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

You have already seen it but must have missed the short comments about it. This is where I got the 4000 CL14 settings that I used for the benchmarks I posted last night. I posted that video over here.

 

TM5 is nice. It is super useful and I use it frequently. The version @Prema tweaked (fixed bad grammar and typos) is even nicer.

 

Last night's benching spree was very fruitful. Had to wait for HWBOT to finish crunching the numbers so I could see the effect. I was #4 in the Enthusiast league for the US, but it's not taking long to catch up in the Apprentice League... started out like #90 or so a few weeks ago. So glad I got that chiller. What a difference a little bit of cold makes for the CPU and GPU. When brother @Johnksss came over last spring and brought his water chiller, it became clear very quickly what needed to happen next.

 

i read your post, but believe it or not: ive actually used tm5 and memtest86 to test my trefi timings worked just fine.

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

 

Memtest86 to test tREFI, ok, cool! You saw instability when tweaking tREFI too high on Memtest86+ then? (Or you mean this Memtest here? https://www.memtest86.com/index.html) That's a comfort if that is the case, because I don't like the idea of not being able to test tREFI beyond finding out if I have OS corruption in a couple of weeks!

 

Oh yeah, thanks, I did see that video, but skipped through his BIOS setting part. Yep, I found it now, it was in the "circle jerk of knowledge" section, ha! (that's his quote by the way!)

 

I dropped tRFC to 470 from 514 and tFAW to 30 from 31. Seems stable despite not seeing any gains with bandwidth or latency (AIDA64) (I blame Windows for that). Ran a few pass with original Memtest86 7.5. Not much parameters I can play with, and my notebook BIOS implentation for tREFI timing are buggy and honestly I dont want to touch it...My stock tREFI set as 11xxx for 32GB kit.

 

yes, correct. with trefi set too high i saw errors popping up in memtest. i meant this one: https://www.memtest.org/

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

 

Oh yep, that was (& is) my go to application for testing RAM overclock stability, I then use HCI Memtest after it passes Memtest86+, HCI catches some errors that slip through the net. I have found this one though, that looks like it does a lot of the same tests of Memtest86+ but with ability to use all physical cores of CPU (to speed up process), as well as a couple of new tests on there, Bit Fade Test & Hammer Test - both of which look very tRFC & tREFI related (they even say so in the description of Hammer Test) ( https://www.memtest86.com/). The Bit Fade Test is where it writes stuff to an address in RAM, holds it there doing nothing for 5 mins+, and then reads it again to see if it reads the same, I imagine that would be tREFI related, and you can also customise that test to have it hold it in RAM for as long as you like. I'm gonna run the Bit Fade Test & the Hammer Test on my RAM all night, with a custom setting on the length of holding time on the Bit Fade test to max out that holding time variable while still completing in say 10 hours. I only found it over the last couple of days, so I can't attest to how well it finds errors, but it 'looks' solid and very similar to Memtest86+.

I've increased tREFI from stock 12480, then to 16383, and now today 32767. I have tested a few runs of Aida, and it seems to increase Read speed by nearly 100MB/s, and maybe decreases latency by 0.x ns, but can't be sure as haven't done enough runs to account for significance. I know I had that weird program crash with tREFI maxed out at 65535, so halved that value for 32767.

 

ull have to manually enable multicore/multithreading on memtest86+, but it is possible to do so and speed up the testing

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

 

http://hwbot.org/submission/3975088_spektykles_xtu_core_i7_7700hq_1134_marks
Just beat Prema and top the first position with my notebook rig...whooo...
update: oh Prema just beat me again with his BCLK OC...well I beat him with stock clock thats something
update 2: http://hwbot.org/submission/3975295_spektykles_xtu_core_i7_7700hq_1150_marks
In your face, Prema
update 3: Im lost again, and this time I will tune the memory tightly to retake the crown as I dont have the fancy of BCLK OC and 4-slot of RAM to compete with Prema

 

Yep, I know about that, but forcing multi threading on Memtest86+ creates a bug where it will throw errors in Test #7 on pass #2 with multi threading enabled. I remember reading (I think from the developers) that this was a Skylake (& onwards?) related issue with forcing multi-threading on Memtest86+.

