Help optimizing Samsung B-Die Ram on Intel x299

[masterdeejay]

Hi!

I have 4x8gb Patriot Viper 4 Blackout 4133MHz PVB416G413C8K

Currently it runs at 4000-15-18-18-38-cr2 1.5v memtest passed for hour+.
I want to stay at 4000 but with better timings.

It is not stable at 15-16-16-38 (failed after 10-15 min) but runs games fine.

I think i must touch the subtiming but that is a lot of settings. I only want to reduce the latency a little because it is very high i think. (i have a lot of bandwidth but i want to balance out for the best bandwidth and good latency)
When i tried to reduce the tRFC to 400 it dropped out two channels. (need more system agent voltage or it is limited by the dimms?)
The dimms are just warm, not hot but i will put a fan on them if needed.
(it is on quad channel not dual)

Cpu is stable and tested so i want to work with the memory overclock.
VCCSA, VCCIO 1.25v
System Agent +350mv (what is the safe max?)
Mesh 1.2v 3200mhz
Vcore 1.2v 45x all core

Some say that VCCSA and VCCIO do nothing for memory on X299. It is true?
By default these kits not starting until i raise the system agent, so when i do a cmos clear i must put my older ram in it and then raise the system agent to even start with this kits. (error code 33 happens)
XMP also not working, i must enter the primary timing manually. I think the motherboard not support this dimms by default so i must be carefull and go slow.

I noticed that the it very good for the perfromance to reduce the latentcy:
https://forums.overclockers.co.uk/threads/skylake-x-memory-benchmarks.18793466/

Mesh oc is completed and stable, but how can i reduce the tRFC?
Should i increase the CAS for CR1?
Or it is better to stay at CAS15 CR2?

 

[oobymach]

There's not much dif in timings the bigger dif comes from clock speed so 4ghz ram speed will automatically be better than 3.2ghz.

Running 4 or 8 sticks is always going to have more latency than 2 sticks, and the biggest latency increase will come from the tRFC so if you can get that down you'll benefit more than decreasing tcl etc.

Disable gear down mode and lower the tRFC until you get issues like bsod or failing workers during P95 then increase tRFC by 8-16 (eg from 256 to 264 or 272) and that should be stable.

 

[masterdeejay]

There's not much dif in timings the bigger dif comes from clock speed so 4ghz ram speed will automatically be better than 3.2ghz.

Running 4 or 8 sticks is always going to have more latency than 2 sticks, and the biggest latency increase will come from the tRFC so if you can get that down you'll benefit more than decreasing tcl etc.

Disable gear down mode and lower the tRFC until you get issues like bsod or failing workers during P95 then increase tRFC by 8-16 (eg from 256 to 264 or 272) and that should be stable.

I dont have gear down mode, it is not ryzen. I can set CR2 or CR1 but it only boots with CR2.
4 sticks is the minimum for quad channel but i think 65ns latency is high for 4000mhz ram and CAS15.
I can't lower the tRFC, it is now at 700, i only tried 400 but failed (boots but win11 not recognizing all 32gb ram, only 16gb when i set to 400). So something is limitig that setting.
Do i need more voltage to VCCSA, VCCIO, Mesh, SystemAgent or to the vdimm?
Or i can simply try Trfc 272 x 2 = 544 because of the 4 channels?

 

[dgianstefani]

Increase voltage to 1.5/1.6 for the memory, I run daily 1.6 with tight timings, it's fine with B die.

MemTestHelper/DDR4 OC Guide.md at oc-guide · integralfx/MemTestHelper

C# WPF to automate HCI MemTest. Contribute to integralfx/MemTestHelper development by creating an account on GitHub.

github.com

Use active cooling for the RAM.

 

[oobymach]

I dont have gear down mode, it is not ryzen. I can set CR2 or CR1 but it only boots with CR2.
4 sticks is the minimum for quad channel but i think 65ns latency is high for 4000mhz ram and CAS15.
I can't lower the tRFC, it is now at 700, i only tried 400 but failed (boots but win11 not recognizing all 32gb ram, only 16gb when i set to 400). So something is limitig that setting.
Do i need more voltage to VCCSA, VCCIO, Mesh, SystemAgent or to the vdimm?
Or i can simply try Trfc 272 x 2 = 544 because of the 4 channels?

