I9 9900k

moorhen2 · 28 Feb 2019 at 16:44

Distracted said:
Really nice. I'm also looking at changing the velocity block or at least the mounting mechanism for it.

How close in temps are your cores with the raystorm vs the ek block?

I must admit to being very disappointed with the Velocity block to be honest, with the Raystorm Pro in the above Realbench run,there was about an 8c difference between the hottest core and the coolest one, the Velocity was much worse, hence changing back to my Raystorm Pro.

Sc00p007 · 28 Feb 2019 at 18:49

moorhen2 said:
I must admit to being very disappointed with the Velocity block to be honest, with the Raystorm Pro in the above Realbench run,there was about an 8c difference between the hottest core and the coolest one, the Velocity was much worse, hence changing back to my Raystorm Pro.

Could it be possible that your velocity has a fault or wasn't mounted correctly? I use the velocity and without delid, using your bench settings, I get the same max temp as yourself and my average is within 1 degree. I feel another block wouldn't knock me down to even 5 degrees lower than yourself, hence why I think there could be something wrong with the velocity or the mount

Edit* Or thinking about it, maybe with direct die contact the difference between blocks increases? I only have experience with IHS mounts.

moorhen2 · 28 Feb 2019 at 19:27

Sc00p007 said:
Could it be possible that your velocity has a fault or wasn't mounted correctly? I use the velocity and without delid, using your bench settings, I get the same max temp as yourself and my average is within 1 degree. I feel another block wouldn't knock me down to even 5 degrees lower than yourself, hence why I think there could be something wrong with the velocity or the mount

Edit* Or thinking about it, maybe with direct die contact the difference between blocks increases? I only have experience with IHS mounts.

When using the Der8auer OC frame, with direct to die block mounting you have to get the mounting right, the EK block is already sitting on the nuts that are on the motherboard posts, before you even try screwing down, so I believe the block is barely touching the Die to start with, but I used some m4 screws to get the block on the die as it should be, so the mount was ok, and running stress tests the performance was not what I expected.

Swapping over to the Raystorm Pro, which sits straight onto the die from the get go without having to mess about, temps were much better as I have stated.

Also the flow rate is superior on the Raystorm IMO.

Sc00p007 · 28 Feb 2019 at 20:56

moorhen2 said:
When using the Der8auer OC frame, with direct to die block mounting you have to get the mounting right, the EK block is already sitting on the nuts that are on the motherboard posts, before you even try screwing down, so I believe the block is barely touching the Die to start with, but I used some m4 screws to get the block on the die as it should be, so the mount was ok, and running stress tests the performance was not what I expected.

Swapping over to the Raystorm Pro, which sits straight onto the die from the get go without having to mess about, temps were much better as I have stated. Also the flow rate is superior on the Raystorm IMO.

Ah ok, wasn't thinking about restriction. I guess if I turn my pump from 1-5 and only see 1 degree difference, that a block with a higher flow rate isn't going to give me such generous returns?

Silent_Scone · 28 Feb 2019 at 21:32

Toying with 4200MHz on the XI Formula (4x8GB). Beyond this on a 4 DIMM board requires a fairly exceptional IMC

VCCIO 1.2v / VCCSA 1.24v

moorhen2 · 1 Mar 2019 at 09:34

Very nice SS, I can do 4133mhz @16-17-17-42 CR2 with two sticks of my 3200mhz cl14 on my Extreme. Not tried any higher on this board as yet, but I know they will do 4234mhz.

Silent_Scone · 1 Mar 2019 at 09:42

That's decent for a T-Top board. Two sticks can often be tougher in this configuration. It helps to have a kit validated for those speeds (4200+), though. The guardband is fairly narrow up there.

moorhen2 · 1 Mar 2019 at 10:26

Continuing with my testing since Deliding and direct die cooling my 9900k, if you take FPU/AVX out of the equation, which I will never use as I don't game or use anything that requires it, this chip will quite happily sit at 5.2ghz with a full load voltage of between1.243v-1.252v, 1.290v set in bios with LLC of 7 with temps ranging from 59c lowest core to 65c highest core. CPU package max 67c. Not bad I think, need to see how much more this chip has to offer now I can keep the temps down.

moorhen2 · 1 Mar 2019 at 12:35

Now running RealBench which does utilise AVX, still 1.243v under full stress load, max core temp is 73c, lowest core is 65c, package temp of 73c.

