*Official* NBR Desktop Overclocker's Lounge [laptop owners welcome, too]

sounds like a perfect man cave to me

 

You do have a few rigs How do you choose?

 

Truly depends on the temp to be quite honest. I haven't touched my X299 rig in a week because of the temps increasing (on average) 10 degrees. I'm currently working on memory timings on my 9900K rig. My sig doesn't include the partial X99 and Z370 builds I have, that I'm likely going to let go on the cheap. My X399 has become my main work/play machine, which may surprise some here.

 

indeed! tell us your reasoning

 

Doesn't surprise me in the slightest... Lol.

Sent from my SM-G900P using Tapatalk

 

I was going to go into a long-winded spiel but I will shorten with one word. Practicality. Regardless of the reviews, hate, love and hype of both X399 and X299, my AMD rig serves it purpose very well. A good platform allows itself to disappear, not getting in the way. When I use my AMD rig, I don't notice it is AMD. It simply works. (This does not mean I have or had issues with X299).

 

If I were to be honest, the last Intel desktop cpu that I purchased separately specifically for a build, would be a Pentium 4. Of course prior to the last two ryzen builds Ive done, the last cpu would have been the Phenom 2

 

New PR..

9900K @ 5.4GHz / 2080Ti
Fire Strike Ultra: 9687

 

3300 mhz tm5 stable, 3333 mhz booting

just a small issue: first boot after setting oc clocks, timings and voltages, as well as hot reboots afterwards go down without a hitch. but when i do a cold reboot with oc settings im getting a black screen. how come the two diverge in this manner? any suggestions? havent had this happen before, either something boots or it doesnt...

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Sounds like an issue of memory training. It trains, then doing reboots are no problem. But a cold boot from fully off requires the system to retrain the memory. So it likely isn't finding a stable mem timing when retraining. One way to get around this is turn mem training off if the BIOS/EUFI has that setting. This way, once a trained and stable state is found, it won't try to run through to find the timings and find a resolve for boot, you just use the prior timing configuration.

 

Try saving the data file locally, then go to hwbot and use the upload method. See if that works.

 

I forgot to mention, if it is boot training at issue and you don't have the option to turn off boot training, there is an alternative: setting up boot voltages! Both Intel and AMD, although Intel seems to have more at times (at least compared to my current MB), allow sometimes for setting higher memory boot voltages, higher SA or VCCIO boot voltages, etc., but still within safe voltage limits. Sometimes, having these higher than the voltages you have set that are tested stable can allow for it to train and POST.

So that is another way you can play with it if you BIOS/UEFI allows you to set a boot voltage. You may want to play with a couple other voltages also to find stability or voltages that allow for a boot solution. One of those is the Vref voltage, which sets where the ram recognizes above that as a 1 and below as a 0.

Just thought I should mention that as another way to tackle the issue if you cannot turn off boot training after you find a stable configuration.

 

aha!i have the feeling this was spot on! found a setting in bios labeled "Mrc fast boot" which is enabled by default. turns out that enabling this actually switches off memory training! so lets see how it goes from here on out

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

I saw the same behaviour on my desktop, and it was indeed the memory training kicking in. I only had the primary timings set. After setting every auto value in the DRAM timing section of the BIOS manually, it stopped doing so.

 

Looks like it shipped early



Sent from my SM-G960U using Tapatalk

 

i am actually adjusting all available ram settings in bios when tuning the ram but i can imagine that still not all settings are exposed so there are still some untouched and have to be trained...

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Disabling memory training on boot didn't help then?

 

oh i actually havent come around to testing that yet haha maybe later today, just so much other stuff going on that i barely get an hour of tuning in each day. and when it comes to RAM, one hour means nothing ill keep you guys posted.

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

3333 mhz tm5 stable, 3400 mhz booting

although to get 3333 mhz error free in tm5, i had to up VDIMM to 1.40V and IO/SA to 1.30V. already checked if 1.45V VDIMM was doable but no dice and 1.30V is already max for my taste for IO/SA. so i guess im slowly reaching my limits here.

edit: yep, i guess thats it for me frequency-wise @3333 mhz. 3400 mhz boots but throws up errors in tm5, plus first reboot threw me a black screen.

not too shabby for a 4 yr old machine! now to check and see how much i can tighten up the timings and dial the voltages back down

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Let me know how it performs, as I'm getting the EK block.

 

Happy Saturday.