My Passmark Memtest86 (note not Memtest86+) run last night passed without problem for 11 hrs last night with the new tREFI of 32767. I might try putting my PC to sleep tonight, to see if it functions ok in the morning, supposedly another good test for tREFI. (I've never put the computer to sleep before though, so can't be sure that it returns from sleep without issue even at stock - sleep settings are notorious for bugs anyway).

 

You guys probably sick of me going on about tREFI, but I decided to go back to the stock Auto setting for this RAM overclock variable (12480, which is the JEDEC standard too). It was just bugging the hell out of me that I couldn't reliably test for errors that would come about due to tREFI being set too high. I know Jaybee was saying he detected instability when testing with Memtest86+, but by tREFI's nature I'm not confident it would catch all errors. The performance gain is arguably immeasurable due to the such small gains, and I can't be bothered to live with the potential OS corruption issues as the likely only & final indicator that it's not stable. Restored my OS partition using a Macrium Reflect image to a time before I was tinkering with tREFI as wanted to be sure to be free from corruption - feels good to be white as snow!

EDIT: tRFC is very tightly optimised on my RAM overclock, which is the charge time, and so this probably lessens the performance effect of tREFI tweaking (which is the length of time between charges).

 

How about this example...

 

That is an excellent example. Thanks for sharing it. CPU performance has very little influence on the overall score in those benchmarks.

Even so... dang, that 2080 Ti is a wicked GPU. Excuse me while I wipe the drool off my chin, LOL.

 

Just depends on the angle the user is coming from, but that 9900K setup still beat a 7980XE-1080TI and a 7960X-1080TI, while Talons 9900K-2080TI did not. Had he been up a few mhz in clock speed.....maybe the outcome becomes the same..... Maybe.

 

Yes, it does depend on the angle of the user. It only beat it in the graphics tests though. And, the graphics tests are what influence the overall score more than anything else in Fire Strike. That's fine for those of us that understand that, but it is misleading to those that do not know to break it down and look at graphics, physics and combined results. In the context of the overall score, Fire Strike is not a useful tool in measuring overall system performance. It is only useful for measuring graphics performance. Those of us that understand that are in the minority. If the metrics were equally weighted, the overall score would not be less on the more powerful systems.

There are a number of benchmarks that I had an overall score higher with my 8700K even though the CPU performance of the 8700K is obliterated by the 4960X. Sometimes the GPU functions more efficiently with fewer CPU cores and threads involved in the operation of the benchmark.

 

Hummm, hard to say. This argument as been around a very long time. That's why firestrike is known as a gpu benchmark and not a cpu one or and overall one.

Just the basics...
Firestrike is a gpu benchmark influenced by cpu speed
01/03/05/06 & Vantage are all cpu benchmarks.
3dmark11 can be ran with 1 core and basically get the same gpu score.
Skydiver is a cpu benchmark.
Cloudgate is cpu benchmark.
Time spy is a gpu benchmark.
Aquamark is a cpu benchmark.

No one test to just some it all up. To many ways to argue.

 

[GUIDE] 3DMARK SCORE CALCULATION - HOW TO CALCULATED YOUR 3DMARK SCORE rog.asus.com

http://akamai-dl.futuremark.com.akamaized.net/3dmark-technical-guide.pdf

 

Thanks, that's cool. Did you see the 3DMark calculator tool the same guy made in this thread?

https://rog.asus.com/forum/showthre...-for-3D-Benchmark-Tests-Calculate-your-Scores

Could be helpful in figuring out where to burn the most calories for a higher overall score.

 

That makes total sense, where they say in that guide the following with regard to how the overall score is calculated from the Graphics & Physics scores:

"For a system where either the Graphics or Physics score is substantially higher than the other, the harmonic mean rewards boosting the lower score. This reflects the reality of the user experience. For example, doubling the CPU speed in a system with an entry-level graphics card doesn't help much in games since the system is already limited by the GPU. Likewise for a system with a high-end graphics card paired with an underpowered CPU."

I agree with that philosophy, plus you have the individual scores for the Graphics & CPU subsystems anyway, so still a meaningful way to compare CPU & GPU specifics.

