benchmark 9900k benefits very much from Cache overclocking (my results)

[K0NG]

since my 9900k is a bad overclocker i tested (again) working with cache overclocks.

and the results are against the ideological "CACHE OC DOES NOT DO ANYTHING" phrases at the internet.

Cache overclocking did not change the power draw and temperature nor did it need a higher voltage (except when core is increased together with the cache)

here are the first benchmark, temperature and powerdraw results:

Aida64 Extreme CPU,Cache,FPU & RAM

All Core, No AVX Offset Fixed Voltage.

Core / Cache Frequency

4.9 / 4.3 1.28v 184W 72°C
4.9 / 4.7 1.28v 189W 73°C

5.0 / 4.3 1.35v 229W 90°C
5.0 / 4.5 1.35v WHEA Error / L0 Cache Error / BSOD

 

Cinebench R15 (Average out of 10 runs)
4.9 Ghz Core 4.3 Ghz Cache = 2073

5 Ghz Core 4.3 Ghz Cache = 2133

4.9 Ghz Core 4.7 Ghz Cache = 2204

Cinebench R20 (Average out of 10 runs)
4.9 Ghz Core 4.3 Ghz Cache = 4919

5 Ghz Core 4.3 Ghz Cache = 5105

4.9 Ghz Core 4.7 Ghz Cache = 5192

Timespy CPU Benchmark (Average out of 3 runs)
4.9 Ghz Core 4.3 Ghz Cache = 11941

5 Ghz Core 4.3 Ghz Cache = 12219

4.9 Ghz Core 4.7 Ghz Cache 12495

Game Benchmarks:
Average FPS after 10 Minutes or a full Race (RTSS)

Witcher 3 Low preset Standing in Novigrad.
5.0 / 4.3 = 139 FPS
4.9 / 4.7 = 147 FPS

Assassins Creed Odyssey standing in Athens.

5.0 / 4.3 = 98 FPS
4.9 / 4.7 = 103 FPS

Project Cars 2 Race with 20 Vehicles @ Côte d'Azur.

5.0 / 4.3 = 238 FPS
4.9 / 4.7 = 264 FPS

ARMA III Start Mission.

5.0 / 4.3 = 101 FPS
4.9 / 4.7 = 114 FPS

Fallout IV Standing in Boston.

5.0 / 4.3 = 96 FPS
4.9 / 4.7 = 105 FPS

1% Lows improved as well up to 15%

in my Opinion.. Dial Back a 100mhz for much less voltage and bump up the Cache until the PC freezes or Crashes. (unstable cache is very obvious since the PC might not even boot or freeze very quickly)
 

[1van]

3 minutes ago, K0NG said:

 

What memory configuration do you have?

[K0NG]

1 minute ago, 1van said:

What memory configuration do you have?

4x8 GB 3600 Mhz CL18

[1van]

1 minute ago, K0NG said:

4x8 GB 3600 Mhz CL18

It's fun, I thought it's something like 4400 CL19

[Jurrunio]

Increasing core voltage can help with cache overclocking. I never recommend cache frequency more than 300MHz lower or any faster than core frequency during overclock too for the performance benefit. Too bad my Sandy Bridge CPU can't do it (Sandy Bridge E could)

[K0NG]

1 minute ago, Jurrunio said:

Increasing core voltage can help with cache overclocking. I never recommend cache frequency more than 300MHz lower or any faster than core frequency during overclock too for the performance benefit. Too bad my Sandy Bridge CPU can't do it (Sandy Bridge E could)

my 8086k had both identical and there was never an issue.
Core was 5.2 and Cache was 5.2 at 1.425v
worked like a charm until i sold it for this 9900k

[K0NG]

3 minutes ago, 1van said:

It's fun, I thought it's something like 4400 CL19

Just a normal Vengeance RGB Kit on XMP

[1van]

1 minute ago, K0NG said:

Just a normal Vengeance RGB Kit on XMP

Is it Samsung, or Hynix?

You might want to tune some timings to see greater results. XMP tRFC and tREFI are usually quite slow.

[K0NG]

13 minutes ago, 1van said:

Is it Samsung, or Hynix?

You might want to tune some timings to see greater results. XMP tRFC and tREFI are usually quite slow.

Micron B Die

i have the ram since a few weeks but i'll check that out sooner or later
 

[redbread123]

I have a 9700kf, will it work with that?

[K0NG]

Just now, redbread123 said:

I have a 9700kf, will it work with that?

of course.
any kind of cache oc does give better results.

try out 100 mhz more, if it's stable another 100mhz etc.

at 4.7 everything is fine but at 4.8 my PC freezes within a few minutes without BSOD.

[redbread123]

Just now, K0NG said:

of course.
any kind of cache oc does give better results.

try out 100 mhz more, if it's stable another 100mhz etc.

at 4.7 everything is fine but at 4.8 my PC freezes within a few minutes without BSOD.

