Jeez 10900k really?

ochadd

Gawd
Joined
May 9, 2008
Messages
979
Maybe we'll see additional badass AIO water coolers come out of this. When a 280mm can't dissipate enough heat you've entered an area where only the fringe will be buying, imho. I sure don't want to listen to the machine cooling a 220 watt CPU.
 

Ready4Dis

[H]ard|Gawd
Joined
Nov 4, 2015
Messages
1,025
I don't know who this processor is for.
Someone who likes.building water cooling setups :). It's a good challenge for them to build a system that won't throttle. I'm 1/2 joking or courses it's still the fastest gaming CPU you can buy, if that's your thing it's for you. If you do other stuff and gaming @ 200fps isn't your thing, AMD is a better buy top to bottom.
 

Falkentyne

[H]ard|Gawd
Joined
Jul 19, 2000
Messages
1,636
Nice. Any luck with overclocking?
Yes, the ES chip can do 5.3 ghz all cores at around 1.310v-1.320v load or something. But I'm having some weird issue where it seems (at least from a few quick tests) that using a cache ratio of x50 makes the system MORE stable than a cache ratio of x49! I was getting CPU Cache L0 errors in Cinebench R20 at x53/x49 at 1.305v load. ES CPU bug? Or am I just cursed? idk...Maybe everything's going to hell because the load temps are above 90C...and we don't talk about fight club when your load temps are above 90C! But when I set cache to x50...it was stable? What...

My retail chip seems to need absurd volts for 5.3 all cores (like 1.4v at full load), but the 4.7, 4.8, 4.9 and 5 ghz voltages seem to be exactly the same load voltage requirement as my ES.
basically, the retail chip falls off at 5.2 ghz, and falls off HARD at 5.3 ghz.
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,263
Yes, the ES chip can do 5.3 ghz all cores at around 1.310v-1.320v load or something. But I'm having some weird issue where it seems (at least from a few quick tests) that using a cache ratio of x50 makes the system MORE stable than a cache ratio of x49! I was getting CPU Cache L0 errors in Cinebench R20 at x53/x49 at 1.305v load. ES CPU bug? Or am I just cursed? idk...Maybe everything's going to hell because the load temps are above 90C...and we don't talk about fight club when your load temps are above 90C! But when I set cache to x50...it was stable? What...

My retail chip seems to need absurd volts for 5.3 all cores (like 1.4v at full load), but the 4.7, 4.8, 4.9 and 5 ghz voltages seem to be exactly the same load voltage requirement as my ES.
basically, the retail chip falls off at 5.2 ghz, and falls off HARD at 5.3 ghz.
Interesting. Perhaps your retail chip was a 'C' batch.
https://www.techpowerup.com/266741/msi-shares-fascinating-insights-into-comet-lake-binning
 
  • Like
Reactions: N4CR
like this

Falkentyne

[H]ard|Gawd
Joined
Jul 19, 2000
Messages
1,636
No, it's probably a B batch. The average chip reaches 5.2 on all cores, not 5.3.
5.3 on all cores is A batch. Remember the Intel turbo boost specifies 5.1 on all cores for a light load on 2 cores above 70C, and 4.9 all cores below 70C and 4.8 all cores above 70C.

The vids on the retail chip are identical to my ES at 4.7-5.1 ghz, but the 5.2 ghz VID is 35mv higher on the retail chip than the ES chip. So I'm guessing the 5.3 point is maybe 50mv higher (there is no new VID programmed on either chip for 5.3 vs 5.2)
 

Falkentyne

[H]ard|Gawd
Joined
Jul 19, 2000
Messages
1,636
Here is the non AVX vf curve for the ES chip. the 94 quality one (this is in top 10%, high A tier).

The red line is the predicted VR VOUT value at 100C needed for non AVX prime95 small FFT (100C means, as if Thermal Velocity Boost Voltage Optimizations are disabled--this setting being enabled reduces the vCPU 1.55mv every 1C starting at 100C

TVB Voltage optimizations have been here since Kaby Lake. This isn't anything new.

"vCPU" is CPU VID when AC Loadline and DC Loadline are both at 0.01 mOhms.

The blue line is the predicted Bios setting at loadline calibration level 4, that Asus guesses that you need for that VR OUT.

If your CPU can do a lower VR OUT than the red line predicts (if it is at 100C), then there is a guardband of voltage leeway.
In my case, for example, at 5.2 ghz, I can do AVX small FFT disabled Prime95 with a load VOUT of 1.235v, at 90C (200 amps), while the red prediction says I need 100C.

For 5.3 ghz, I have no cooling headroom to test that. But Battlefield 5 and Minecraft (Java, loading to main menu loop test) seems to need about 1.325v load, although AIDA64 Stress FPU needs less. This might be due to violent transients that BF5 and Minecraft loading causes. Notice the prediction says I need about 1.356v.

nonavx_vfcurve2.jpg



Here is the AVX prediction chart. (Predicted based on AVX small FFT prime95).



avx_vfcurve2.jpg
 
Top