Hmm, b-dies usually go a lot lower than 400trfc, my single rank kit likes to be around 130ns which is 256trfc at 3800mhz, can you test with just one kit of ram and see what it sets to on auto. Maybe your kit is dual rank sticks? Try again with 500trfc and then 600 if that doesn't work, you can adjust by lower amounts once you've found the stable area.

 

[Deleted member 185158]

I dont have gear down mode, it is not ryzen. I can set CR2 or CR1 but it only boots with CR2.
4 sticks is the minimum for quad channel but i think 65ns latency is high for 4000mhz ram and CAS15.
I can't lower the tRFC, it is now at 700, i only tried 400 but failed (boots but win11 not recognizing all 32gb ram, only 16gb when i set to 400). So something is limitig that setting.
Do i need more voltage to VCCSA, VCCIO, Mesh, SystemAgent or to the vdimm?
Or i can simply try Trfc 272 x 2 = 544 because of the 4 channels?

Command rate capable will depend on the PCB revision of the memory.

Use Taiphoon Burner to check Eeprom. It'll tell you the PCB revision. Generally A2 is good for only CR2 while A0 and A1 have an easier time with CR1.

Here's what the differences look like physically




Also, B-Die scales really well with v-dimm. So much that you can be conservative with SA and IO voltages.

 

[masterdeejay]

I have reduced the tRFC 600 and its working.
with 544 one channel lost
with 400 two channel lost
default is 700 (is it possible that quad channel doubles the Tref?)
So 600 quad channel is like 300 in dual channel?



with 544 one ram is missing:




With CR1 its not booting, error code 33
A0 all dimms.

 

[Deleted member 185158]

I have reduced the tRFC 600 and its working.
with 544 one channel lost
with 400 two channel lost
default is 700 (is it possible that quad channel doubles the Tref?)
So 600 quad channel is like 300 in dual channel?

View attachment 287802View attachment 287801View attachment 287800View attachment 287799View attachment 287798

with 544 one ram is missing:
View attachment 287803



With CR1 its not booting, error code 33
A0 all dimms.
View attachment 287797

That's 7th Gen memory controller for ya....

 

[masterdeejay]

That's 7th Gen memory controller for ya....

Desktop skylake is 6gen. This is skylake-x, total different. Yes it is almost maxed at 4000mhz. I have a big custom loop so voltage and temperature is not an issure.

 

[Deleted member 185158]

Desktop skylake is 6gen. This is skylake-x, total different. Yes it is almost maxed at 4000mhz. I have a big custom loop so voltage and temeprature is not an issure.
View attachment 287807
View attachment 287808

Nice setup man! Holy ship!!!

The memory controllers are hit and miss. And yours doesn't like the 4 sticks, which is pretty normal for all setups, even if it was ddr2....

Performance top = 2 sticks, single ranked for B-Die.

I say pull 2 sticks and keep pushing. If you're trying to be competitive anyways.

 

[dgianstefani]

Nice setup man! Holy ship!!!

The memory controllers are hit and miss. And yours doesn't like the 4 sticks, which is pretty normal for all setups, even if it was ddr2....

Performance top = 2 sticks, single ranked for B-Die.

I say pull 2 sticks and keep pushing. If you're trying to be competitive anyways.

Dual ranked two sticks is actually optimal.

I went from 2x8 SR 4000/14 to 2x16 DR 3933/14 and it's still faster.

But on a quad channel setup, 4xSR is optimal.

Your spec numbers may seem better with fewer ranks, but the actual performance in applications and games is still higher with two ranks per channel.

 

[Deleted member 185158]

Dual ranked two sticks is actually optimal.

I went from 2x8 SR 4000/14 to 2x16 DR 3933/14 and it's still faster.

But on a quad channel setup, 4xSR is optimal.

Your spec numbers may seem better with fewer ranks, but the actual performance in applications and games is still higher with two ranks per channel.

