Overclock.net › Forums › Intel › Intel CPUs › Skylake-X/Kaby Lake-X Combined Discussion
New Posts  All Forums:Forum Nav:

Skylake-X/Kaby Lake-X Combined Discussion - Page 436

post #4351 of 6051
So, when I built my 7820X system, I ordered two kits of G.Skill DDR4 3600 15-15-15-35-2T. One of the kits I received was a 16-16-16-36-2T kit, but I dealt with it and used it because one stick in the 15-15-15-35 kit I received was faulty and had to be RMA'd.

I have both sets now, so I have the option to return the 16-16-16-36 kit for the correct kit. Should I bother? I can run the kits together in quad-channel at 1.395V at DDR4 4000 17-17-17-37-1T. The 15-15-15-35 kit doesn't really seem to overclock any better. They're both clearly B-Die.

Also, what's everyone getting for memory latency in AIDA64? At DDR4 4000 17-17-17-37-1T, I'm averaging around 53ns. I've seen as low as 50.7ns. Am I realistically going to drop latency by any meaningful amount?
Quote:
Originally Posted by surfinchina View Post

Yes exactly.
I think the headroom for overclocking in all of these i9s is way better than any chip I've had in the past.
So much for the "Intel rushing out an answer to the threadripper" stories...

The "rushing out a response bit" is partially true. The X299 release got pushed forward by two months and resulted in a relatively poor selection of motherboards, most of which had glaring issues. The HCC parts were also not originally on the roadmap, and were almost surely a reaction to Threadripper.

Also, the headroom on the i9s isn't terribly amazing. They're pushing overclocks that are, proportionally, similar to Nehalem and Sandy Bridge. The fact that they all seem like they could hit 5.0 GHz at reasonable voltages if temperatures allowed isn't surprising when you realize they're built on the same process as Kaby Lake.
Edited by pantsaregood - 10/17/17 at 9:30pm
post #4352 of 6051
Quote:
Originally Posted by surfinchina View Post

Yes indeed.
I understand the heat. Even my old 5960 produced a lot of it and my 7900 has 2 more cores, so 25% more watts drawn I guess.
I think that it's ok to produce a CPU that needs a basic AIO when not overclocked.
Now, if you want to overclock you need a proper watercool which includes water VRM cooling and if you want to get a great overclock you need to delid. I also don't see a problem with that!
But then I like the hardware thing more than the software thing...
Oh, don't get me wrong, I've been pleading out-loud for this chip... Leave the cooling to us! As I suspected, its only a partial solution since the VRM and TIM have to support it...

I've been scrambling to get to a fully functional testbed with this chip today because its been gnawing at me for months to know what their mesh vs ring did to heavy IPC apps...

Answer: It hurt them...

Still need to shake out the potential causes beyond the mesh/L3 configuration (like the potential for micro-throttling), but...

I have this application that scales reasonably well out to 18 cores on my 2x2696v3 system @3.8GHz. FAR from linear as has a lot of IPC. So, by 18 cores you are hitting the law of diminishing returns HARD in terms of adding cores to gain performance. 20 cores still runs at roughly the same clock speed, but is no faster in wall-clock time.

Now, I'm running the same application on this single socket chip @4.2GHz all-core (3GHz mesh, 3200 CAS14 memory). I can run it for 2 minutes or 2 days, that part is entirely linear, so I ran a short one:

Here's the first result:
7980xe 18 cores @ 4.2GHz: 1m:45s
2x2696v3 18 cores @ 3.8Ghz: 1m:27s

Again, lots of variables to remove from this equation before I call it conclusive, but.... It appears my concerns about the mesh being essential for long-term scaleability, but pretty painful right now at this core count for apps with heavy IPC may be justified...

Fun stuff! biggrin.gif
post #4353 of 6051
Quote:
Originally Posted by cekim View Post

Oh, don't get me wrong, I've been pleading out-loud for this chip... Leave the cooling to us! As I suspected, its only a partial solution since the VRM and TIM have to support it...

I've been scrambling to get to a fully functional testbed with this chip today because its been gnawing at me for months to know what their mesh vs ring did to heavy IPC apps...

Answer: It hurt them...

Still need to shake out the potential causes beyond the mesh/L3 configuration (like the potential for micro-throttling), but...

I have this application that scales reasonably well out to 18 cores on my 2x2696v3 system @3.8GHz. FAR from linear as has a lot of IPC. So, by 18 cores you are hitting the law of diminishing returns HARD in terms of adding cores to gain performance. 20 cores still runs at roughly the same clock speed, but is no faster in wall-clock time.

Now, I'm running the same application on this single socket chip @4.2GHz all-core (3GHz mesh, 3200 CAS14 memory). I can run it for 2 minutes or 2 days, that part is entirely linear, so I ran a short one:

Here's the first result:
7980xe 18 cores @ 4.2GHz: 1m:45s
2x2696v3 18 cores @ 3.8Ghz: 1m:27s

Again, lots of variables to remove from this equation before I call it conclusive, but.... It appears my concerns about the mesh being essential for long-term scaleability, but pretty painful right now at this core count for apps with heavy IPC may be justified...