EDIT: and here's my Timespy Score I did this morning ( http://www.3dmark.com/spy/4964778) (Spectre fixes not disabled). It's pretty well balanced between GPU & CPU, but could do with a bit more CPU, which rings true when I think of the Amiens map in BF1 where I'm CPU limited (albeit at 125-144fps). But also, a bit GPU limited on BF1 as not running maxed out on I think two of the graphics settings.

EDIT #2: You might notice in my Timespy link that I've got my CPU at 4.65Ghz now - that's a result of some successful BLCK fine tuning I did yesterday, up from 4.6Ghz to 4.65Ghz. As an added bonus it speeds up the RAM by about 1% in speed, giving about 250MB/s increase in Read speed in Aida64, as well as a drop in RAM latency by about 1 ns, I'm at 41ns latency in Aida64 now. Ha, this sure trumps the ridiculously ineffective tREFI experiements I was trying a few days ago! (New CPU & RAM overclock tested in Prime95 v26.6, x264 custom loop overnight, and HCI Memtest 600% pass.)

 

Just picked this PSU up for $149.99 because why not?

https://www.evga.com/products/product.aspx?pn=220-T2-1600-X1

 

Posted about this late Mars this year. Now we are soon middle of Nov. The days have passed but it is not far away that the internet fiber cable rolls past my house now Seems I can get it within 1 month. Yeah, it will be a lot easier with a working internet.

 

were actually super happy with our 200 Mbit cable connection here, but our ISP decided to switch to fibre exclusively, so now were gonna be "forced" to upgrade to a 500 MBit fibre connection soon i figured going straight to 1 GBit or even 2/5/10 GBit would just be silly

 

Too bad my ISP throttled the fibre connection from advertised 10Mbps to 10Kbps-50Kbps and I just disconnected it since they couldn't resolve my issue even after I hinted them what to do.

 

OK, I finally decided to order a set of these. Should be here sometime today. I hate spending money on stuff like this, but the stock cables are just too rigid and unwieldy to submit to my will for cable management. I went with all black because there are too many nice color options available and I know I will get sick of looking at the same color every day if I choose one. @Johnksss is this the same kit you bought (not counting the color option)?

 

Nice! Let me know how you like them, I recently did the "cable extensions" aka fake cable mods, and they work/look great but now I have to manage the regular cable and the extesions so no bueno. I used my EVGA discount and got the Titanium 1600w PSU coming this week and might just go full stop on new PSU and nice sleeved cables.

 

Hey @Mr. Fox . How you doing Bro?

 

Yes, that is the one I ordered.

 

I will let you know later tonight. I got them, but I haven't opened the box yet.

Cool, glad I got the right set.

 

OK, I replaced all of my EVGA cables with these and they are great. Very good quality, and they are much easier to manage than the stock EVGA cables. The free cable combs Amazon included are not very good. They are too loose and do not grab the cables well, and I ended up stacking them together to help stop them from slipping. That part doesn't look great. But, hey... they were free so I'm not going to be too critical. They're better than nothing. I will get some better combs later that fit tighter on the thinner cables.

 

I need an advice here: Should I pick Titan Xp Collector Edition or RTX 2080 (EVGA XC Ultra)? Both priced at 850$ here. Well my current games doesnt take advantage of all the AI buzzwords but I need it as future-proof as possible. Raw power the Titan Xp has the advantage but its true that driver optimizations in the future will make 2080 an obvious choice over the top Pascal card?

 

I think the last comment you made is your best answer. True statement, but focus on the word "future" to keep things in perspective today. The 2080 has nothing on the 1080 Ti in terms of performance except ray tracing and deep learning, but neither of those things mean too much right now. If you have a 1080 Ti already, don't waste money on a 2080.

