Question Poor performance after upgrading CPU ?

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Oct 13, 2021
43
8
45
3
Hopefully, your experience is just a little bit better and faster now. Personally, I'm out of recommendations for increasing the frequency and stability for your DRAM. If you haven't I would disable power down mode for the last push for stability. Keep GDM on, Power Down mode off, and try to lower tRC as much as possible. This is just about the extent of my limited memory knowledge at how old I am.
 

mamasan2000

Distinguished
tFAW is working in conjunction with tRRDS and L. Stands for Fourth Active/Activate Window. So you gotta multiply something with 4. And that is your tRRD. 4-4-16 commonly works. Sometimes you might have to use 4-6-16. Why didn't I use 24 instead? Because 4*4 = 16. When RAM has to do the 6 cycles, it ignores tFAW since that is lower than 64 = 24. Meaning, it's more efficient with the lower tFAW.
Yours are 7 and 9. It's going to ignore your tFAW. 74 = 28, 94 = 36. Both are higher than 24. These numbers are cycles. Higher = slower. I think every DRAM command is preceded by an Activate-command so these 3 values are used all the time.
S stands for Short and L stands for Long.

Something seems messed up with your system.

Power Down mode increases latency since RAM tries to save power and there is a lag between Power On and Off. I've never had it turned on. RAM draws around 3 watts, who cares?
GDM disabled also gives lower latency, especially GDM off and 1T commandrate. But it is harder to stabilise RAM with GDM off. So it's a convenience to leave it on.
Goes from GDM On -> GDM off 2T -> GDM off 1T in terms of difficulty.
 
Last edited:
tFAW is working in conjunction with tRRDS and L. Stands for Fourth Active/Activate Window. So you gotta multiply something with 4. And that is your tRRD. 4-4-16 commonly works. Sometimes you might have to use 4-6-16. Why didn't I use 24 instead? Because 4*4 = 16. When RAM has to do the 6 cycles, it ignores tFAW since that is lower than 64 = 24. Meaning, it's more efficient with the lower tFAW.
Yours are 7 and 9. It's going to ignore your tFAW. 74 = 28, 94 = 36. Both are higher than 24. These numbers are cycles. Higher = slower. I think every DRAM command is preceded by an Activate-command so these 3 values are used all the time.
S stands for Short and L stands for Long.

Something seems messed up with your system.

Power Down mode increases latency since RAM tries to save power and there is a lag between Power On and Off. I've never had it turned on. RAM draws around 3 watts, who cares?
GDM disabled also gives lower latency, especially GDM off and 1T commandrate. But it is harder to stabilise RAM with GDM off. So it's a convenience to leave it on.
Goes from GDM On -> GDM off 2T -> GDM off 1T in terms of difficulty.
right, ill check that tFAW later, first gen didnt alow me to reduce RRD at all...and faw around 40 was running stable
RRD is at 4/6 now for S/L
power down is disabled
gdm... hmmm i doubt it will run in 1t with gdm off..its dual rank modules and 2t with gdm off is slower then gdm on 1t
but ill check it later if CL15 with gdm off would work (that could offset latency diff)
 
so this seems to be best result for my cpu and ram
didnt noticed before that amd power plan goes like 4175MHz on all cores with 105watts on PPT
so that ipc gain was there...just frequency wasnt as expected

PBO is not enabled, all stays within AMD power limits
it is boosting now to 4.6GHz on light tasks (up to 25% cpu usage with P2 profile) 1,424volts
and 4443MHz CCX1/4418MHz CCX2 on heavy tasks (above 26% with P1 profile) 1.328volts
theres minimal performance gain with raising voltage (for more clock, as i would be hitting 140A EDC limit
voltage is lower than AMD gave me, and raised frequency to stay within rated 105watts (stress test or cinebench doesnt go above 100 watts)

https://www.userbenchmark.com/UserRun/47132492
cpu-z validatation updated in signature
cinebench is also okay
 
this might by silly question from me, but i couldnt find any answer on the internet