Alright, after I do my build I'll see, but this build is hopefully gonna last me through high school and more, so I don't wanna risk anything (I know it probably won't damage anything, but I'm fine) lol

[K0NG]

voltage, heat and current kills your Hardware. not the clockspeed.

if you can run a CPU at 1.25v at 5 Ghz all core it will work as long as 3 Ghz at 1.25v

i never saw any kind of degradation even at high voltages and high LLC (my 8086k ran at 5.2 Ghz at 1.425v for over 15 months with no issues at all)

keep it cool and do not try to force another 100Mhz with crazy voltages and you'll be fine.. (except you are for 20 years in school)

 

[Jurrunio]

47 minutes ago, K0NG said:

my 8086k had both identical and there was never an issue.
Core was 5.2 and Cache was 5.2 at 1.425v
worked like a charm until i sold it for this 9900k

not saying there's a problem, but 100MHz of core clock is worth more than 100MHz of uncore at that point. 5.3GHz core 5.1GHz uncore is probably faster than both at 5.2GHz

[Falkentyne]

Overclocking cache ratio is more tricky on 9900k and other hyperthreaded chips because the L3 cache is used heavily in hyperthreading as well, for the hyperthreaded (logical cores), and the CPU L0 cache is the virtualized register store that sits on top of the cores, and it's required for the logical threads to think they are operating on real instruction registers.  It's more common to get L0 errors when HT is enabled than when HT is disabled (usually then you just BSOD or the program crashes).  VCCIO can help reduce L0 errors if it's caused by IMC related issues, other times vcore is needed).

 

A good way to test for IMC / L3 stability is prime95 29.8 build 6 custom 112k in place FFT (AVX disabled).  Several users have tested this successfully in the DDR4 overclocking thread.  If "Small FFT" preset tests 100% stable but 112K FFT gives CPU L0 errors (in HWinfo64) or system service exception randomly, then you need more VCCIO (Vcore may help, but sometimes a LOT of vcore is needed).

 

https://www.overclock.net/forum/5-intel-cpus/1569364-official-intel-ddr4-24-7-memory-stability-thread-1048.html#post28193324

 

https://www.overclock.net/forum/5-intel-cpus/1569364-official-intel-ddr4-24-7-memory-stability-thread-1044.html#post28189018

[K0NG]

overclocking cache is not tricky at all..

i did that on several CPUs now (4790k, 8600k, 8086k and now the 9900k)

every time there were 2 scenarios.
100% stable and absolutely fine in every single stress test or immediately crashing and freezing the whole PC within seconds after booting.

same with the 9900k
4.7 = Primestable
4.8 = Freezing without BSOD while booting.

[Falkentyne]

It is tricky because I've seen some people unable to go past x44 cache on their 9900k without needing TONS of voltage.

It happens.  Just because it didn't happen to you doesn't mean it doesn't happen.

 

And I also can't replicate your findings.  Miniscule improvements here.

5.1 / 4.7:

CB 15: 2273 (all five loops)

CB 20: 5409 (looped 3600 seconds, stopped after 5 times as the scores were the same).

 

5.1 / 4.8:

CB 15: 2285

CB 20: 5411 (scores the same all 5 loops).

 

I did not use realtime priority.  Just rebooted windows, waited a minute then ran.

[K0NG]

So.
i did All tests again and included Blender from Steam with the BMW CPU Render Scene. (made a screenshot from the 4.9 / 4.7 result because it was ridiculous.

Results from Yesterday and today on a freshly installed windows 10 pro copy. (and i was able to go a step lower on the voltage but the room was much warmer)

 

Intel I9 9900k R0 Stepping no Delid cooled with a Corsair H115i (280mm) all in one closed loop cooler.
Mainboard: ASRock Z390 Extreme4 all Powerlimits maxed out and Powersavings disabled.

All Overclocks are All Core, No AVX Offset and fixed Voltage.

Aida64 Extreme CPU,Cache,FPU & RAM Temperature and Power Draw. (Average Temperature after 10 Minutes with 22°C Ambient Temperature)
Core / Cache Frequency

4.9 / 4.3 1.264v 184W 75°C
4.9 / 4.7 1.264v 189W 77°C

5.0 / 4.3 1.35v 229W 90°C
5.0 / 4.5 1.35v WHEA Error / L0 Cache Error / BSOD

 

Cinebench R15
4.9 Ghz Core 4.3 Ghz Cache = 2073

5 Ghz Core 4.3 Ghz Cache = 2133

4.9 Ghz Core 4.7 Ghz Cache = 2171

Cinebench R20
4.9 Ghz Core 4.3 Ghz Cache = 4911

5 Ghz Core 4.3 Ghz Cache = 5115

4.9 Ghz Core 4.7 Ghz Cache = 5172

Blender Benchmark BMW (Version 2.80.75)

4.9 Ghz Core 4.3 Ghz Cache = 187 Seconds

5 Ghz Core 4.3 Ghz Cache = 175 Seconds

4.9 Ghz Core 4.7 Ghz Cache = 153 Seconds

Timespy CPU Benchmark
4.9 Ghz Core 4.3 Ghz Cache = 11984

5 Ghz Core 4.3 Ghz Cache = 12192

4.9 Ghz Core 4.7 Ghz Cache 12401

Average FPS after 10 Minutes or a full Race for Project Cars 2 (RTSS)

Witcher 3 Low preset Standing in Novigrad.
5.0 / 4.3 = 138 FPS
4.9 / 4.7 = 147 FPS

Assassins Creed Odyssey standing in Athens.

5.0 / 4.3 = 100 FPS
4.9 / 4.7 = 105 FPS

Project Cars 2 Race with 20 Vehicles @ Côte d'Azur.

5.0 / 4.3 = 244 FPS
4.9 / 4.7 = 261 FPS

ARMA III Start Mission.

5.0 / 4.3 = 99 FPS
4.9 / 4.7 = 112 FPS

Fallout IV Standing in Boston.

5.0 / 4.3 = 98 FPS
4.9 / 4.7 = 105 FPS

 

 

[Patklze]

whats the stock cache speed ?