2x single ranked 4600mhz won't be touched by 4000 in quad dual rank.

And if we are gaming with this set up, any current gen Intel chip under 200$ is faster.

Any ways, I'll ask it out loud, why try for benchmarking figures on a gaming rig?
And why use a modern gpu with system hardware that old?
Why ask why. I just give information. OP does what they need to do.

 

[masterdeejay]

im experimenting with AI, learning, virtual machines... sometimes i use this rig for those things
now im using stable diffusion with pytorch and Intel Arc AI acceleration
And i also use a few tesla cards for my projects, but i have a few rigs for these.
Yes im also gaming but with an Intel A770 16gb, and this cpu is not botlenecking it (it is just an RTX3060/3070 performance card). I playing on windowed 60-75hz (not even 1080p) because i need the screen for another things when im playing.
I need a lot of pcie lanes for multiple cards, that is why a new ryzen or intel is not an option

and stable diffusion ai can use all cores


This is from techpowerup logo and ShrimpBrime mixed

 

[dgianstefani]

2x single ranked 4600mhz won't be touched by 4000 in quad dual rank.

And if we are gaming with this set up, any current gen Intel chip under 200$ is faster.

Any ways, I'll ask it out loud, why try for benchmarking figures on a gaming rig?
And why use a modern gpu with system hardware that old?
Why ask why. I just give information. OP does what they need to do.

Dual channel 4600 SR does not beat quad channel 4000 SR.

And good luck getting more than 4133 on SkylakeX lmao.

 

[ThrashZone]

Hi,
I'd personally just be happy with tight c16 timings and way less dimm voltage at 4000 only 1.45v should be plenty.
You should be able to push rfc down to 350 or 20-30 above where it won't post.
cke down to 7

refi double what it is at minimum max is somewhere around 45k

 

[Deleted member 185158]

Dual channel 4600 SR does not beat quad channel 4000 SR.

And good luck getting more than 4133 on SkylakeX lmao.

Right I already stated that. The memory controller....

And no competitive overclocker uses a 4 stick board for overclocking. Well some do, but pricing man.... That's why you buy into the Asus Apex, run 2 sticks, single rank....

For a daily driver, quad channel doing some work, I don't think it's worth the effort to find benchmarking numbers for sub 1% differences.

 

[ThrashZone]

Hi,
Not sure this is cruncher stable but benchmarked and games okay
I have a wonkey one that benched better but this one is more stable but both are T1 just loosen up a little
Write latency c15 lol good old asus bios

[2019/07/19 21:55:40]
Ai Overclock Tuner [Manual]
CPU Strap [100]
BCLK Frequency [100.0000]
ASUS MultiCore Enhancement [Auto]
AVX Instruction Core Ratio Negative Offset [3]
AVX-512 Instruction Core Ratio Negative Offset [5]
CPU Core Ratio [Sync All Cores]
ALL-Core Ratio Limit [48]
Min. CPU Cache Ratio [Auto]
Max. CPU Cache Ratio [32]
BCLK Frequency : DRAM Frequency Ratio [Auto]
DRAM Frequency [DDR4-4000MHz]
TPU [Keep Current Settings]
CPU SVID Support [Auto]

DRAM CAS# Latency [16]
DRAM RAS# to CAS# Delay [17]
DRAM RAS# PRE Time [16]
DRAM RAS# ACT Time [36]
DRAM Command Rate [Timing 1T]
DRAM RAS# to RAS# Delay [4]
DRAM RAS# to RAS# Delay L [6]
DRAM REF Cycle Time [373]
DRAM Refresh Interval [32767]

DRAM WRITE Recovery Time [Auto]
DRAM READ to PRE Time [4]
DRAM FOUR ACT WIN Time [18]
DRAM WRITE to READ Delay [6]
DRAM WRITE to READ Delay L [10]
DRAM CKE Minimum Pulse Width [8]
DRAM Write Latency [16]