Silent_Scone · 2 Mar 2019 at 09:19

4266 on Formula XI

VCCSA 1.26v
VCCIO 1.23v

moorhen2 · 2 Mar 2019 at 12:03

Looking good SS, what ram are you using, rated frequency I mean, probably Samsung B-Dies from G-Skill ?

Silent_Scone · 2 Mar 2019 at 14:27

Indeed, you need a validated kit to hit these speeds on these boards.

4266C17 (4x8GB)

moorhen2 · 2 Mar 2019 at 15:01

Silent_Scone said:
Indeed, you need a validated kit to hit these speeds on these boards.

4266C17 (4x8GB)

And a good IMC as well.

moorhen2 · 2 Mar 2019 at 21:26

Off we go at 5.3ghz, I know its only Aida64 and not utilising FPU/AVX, but it gives me a baseline to work on, stress load voltage of 1.314v, max core temp of 77c so far, cores tend to level out after about 10 minutes in general, so don't expect them to get much higher, we will see.

I could run RealBench with AVX and Aida64 non FPU/AVX at the same voltages when at 5.2ghz.

tyler_jrb · 2 Mar 2019 at 22:14

moorhen2 said:
I must admit to being very disappointed with the Velocity block to be honest, with the Raystorm Pro in the above Realbench run,there was about an 8c difference between the hottest core and the coolest one, the Velocity was much worse, hence changing back to my Raystorm Pro.

Me too, I’ve gone back to my nickel Supremacy EVO which in my opinion is better quality, bigger springs, thicker clamp plate, better microfin machining.

Also the mobo backplate that came with it wasn’t even threaded, just drilled holes.

After a bit of testing, doesn’t perform as well as my original supremacy block by quite a bit so going to swap back to that tomorrow and probably return it back to ocuk due to the non threaded backplate. Just doesn’t seem to contact right on the IHS.

I snapped up a 2nd hand 9900k 6.8-9ghz (ln2) which arrived today so got that to fit and test tomorrow when I get home from work. Also a delid rockit kit and copper IHS ordered so I can do that in a week or so.

Nice scores btw! If mine comes anywhere close to yours, I’ll be happy

Silent_Scone · 3 Mar 2019 at 10:11

moorhen2 said:
And a good IMC as well.

Depends what you want to do with it. A lot of samples will POST up there, IMC variance comes down to how conditional it is. The stability window on VCCSA/IO also can become fairly narrow, too.

Whilst I've got you in quotes, here's 30 minutes of RB on my so-so spud. Things could perk up if I moved the setup to the bench table and improved temps, but it's currently home to my X299 system.

LLC6 / Adaptive / 1.385v / RB load 1.315v
Package TDP 195w / 2x 360mm EK Slim Line
VRM Crosschill / 2080TI/ EK Acetel Full Copper CPU
Uncore: 45
Core: 50 (no AVX offset)
DRAM: 4200 CAS 17-18-18-2T 1.4v
VCCIO 1.23 / VCCSA / 1.25
Water T: 33c

moorhen2 · 3 Mar 2019 at 10:59

Very nice, but those temps are horrible. I take it your memory requires high'ish SA and IO voltages, when I run my 3200 kit at 4133 I only need 1.150v on both for HCI stability.

Silent_Scone · 3 Mar 2019 at 11:40

Realbench is based on Package, it is what it is and water temps are fair. Try those voltages with Google Stress app or Ramtest. I use this system daily so the odd synthentic bench means very little to me when it comes to CPU load.

moorhen2 · 3 Mar 2019 at 12:18

Silent_Scone said:
Realbench is based on Package, it is what it is and water temps are fair. Try those voltages with Google Stress app or Ramtest. I use this system daily so the odd synthentic bench means very little to me when it comes to CPU load.

If I can pass 1000% Coverage with HCI at these IO and SA voltages that's good enough for me.

Silent_Scone · 3 Mar 2019 at 13:25

HCI is better at testing the memory/uncore interaction. If it’s setup correctly (one instance per thread with 90% available memory) then it’s still very much a valid test of stability.

Stress app test isolates the DRAM subsystem better, though. An hour can be fairly telling (plus it’s easier to run than HCI, and coverage is quicker).

A higher VID and lower loadline on this box hasn’t had much of an impact, the end result is similiar at these kind of loads (RB). On a better CPU I may have been bothered to delid.