 

If I had to build a system now I'd probably put a good 3rd party version of the RTX 2070 in there, but I wouldn't build one now with this gen, and 1080ti/TitanXp is old school now for a new build. In your case I'd choose the 2080 over the 1080ti and also over the TitanXp if they're the same price. RTX 2080 performs better than 1080ti in most cases (not by much) plus it has ray tracing/DLSS & is more future proof. (See here for comparisons against 1080ti & other cards: https://www.guru3d.com/articles_pages/geforce_rtx_2080_founders_review,39.html)

 

Still depends on what GPU he has now (if any). There would no point in spending money on a 2070 downgrade or a 2080 sidegrade if he has a 1080 Ti and it would make more sense to do nothing. If and when ray tracing and DLSS actually become important and meaningful one can spend money (and possibly a couple hundred less of it) when that day comes. It may turn out to be the plastic carrot dangling at the end of a stick like DX12 was to sucker game-crazy dumb-dumbs into accepting Windoze OS X. NVIDIA is pretty famous for inventing awesome things that never receive the broad acceptance they deserve. PhysX and SLI are two examples of awesome things that truly deserved to do better and never did. Ray tracing and DLSS may turn out the same way. Too early for anyone to say with any level of accuracy right now. But, I sure as hell wouldn't rush out and blow a wad of huge cash on pie in the sky technology that may not materialize before the overpriced GPU becomes obsolete, if ever. The only valid reason to upgrade from a 1080 Ti to 2080 Ti right now is for overclocked benching. It's a total waste of money for gaming to spend over $1,200 for a baby boost in performance anywhere from zero to 20 FPS depending on the game title. That is, of course, unless you are hell bent on playing games in 4K for some reason. Rumor has it that 2080 Ti does a great job of that, which is a first. Until now that has been pretty much a slide show proposition unless you turn off all of the eye candy.

 

@

@Mr. Fox in the picture on the lower right corner a tiny bit of LM got on it and used alcohol to remove it. I subsequently sold it and the person I sold it told me he is getting code 53. Did I ruin the cpu? I told him to check his socket pins as I think that's a memory code. He said he struggled to get his water block off; it was like it was glued on. The chip is a 7980xe and he has an x299 board. Thanks. Sorry but I guess being a newbie I can not post pictures.

 

I will send the pic to your email.

 

https://www.3dmark.com/compare/spy/4965024/spy/5005158/spy/4968288

https://www.3dmark.com/compare/spy/4945324/spy/4998591/spy/4968407

https://www.3dmark.com/compare/fs/16916824/fs/17012571/fs/16975571

 

Yeah, I agree, I wouldn't buy a Turing card now, I'd wait for the next generation. I don't know what card he has now, if he already has a 1080ti then it's absolutely not worth buying a 2080 (and not really a 2080ti). If a person had no PC now and was building a new one, I'd probably build one based around an RTX 2070 as they're reasonable value (just) and pretty good performance with the added bonus of being new tech allowing for raytracing, etc. But, if someone has a good enough GPU for now, then I wouldn't recommend buying any of the RTX cards full stop, because I see the next generation of NVidia cards after Turing to be the ones to get due to die shrink & better ray tracing performance (if it has turned out that ray tracing is a thing). That's why I'm probably gonna hang onto my GTX 1070 until the next gen following Turing.

He mentioned spending the same amount of money on an RTX 2080 or a Titan Xp, and if he's dead set on getting one or the other (regardless of which GPU he's got now), then I'd choose the RTX 2080.

 

Yes, agree, if one were starting from scratch on a new build I would buy new tech as well. I would not buy an obsolete GeForce GPU or any Radeon GPU. But, I would also recommend waiting if at all possible and only buy one of the high-end GPUs built with the intent of overclocking even if one did not intend to overclock just for the sake of durability and resale value. Unfortunately, the best built GPUs are almost impossible to find now. You can only find the ordinary reference design and FE crap right now.

The exception would be if you have some kind of "emergency" and have to buy now, then I would get the cheapest GPU available just to tide one over until things change with the Turing GPUs. Like a low-end 1060 or an RX 580 used on eBay in the $150 price range and run the guts out of that until you can get something special and the all the new GPU bugs are worked out.

 

I seriously doubt that what is showing in that photo would cause any harm whatsoever. There was liquid metal literally all over the 7960X when I got it that I had to clean off. It should have been cleaned up and all liquid metal removed from the exterior of the CPU prior to shipping, but liquid metal can only cause harm if it creates a short circuit. Even then it's not a given. It depends on what circuit gets shorted. It may not cause any permanent damage depending on the circumstances.