somehow ive managed to make FCLK, MCLK, UCLK downclock/boost alongside with power profiles
zentimings doesnt show FCLK, MCLK, UCLK frequency anymore , but multipliers which change alongside with min/max/boost profiles
it somehow droped latency

from zen timings which doesnt shows frequency anymore...but some numbers which does change during pc use


from hwinfo64

from sandra

SiSoftware Sandra

Benchmark Results
Memory Latency : 37.7ns (37.5ns - 38.2ns)
Decimal Numeral System (base 10) : 1s = 1000ms, 1ms = 1000µs, 1µs = 1000ns, etc.
L1D (1st Level) Data Cache : 26.0clocks (25.0clocks - 26.0clocks)
L2 (2nd Level) Unified/Data Cache : 26.0clocks (25.0clocks - 27.0clocks)
L3 (3rd Level) Unified/Data Cache : 164.0clocks (157.0clocks - 170.0clocks)
Speed Factor : 43.30
Results Interpretation : Lower Scores mean Better Performance.

Performance vs. Power
Chipset(s)/Memory Power : 40.56W
Memory Latency : 0.93ns/W
Results Interpretation : Lower Scores mean Better Performance.

Capacity vs. Power
Memory Capacity : 403.94MB/W
Results Interpretation : Higher Scores mean Better Performance.

Performance vs. Speed
Memory Latency : 0.01ns/MHz
Results Interpretation : Lower Scores mean Better Performance.

Detailed Results
2kB Range : 0.9ns / 4.0clocks
4kB Range : 0.9ns / 4.0clocks
8kB Range : 0.9ns / 4.0clocks
16kB Range : 0.9ns / 4.0clocks
32kB Range : 0.9ns / 4.0clocks
64kB Range : 1.8ns (0.9ns - 2.1ns) / 8.0clocks (4.0clocks - 10.0clocks)
128kB Range : 2.2ns (0.9ns - 2.4ns) / 10.0clocks (4.0clocks - 11.0clocks)
256kB Range : 2.3ns (2.2ns - 2.4ns) / 11.0clocks (10.0clocks - 11.0clocks)
512kB Range : 2.3ns / 11.0clocks (10.0clocks - 11.0clocks)
1MB Range : 5.0ns (4.3ns - 5.4ns) / 23.0clocks (20.0clocks - 25.0clocks)
2MB Range : 5.5ns (5.3ns - 5.9ns) / 25.0clocks (24.0clocks - 27.0clocks)
4MB Range : 5.6ns (5.4ns - 5.7ns) / 26.0clocks (25.0clocks - 26.0clocks)
8MB Range : 5.6ns (5.4ns - 5.8ns) / 26.0clocks (25.0clocks - 26.0clocks)
16MB Range : 6.0ns (5.7ns - 6.4ns) / 27.0clocks (26.0clocks - 29.0clocks)
32MB Range : 35.6ns (34.7ns - 36.3ns) / 163.0clocks (154.0clocks - 167.0clocks)
64MB Range : 37.7ns (36.9ns - 38.2ns) / 173.0clocks (169.0clocks - 175.0clocks)
256MB Range : 37.7ns (37.5ns - 38.0ns) / 173.0clocks (172.0clocks - 175.0clocks)
1GB Range : 37.7ns (37.5ns - 37.9ns) / 173.0clocks (172.0clocks - 174.0clocks)

Benchmark Status
Result ID : AMD F19 (Ryzen3/TR3 Matisse) Host Bridge; 2x 8GB G.Skill F4-3200C14-16GTZR DDR4 (3.5GHz 9% OC 128-bit) PC4-25600 (16-16-16-32 6-48-12-12)
Revision : P6.00 46.62
Computer : ASRock X370 Gaming X
Platform Compliance : x64
No. Threads : 1
Processor Affinity : U0-M0C0T0 U2-M0C1T0 U4-M0C2T0 U6-M0C3T0 U8-M1C0T0 U10-M1C1T0 U12-M1C2T0 U14-M1C3T0 U1-M0C0T1 U3-M0C1T1 U5-M0C2T1 U7-M0C3T1 U9-M1C0T1 U11-M1C1T1 U13-M1C2T1 U15-M1C3T1
System Timer : 10MHz
Page Size : 4kB
Access Pattern : In-Page Random Access Pattern
Type : Global Data Memory