 

Primochill must be really close to my house, as it arrived today. Thinking it wouldnt ship until the 25th, I havent bothered to order any fittings yet . So my testing will have to wait until they arrive . @Rage Set what is the highest gpu clock you've been able to achieve for benching? So far the highest Ive benched is 2076mhz.

 

I really like Primochill products. I am looking forward to your before/after results.

 

I've been benching it at 1940 to 1950. I've played around 2100 but for that frequency, I need a better cooler. However, I did not test the core without pushing the memory, so I might be able to achieve 2100+ with the stock cooler. I'm going to test it now.

 

Incoming... finally in stock at PPCS. I'll be on a business trip this week, so maybe it will be installed next weekend.

Hoping it will bring the temps down enough for running Cinebench at 5.3-5.4GHz, maybe CPU-Z validation at 5.5GHz. I guess we'll see.

https://www.performance-pcs.com/delidding-tools-accessories/der8auer-skylake-x-direct-die-frame.html

 

Tested at 2100 with an undervolt and it bounces around 1988 to 2100. So yes, this card can hit 2100+ with the right water cooler setup.

EDIT: On FS, I was able to keep it fairly steady at 2100. The funny thing is, without any memory OC, I was able to beat my last score by 600 points. I can only imagine how well this card will perform with better cooling.

https://www.3dmark.com/fs/18833886

 

Great stuff... I have the EK block on my 1080Ti and it's solid.

Happy Saturday to you as well. What has been your highest Graphics for Fire Strike? I'm trying to break 40500 on plain water and have been scratching the surface. Hopefully any day now. Same with Time Spy, hoping to hit 17000 on plain H20 in this custom loop...

I'm eager to fire up the x299 Dark board, but I'm not finished with the current desktop yet.

I might get some benching in this weekend we'll see. The RAM tuning has been a pain as usual lol.

Very exciting. I have a 9900K prototype bare die kit in route to me and it sounds very promising. I'll be doing some extensive testing and gathering data. Very much looking forward to it... I'll link everything once the item is available and live as it is not public yet. Looking forward to your skylake x data...

With the bare die kit I'm very optimistic that I'll be able to hit 5.5GHz and hopefully stable enough for some benching. Right now 5.4GHz on the non-delidded 9900K is holding pretty well for bench sessions...

 

A little over 41K is my highest using chilled water and shunt mod.

 

I will be looking at your results. I was curious about this solution before but I am afraid of applying too much mounting pressure.

 

As it is now, I use very little pressure. I just apply a modest amount of tension equally on the 4 springs. No need in cranking it down real hard as long as the contact is uniform and the pressure is firm. Direct die contact should also dramatically reduce the amount of liquid metal that needs to be used. I'll probably apply another coat of clear varnish to the surface mounted components, just for good measure. Maybe a layer of Kapton tape as well.

 

I got one of those a couple of weeks ago along with this https://www.asus.com/us/Motherboards/ROG-RAMPAGE-VI-EXTREME-OMEGA/ They are both going in Tuesday. Got the board for a great price; $625

 

Looks like a sweet mobo.

 

Nice board. Work machine? All them dimm slots populated could amount to more storage than what people have in their systems for storage lol. Good stuff...

 

Thats your best with the 9900K, right? I was able to match my watercooled 1080ti today.
https://hwbot.org/submission/4110646_

 

User Ictalley0109 used something similar (or maybe the same thing), and he got some temperature improvements, but only very small, but his testing wasn't that comparable as he was using normal thermal paste for the configuration that used the IHS, but was using liquid metal for the direct die contact configuration - so that was best case for the direct die contact configuration, yet only saw small improvements:
http://forum.notebookreview.com/thr...-lake-cpus-z390.811225/page-145#post-10882464
He also didn't seem to control or monitor his room temperature, and I got the impression he didn't preheat his loop until it reached steady state before doing the comparisons, so I don't think the temperature comparisons are particularly valid.

I replied in one of my posts (somewhere in the above link), theorizing that with direct die contact you would probably want to make sure that your water block was completely flat - I think flatness of the water block is more important when using direct die contact due to the small contact patch, whereas with an IHS involved it probably doesn't matter as much if your water block is not totally flat because the heat transfer area is widened & more spread out when using an IHS. What do you reckon to this idea? Have you checked your water block for flatness, I've seen reviewers place the edge of a metal ruler against blocks to look for gaps & unevenness in the surface? How about lapping the water block surface so it's perfectly flat (+mirror finish)?