DRAM CLK Period [Auto]
Attempt Fast Boot [Auto]
Attempt Fast Cold Boot [Auto]
DRAM Training [Auto]
WR CRC feature Control [Auto]
Duty Cycle Training [Auto]
Read Vref Centering [Auto]
Eye Diagrams [Auto]
Turnaround Time Optimization [Auto]
PDA [Auto]
Write Vref Centering [Auto]
Enhanced Training(CHA) [Auto]
Enhanced Training(CHB) [Auto]
Enhanced Training(CHC) [Auto]
Enhanced Training(CHD) [Auto]

CPU Load-line Calibration [Level 4]
CPU Current Capability [140%]
CPU VRM Switching Frequency [Auto]
VRM Spread Spectrum [Disabled]
CPU Power Phase Control [Standard]
CPU Power Thermal Control [125]

CPU Input Boot Voltage [Auto]
CPU System Agent Boot Voltage [Auto]
CPU VCCIO Boot Voltage [Auto]

Enhanced Intel SpeedStep Technology [Enabled]
Turbo Mode [Enabled]
Long Duration Package Power Limit [600]
Package Power Time Window [127]
Short Duration Package Power Limit [Auto]
CPU Integrated VR Current Limit [Auto]
CPU Integrated VR Fault Management [Disabled]
CPU Integrated VR Efficiency Management [High Performance]

VIN Tracker [Auto]
Self-tracked Clocking [Auto]
PCIe Gen3 PLL Clock Control [Enabled]
PLL Post Divider Adjust [Auto]
Change PLLTRIM Prefix [+]
Change PllTrim Value [Auto]
Change MC-PLLTRIM Prefix [+]
Change MC-PllTrim Value [Auto]
Turbo Residence Tweak 0 [8]
Turbo Residence Tweak 1 [6]
Turbo Residence Tweak 2 [4]
Turbo Residence Tweak 3 [2]
Initial BCLK Frequency [Auto]
CPU BCLK Amplitude [Auto]
CPU BCLK Spread Spectrum [Auto]
CPU BCLK Frequency Slew Rate [Auto]
CPU BCLK Slew Rate [Auto]
PCIE/DMI Slew Rate [Auto]
CPU AUX1 Voltage [Auto]
CPU AUX2 Voltage [Auto]
CPU Input Reset Voltage [Auto]
VTTDDR Voltage(CHA, CHB) [Auto]
VTTDDR Voltage(CHC, CHD) [Auto]
VPPDDR Voltage (CHA, CHB) [Auto]
VPPDDR Voltage (CHC, CHD) [Auto]

CPU Core Voltage [Adaptive Mode]
- Offset Mode Sign [+]
CPU Core Voltage Offset [Auto]
Additional Turbo Mode CPU Core Voltage [1.240]
CPU Cache Voltage [Manual Mode]
CPU Cache Voltage Override [1.180]

Uncore Voltage Offset [Auto]
Uncore Voltage Offset Prefix [+]
CPU Input Voltage [1.930]
DRAM Voltage(CHA, CHB) [1.4100]
DRAM Voltage(CHC, CHD) [1.4100]
CPU VCCIO Voltage [1.13125]
CPU System Agent Voltage [0.93500]
PCH Core Voltage [1.01250]

PLL Reference Offset Mode Sign [+]
PLL Reference Offset Value [Auto]
Hyper-Threading [ALL] [Enabled]
Max CPUID Value Limit [Disabled]
Execute Disable Bit [Enabled]
Hardware Prefetcher [Enabled]
Adjacent Cache Prefetch [Enabled]
VMX [Enabled]
Boot performance mode [Max Performance]
Maximum CPU Core Temperature [Auto]

Enhanced Intel SpeedStep Technology [Enabled]
Turbo Mode [Enabled]
Autonomous Core C-State [Auto]
Intel(R) Speed Shift Technology [Auto]
MFC Mode Override [MFC Driver Override]
MSR Lock Control [Enabled]
PCI Express Native Power Management [Disabled]
PCH DMI ASPM [Disabled]
ASPM [Disabled]
DMI Link ASPM Control [Disabled]
PEG - ASPM [Disabled]
Intel® VT for Directed I/O (VT-d) [Enabled]
MCTP [Disabled]
ACS Control [Disabled]