Processor
Model : AMD Ryzen 7 3800X 8-Core Processor
URL : https://www.amd.com
Speed : 4.6GHz
Min/Max/Turbo Speed : 2.22GHz - 4GHz - 4.6GHz
Modules per Processor : 2 Unit(s)
Cores per Processor : 4 Unit(s)
Threads per Core : 2 Unit(s)
Front-Side Bus Speed : 101MHz
Revision/Stepping : 71 / 0
Microcode : 870F10-1021
Latest Version : 870F10-21
L1D (1st Level) Data Cache : 8x 32kB, 8-Way, Exclusive, 64bytes Line Size, 2 Thread(s)
L1I (1st Level) Code Cache : 8x 32kB, 8-Way, Exclusive, 64bytes Line Size, 2 Thread(s)
L2 (2nd Level) Unified/Data Cache : 8x 512kB, 8-Way, Fully Inclusive, 64bytes Line Size, 2 Thread(s)
L3 (3rd Level) Unified/Data Cache : 2x 16MB, 16-Way, Exclusive, 64bytes Line Size, 8 Thread(s)
Rated Power (TDP) : 105.00W

Memory Controller
Speed : 1.75GHz (100%)
Min/Max/Turbo Speed : 875MHz - 1.75GHz

Computer
Name : ASRock X370 Gaming X
Computer BIOS : P6.00

Chipset
Model : AMD F19 (Ryzen3/TR3 Matisse) Host Bridge
URL : http://www.amd.com
Revision : A1
Front-Side Bus Speed : 17x 101MHz (1.72GHz)
Width : 256-bit
Maximum Bus Bandwidth : 53.66GB/s

Logical/Chipset Memory Banks
Version : 46.62
Bank : 8GB DDR4 16-16-16-32 6-48-12-12 1T
Bank : 8GB DDR4 16-16-16-32 6-48-12-12 1T
Channels : 2
Width : 128-bit
Memory Bus Speed : 2x 1.75GHz (3.5GHz)
Min/Max/Turbo Speed : 2x 875MHz (1.75GHz) - 2x 1.75GHz (3.5GHz)
Multiplier : 52/3x
Min/Max/Turbo Multiplier : 52/6x - 52/3x
Integrated in Processor : Yes
Maximum Memory Bus Bandwidth : 54.72GB/s

Memory Module(s)
Memory Module : G.Skill/Samsung F4-3200C14-16GTZR 16GB DDR4 2Rx8 PC4-25600U DDR4-3200 (15-15-15-36 4-51-16-5)
Memory Module : G.Skill/Samsung F4-3200C14-16GTZR 16GB DDR4 2Rx8 PC4-25600U DDR4-3200 (15-15-15-36 4-51-16-5)

Performance Enhancing Tips
Warning 5010 : Cannot use Large Memory Pages due to lack of privileges.
Notice 242 : Dynamic OverClocking/Turbo engaged. Environment conditions may influence performance.
Tip 229 : CPU microcode update available. Check for an updated System BIOS with updated microcode.
Tip 3 : Double-click tip or press Enter while a tip is selected for more information about the tip.

from aida
had to slightly descrease ram base clock to 3466..as during boost it would go over 3600mhz...which gave me just bsods
ram is boosting to 3574Mhz now stable :)

so, is this normal? coz i cant find any single post on google about running FCLK, MCLK, UCLK in some unlocked state
and till now i didnt even know that desktop ram can change frequency on the fly
which is a bit weird, since ram timing settings doesnt have configuration for different clocks (as in GPUs)
 

ASK THE COMMUNITY

TRENDING THREADS