 

When using an IHS with the waterblock we would apply LM to both surfaces. With the direct die contact do you think we need not apply it to the waterblock, only the die? Thank you.

 

I would put also on the waterblock to be sure Liquid Metal suck into the copper.

 

Yes, what @Papusan said. Always apply it to both contact surfaces.

 

Lapping the water block would be good. It will be interesting to play with. I would be doing the opposite and pre-chilling my loop (as I already do) by removing the radiators (easy with quick disconnect fittings) and letting the chiller circulate water through the CPU and GPU until they reach the lowest possible temperatures with the system turned off. The challenge with so many cores is moving heat off the CPU fast enough, and the IHS seems like the biggest impediment. It seems to act as an insulator under extreme conditions. Will have to be more cautious of condensation with a naked die and chilled water, since the IHS is also like an umbrella.

 

https://www.3dmark.com/fs/18695235

 

I was browsing HwBot Radeon firestrike scores, and noticed the guy ranked above me has an invalid score according to 3Dmark. It shows it on his screenshot. Does HwBot accept scores that are invalid? I was always under the impression that it needed to be a valid score to count?

 

Does it have a link to 3dmark?

 

Yes, it says something about tessellation load modified
https://www.3dmark.com/3dm/33632823

 

Then I suspect the answer is no. It's not valid. You can not mod the Tessellation of a benchmark, last I checked.

Edit
Seems you can on some benchmarks now.

 

Thanks for clarifying. I was wondering how he achieved a score more than 1100 points higher than me with the same hardware. Even @Rage Set with his 9900K barely beat this score. Speaking of Ragesets score, you need to fix it. For some reason it doesnt show up in the Radeon rankings. Your XE shows up at the top, but the 9900K is not on the list??

Sent from my SM-G960U using Tapatalk

 

Bro I'm just hoping I can use my windows 7 usb stick that I made with the apex and install 7 on the board. Shouldn't be a problem; both x299 chipsets and I saw one guy on HWBOT with an Omega running W7.

 

Yep, in terms of experimentation to compare before naked die with after naked die, then as long as you do the same thing then that's valid - ie you said you pre-chill to lowest possible temps & then run benchmark. As long as everything is the same then you compare before & after. If you lap your water block then do show some pics, mirror finishes are the ultimate photo opportunity! Problem is if you do before & after and the after includes lapping then it's not a true before & after comparison of the benefits of naked die, but that's not the end of the world you'd at least see be able to see the positive effects of both naked die & lapping together. To be honest, given the small contact area with a naked die, I think that goes hand in hand with making sure that everything is as flat as possible, to me it seems like there's no room for errors when you go down that route.

 

I probably won't do the lapping at first. I will run some Cinebench before installing the die frame at a "typical" overclock like 4.9 or 5.0, record the temps and then install the die frame. The repeat and record the temps running bare die. Then I will play around with max overclock capacity. And, then later on (some point in the future) see what improvement lapping the water block yields. If I do not see as much improvement as I hope to see with bare die, then I will probably be more motivated about trying the water block lapping sooner than later. If the bare die results are amazing, probably won't make it a huge priority.

 

No reason to believe it won't work exactly the same as it always had in the past. ASUS is smarter than that. Not supporting Windows 7 would ultimately mean the death of their success track record among people that love competitive benching.

 

Might have something to do with AMD drivers allowing it to be disabled without detection by 3DMark Systeminfo and NVIDIA drivers detect it with Systeminfo, which allows AMD GPUs to have inflated benchmark scores without being able to tell the mod has been applied. Not sure when it got changed on HWBOT, but the LOD/tess mod is valid for some benchmarks. 3DMark 11, Vantage, Fire Strike, Sky Diver are valid. For Time Spy and some others it is not. You have to look at the rules for each benchmark. I think this is a change from what the rules used to be. So, always check the rules. If you look at the scores for the people at the top of the food chain at HWBOT, their top scores usually have the LOD mod except where prohibited.

I only know this because I was trying to figure out why some people had much higher benchmark scores than I did with the same or lower clocks. When I started digging deeper, I found the rules allowed for it. Those that allow it look like crap when you are running them.

Fire Strike rules... https://hwbot.org/news/9039_application_52_rules/

Time Spy rules... https://hwbot.org/news/13722_application_198_rules/