
System Memory support:
• Supports Quad-Channel DDR4 up to 4000MHz+ Skylake-X (OC), and 4133MHz+ Kaby Lake-X (OC).
• Supports up to 64GB of DDR4 memory.
BIOS 1.19 Update:
• Support next-gen processors
• New OC Robot (Turbo 3.0) - adds PerCore mode, and allows to set target temperature
• Add PerCore voltage control for Vcore
• Add AVX2 for built-in stress test
• Support 32GB memory modules
• Unlock TjMax setting range from 105 to 110
Just sayin'.
You are quoting two different documents issued at two different times. I also think without knowing what exactly is causing this your sayin' is not contributing anything to the topic on hand and is venturing into territory that only authorized EVGA employee should be clarifying.
|
bob16314
CLASSIFIED ULTRA Member
- Total Posts : 7859
- Reward points : 0
- Joined: 2008/11/07 22:33:22
- Location: Planet of the Babes
- Status: offline
- Ribbons : 761
Re: Low Memory Benchmark Woes
2020/03/26 20:36:50
(permalink)
It's all 'in-your-face' if you read your mobo manual and the BIOS update notes.
The board officially supports up to 64GB of DDR4 memory and BIOS 1.19 added support for 32GB sticks..Not hard to understand.
So I'll say it again, while it might work, I would not be surprised in the least bit that running more RAM than is officially supported by the mobo (or BIOS update to support more, which none do) has some kind of negative impact on something.
That is all..Carry on.
* Corsair Obsidian 450D Mid-Tower - Airflow Edition * ASUS ROG Maximus X Hero (Wi-Fi AC) * Intel i7-8700K @ 5.0 GHz * 16GB G.SKILL Trident Z 4133MHz * Sabrent Rocket 1TB M.2 SSD * WD Black 500 GB HDD * Seasonic M12 II 750W * Corsair H115i Elite Capellix 280mm * EVGA GTX 760 SC * Win7 Home/Win10 Home * "Whatever it takes, as long as it works" - Me
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 20:45:06
(permalink)
bob16314
It's all 'in-your-face' if you read your mobo manual and the BIOS update notes.
The board officially supports up to 64GB of DDR4 memory and BIOS 1.19 added support for 32GB sticks..Not hard to understand.
So I'll say it again, while it might work, I would not be surprised in the least bit that running more RAM than is officially supported by the mobo (or BIOS update to support more, which none do) has some kind of negative impact on something.
That is all..Carry on.
Like I said, I will leave it to EVGA to provide exact answer.
|
ty_ger07
Insert Custom Title Here
- Total Posts : 16602
- Reward points : 0
- Joined: 2008/04/10 23:48:15
- Location: traveler
- Status: offline
- Ribbons : 271
Re: Low Memory Benchmark Woes
2020/03/26 20:48:18
(permalink)
Good luck. If you don't want to accept the evidence which is already available, remember to be patient.
ASRock Z77 • Intel Core i7 3770K • EVGA GTX 1080 • Samsung 850 Pro • Seasonic PRIME 600W Titanium
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 21:11:16
(permalink)
ty_ger07
Good luck. If you don't want to accept the evidence which is already available, remember to be patient.
Am I really? Take a look here: https://images.evga.com/products/aplus/1/151-SX-E299-KR_EN.jpg It clearly says "128 GB on Cascade Lake-X" So to me it is clear bob and you are jumping to assumptions based on few documents that are still out of date.
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 22:45:59
(permalink)
chinobino.
Do you get the same negative performance with more DIMMs added when using lower frequencies? i.e. 2133, 2400, 2666
I just tried using 'Auto' memory settings, which puts G.Skill kit at 2666 MHz, and patter is same: 14215 in dual, 14338 in triple, and 9481 in quad.
|
ZoranC
FTW Member
- Total Posts : 1099
- Reward points : 0
- Joined: 2011/05/24 17:22:15
- Status: offline
- Ribbons : 16
Re: Low Memory Benchmark Woes
2020/03/26 22:46:32
(permalink)
Here are memory settings I get when I pick XMP Profile and reboot:
|