Fun stuff! biggrin.gif

What app?

Great test btw :
post #4354 of 6051
Thread Starter 
Ah one of the reasons I love Asus.
The free mouse give away for people who bought a Asus x299 motherboard or ROG GTX1080ti, I bought both.

Nice little mouse for free biggrin.gif

post #4355 of 6051
Apex is finally available here at e500,- . Grrrrr I've had so many problems with ASUS but it's still tempting because it has a monoblock available and the Taichi doesn't.

Btw, are there actually any monoblocks for the R6A yet?!
Edited by Martin778 - 10/18/17 at 1:58am
post #4356 of 6051
Quote:
Originally Posted by Martin778 View Post

Apex is finally available here at e500,- . Grrrrr I've had so many problems with ASUS but it's still tempting because it has a monoblock available and the Taichi doesn't.

Btw, are there actually any monoblocks for the R6A yet?!
There is, but no stock at EK
https://www.ekwb.com/shop/catalogsearch/result/?q=R6e
CaseLabs Panda
(29 items)
 
Bench table
(18 items)
 
Fractal White
(17 items)
 
CPUMotherboardGraphicsRAM
5960X Rampage V Extreme KFA² GTX 980Ti HOF Watercooled Kingston HyperX HX426C13PB2K4/16 
RAMHard DriveHard DriveCooling
G.Skill TridentZ F4-3200C14Q-32GTZ Kingston v300 Intel SSD 535 EK-Supremacy EVO - Full Nickel 
CoolingCoolingCoolingOS
EK-Coostream XE 360 EK-Vardar F4-120ER EK-XRES 140 D5 Vario (incl. pump) Win10 / Win 8.1 / Win 7 / Linux Mint 
MonitorKeyboardPowerCase
Dell U2515H Ducky Channel Shine III MX black - white LED /... Enermax Platimax 1000w dimastech easy v3 
MouseMouse Pad
logitech G9X HyperX 900x400 / Steelseries QCK 
  hide details  
Reply
CaseLabs Panda
(29 items)
 
Bench table
(18 items)
 
Fractal White
(17 items)
 
CPUMotherboardGraphicsRAM
5960X Rampage V Extreme KFA² GTX 980Ti HOF Watercooled Kingston HyperX HX426C13PB2K4/16 
RAMHard DriveHard DriveCooling
G.Skill TridentZ F4-3200C14Q-32GTZ Kingston v300 Intel SSD 535 EK-Supremacy EVO - Full Nickel 
CoolingCoolingCoolingOS
EK-Coostream XE 360 EK-Vardar F4-120ER EK-XRES 140 D5 Vario (incl. pump) Win10 / Win 8.1 / Win 7 / Linux Mint 
MonitorKeyboardPowerCase
Dell U2515H Ducky Channel Shine III MX black - white LED /... Enermax Platimax 1000w dimastech easy v3 
MouseMouse Pad
logitech G9X HyperX 900x400 / Steelseries QCK 
  hide details  
Reply
post #4357 of 6051
Quote:
Originally Posted by Martin778 View Post

Apex is finally available here at e500,- . Grrrrr I've had so many problems with ASUS but it's still tempting because it has a monoblock available and the Taichi doesn't.

Btw, are there actually any monoblocks for the R6A yet?!

EK has one for both RVI Extreme and the RVI Apex, same monoblock works for both.

I have the apex, but will stick to my EK HF and a fan @ the VRMS instead. I prefer the look more.
    
CPUMotherboardGraphicsRAM
8700k Apex X TITAN X (Pascal)  Trident Z 4400MHz 
Hard DriveCoolingCoolingOS
750 NVMe 1.2TB EK HF/rx360/rx120 EK FC/GTs/mcp655 Windows 10 Pro  
MonitorPowerCaseAudio
BL3200PT 1600 T2 800d Pro Audio 2.1 + DragonFly Red 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
8700k Apex X TITAN X (Pascal)  Trident Z 4400MHz 
Hard DriveCoolingCoolingOS
750 NVMe 1.2TB EK HF/rx360/rx120 EK FC/GTs/mcp655 Windows 10 Pro  
MonitorPowerCaseAudio
BL3200PT 1600 T2 800d Pro Audio 2.1 + DragonFly Red 
  hide details  
Reply
post #4358 of 6051
I've just got a link sent to me that apparently there IS a monoblock for the Taichi too, but from Bitspower. Doesn't seem to be available yet, however.
https://www.facebook.com/Bitspower/posts/1463828430350436
Changing the board would be another 250 euro's after selling the Taichi and then another 120 for the monoblock...

+
To everyone who delidded an SKL-X, how much liquid metal did you use, was the 1g tube enough?
Edited by Martin778 - 10/18/17 at 2:29am
post #4359 of 6051
Quote:
Originally Posted by Martin778 View Post

To everyone who delidded an SKL-X, how much liquid metal did you use, was the 1g tube enough?

1g is plenty
post #4360 of 6051
Quote:
Originally Posted by haavard View Post

1g is plenty
Yep. You won't need more. It is more than enough!
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › Skylake-X/Kaby Lake-X Combined Discussion