If it had liquid metal on the contacts and the guy installed it that way without cleaning it up first, then he is not very smart and that is not your fault. It would be his fault for not paying attention if he just slapped in into the mobo without checking anything and it caused a problem. If he did clean it up first and what is shown in the photo is exactly how it looked when he installed it, then I think it is totally a non-issue.

But, you are correct about Q-code 53 if it is an ASUS X299 board. Codes in the 50-55 range are memory initialization errors and he should either clear the CMOS to set BIOS defaults or try different RAM modules. That doesn't seem like a CPU problem, but it could be an issue if liquid metal is bridging some of the pins or he bent some pins installing the CPU.

 

Thank you @Mr. Fox " There was liquid metal literally all over the 7960X when I got it that I had to clean off." Sorry about that.

 

No worries on that. It was not an issue for me. It was merely a minor inconvenience that required 10 minutes to resolve. I used some sterile alcohol pads and an old toothbrush with alcohol to clean it up and it was as good as new. But, it is relevant to your question and it's not very likely there is anything wrong with the CPU you are asking about. Excluding physical damage, it takes something pretty serious to cause permanent harm an Intel CPU.

 

Now the guy is telling me he got some MX-4 on the socket pins. Should he try to remove it with a Q tip and some alcohol? I told him to be really careful if he does that as the LGA 2066 pins bend if you breathe on them.

 

No, never use a Q-tip on the CPU socket. The cotton fibers can snag the pins and bend them, causing new problems, or they can leave behind cotton fibers that are snagged on the pins even if they do not get bent. MX-4 is not going to short anything and probably won't cause any harm being on the pins. People that use sub-zero cooling put things like Vaseline in the CPU socket.

I would unplug the PSU from the wall and pour alcohol into the socket, then use a very soft bristled paint brush (a small one like the size of the brush that comes with CLU, or like what kids use for water coloring) to clean it up if he is worried about it. If you are really gentle and only allow the bristles (not the handle of the brush) to come into contact with the pins they should not get bent. Still need to be gentle in case one of the bristles somehow snags on one of the pins while cleaning.

I would let it dry for at least an hour to be sure the socket has no moisture before putting the CPU back into the socket. Even though it won't hurt anything with non-conductive paste on the pins, it would be easier to clean up now while it is soft and gooey before it gets dried out and hardened. I would be surprised if the MX-4 on the pins has anything to do with his code 53 issue.

 

After having 2 2080 Ti fail on me in less than one month I considered full return and just cutting my losses this gen and grabbing a $499 2070. This however realizing that a used 1080 Ti would be the best option.

EVGA tech convinced me that it was just bad luck and said they would send me a brand new replacement. This was last week and I’m hoping it ships early this week.

 

Well, because of all the chatter about it it certainly feels like there is lots of bad luck going around. According to the AIB vendors and system builders (including der8uer/CaseKing) that are willing to speak about it, less than 2% of the 2080 Ti GPUs and complete systems sold with 2080 Ti have been returned due to GPU failures. The failure percentage seems to be very low statistically, but it is still about twice more than what we are used to seeing with NVIDIA GPUs. Failure of NVIDIA GPUs has been historically rare, and based on the numbers reported it is still rare.

I am glad you are having EVGA send you a replacement and you are sticking with 2080 Ti. I think that is smart since you have already spent the money. I doubt you would be happy with a 2070. (I know I would not.) EVGA is a great company and even if the one they send you fails they will replace it again, as many times as it takes for you to get a good one. @Talon are they doing an advance RMA, cross-shipping, or do you have to ship the bad GPU in to them first?

 

They are cross shipping the GPU to me, so advanced RMA. Their support has been great. The tech I spoke with on the phone followed up with me later that day with more things to try. Ultimately they didn’t work and after I exhausted ideas I concluded it’s hardware failure.

The card now artifacts on the desktop with mouse movements. I tried everything I could think of including moving the GPU to a different slot.

 

Yeah my previous GPU was stolen (sent it for repairs, lost on the road) and now Im using RX 560 as an emergency card. So maybe I will pick RTX 2080 unless the owner of an used Titan Xp I talked with drop some price more.

 

RX2XXX will be better at keeping value over time, but RX2XXX are dying like flies in a fire at this moment, maybe wait for now...?