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

Or

We all were the beta testers and they finally came to their sense and said....Let them have what they been asking for. Now. If that were there truth, then I would really like to do them in at that point!

Also this price is suppose to be higher than the original 2080 TI, well that cost was $1250 and up to begin with. Averaging around $1400 with no water block and custom pcb.

Even Nvidia is selling the founders edition for $1200 as of today's date.

So it would also seem like some will be trying to rip the vbios from this new super card and testing it out on theirs if things are really that different.

I guess we shall see next week....

 

Yup, it will be interesting to find out.

If there is a 2080 Ti Super, I'll definitely cross flash it to see if it works. The K|INGP|N vBIOS does, which is surprising. (Wasn't that a first?) If the 2080 Ti Super vBIOS cross flashes with no brick and unlocks the same performance then it will be a non-issue. And, that would be kind of typical for NVIDIA, since they do more smoke and mirrors with firmware and drivers than they do with hardware.

The cost is one thing. Not great, but we know and accept that... it's part of the hobby. But, the treachery aspect of it is what I find disturbing. Treating customers right is important, and that just ain't right. But, if a firmware cross-flash levels the playing field and magically makes it a "Super" as far as overclocking and benchmarks are concerned, then I'll not care any more and it won't matter.

 

True.That is a first! The earlier versions of the K|INGP|N vBIOS versions never worked right on any other card other than itself.

Yep, talking about the treachery part. They are good for that one. That's for sure, but at the same time...They are good for the hype and fizzle in real life performance come launch day. That's usually another 3 to 6 month journey in to a performance increase...

Also i'm already use to the pricing nonsense...Although they did come down for a month or so, but seems that they are going right back up.

 

They will eventually slit their own throats on the pricing. If it continues on the current path it will become a situation that I just can't do it any more. And, that applies to a lot of people. Things are getting to the point that pursuing the hobby amounts to utter foolishness unless you're so wealthy that money is no object. It has always been expensive, but the cost now is unprecedented and reaching the point of being unacceptable.

That custom PCB and no available water block situation you mentioned in a previous post is a real pisser. No point in owning it if you have to run an air cooled throttle-meister, chintzy AIO, or have to cobble together a makeshift /homemade water cooling solution. That sucks. If it's super expensive it should also be super nice.

 

That's why people bench old hardware... The more people in a category the easier to obtain points. So using your cpu on older hardware would net far more than you just benching your 2080 ti. Why? Because those points are far harder to lose.

 

If I had to guess, probably the CPU and chipset. Those are pretty poor numbers. The latency is atrocious, too. Maybe @ajc9988 can provide some pointers for cleaning those numbers up, since he is one of the only AMD CPU overclockers in our community.

 

I have very high reservations against Nvidia releasing a new 2080 TI Super, when they are obviously looking to a new process node. Nvidia has refreshed their cards but not their top consumer card so soon into its life cycle.

Then again, this is the first time they released a TI variant during the launch of a new product line. Perhaps, they know something about AMD that we don't.

 

I'll be able to work with you a bit on it, but Zen is a different animal than Intel, and your numbers will look a bit worse comparatively. Here is mine in sub-optimal conditions of hundreds of tabs open, word, excel, and snipping tool open:

That is at 4.2GHz (ignore it reading 4.1GHz, it is an error due to having virtualization turned on for the CPU in Windows and the BIOS, it reads correctly when turned off) with memory at 3466@CL14-15-15, single rank DIMMs (which are easier for higher speeds).

I may not be able to help until this evening, tomorrow, or this weekend, though.

Also, for cache speeds, because of two dies, mine are doubled, same with the memory speeds (since quad channel instead of dual channel on the mainstream platform).

What work have you already tried on OCing the ram? Can you post an image of Ryzen Timing Checker (RTC) written by the Stilt and downloadable from techpowerup?

 

Did you see the change from a 64 Wave to 32 wave and increasing the SIMD from 16 to 32 so that now for WAVE 32 it takes one cycle to process instead of 2 cycles and instead of Wave 64 taking 4 cycles, it finishes in 2 cycles?

They also did a large reworking of the cache structure, which should help and is why increasing the bandwidth on prior cards helped so much to increase performance. Here is a video discussing some of that from GN with David Canter:

 

Sometimes user error paints a gloomier picture than necessary. I wouldn't write it off too early. When I said "probably the CPU and chipset" that was not to imply they were no good as much as the fact that they are different. Poor numbers are frequently a reflection of user error or lack of knowledge about what to do to fix them.

I hope your reservations are correct. I share them. But, again... if the vBIOS will cross-flash and produce basically the same results I won't care all that much. There's no loss in that scenario as far as I am concerned. But, the element of dirty pool is still there. And, some of the purists and worry worts won't view cross-flashing firmware as being acceptable.

Awesome. If anyone can help, I knew it would be you. Thanks for jumping in.

 

Your sub timings do not line up with your standard timings.

 

Interesting video. One of the statements David Canter made that I really like, although taken somewhat out of context in my application of it, was (at 21:43) :
"If you're making things for consoles and primarily desktop PCs power is not necessarily the most critical thing."

I just hope that brute PC performance remains the primary driving factor and they don't get too caught up in the "following the money" nonsense so to speak. The focus on reducing power to accommodate mobile garbage might help mobile, but not PC. The inordinate focus on mobile garbage has become the bane of our existence in PC enthusiast world. I prefer the concept of doing more with more, not more with less. The latter smacks of compromise, (which sucks,) while the former leaves no room for compromise... power consumption be damned. If it uses lots of power, so be it. So what. Just give us brute performance and let the power numbers fall where they may. Who cares about mobile. (Just my opinion.) But, they'll probably follow the money because that's what they have to care about more than anything else. No money, no product. If they allow the tail to wag the dog by partnering with Samsung, it might result in more mediocrity in the PC landscape only for the sake of accommodating the advancement of mobile crap.

 

If you are referring to the tRAS, 34 does seem really high for CL14. I would be thinking that should fall somewhere in the 20's. But, I am not sure what works for Ryzen.

Yeah, that is certainly moving things in the right direction. Major improvement. Good job.

I don't know if this will work for Ryzen or not, but here are the timings I use for 4000 CL15 if you want to try any of them to see if they work better or worse.

 

That is one of the 7 or so others that need to be adjusted accordingly. Just setting 14-14-14 can be worse than running 16-16-16

This runs better on Z390 than it does on X299. No way I can pull of this on there.

 

Looks pretty serious for the new refresh/upgrades to the 20X0 Nvidia cards:

https://www.notebookcheck.net/Nvidi...ew-RTX-2080-Ti-and-2070-Ti-GPUs.423799.0.html

 

The backtracking begins. There likely will be refreshes of the 2060 through the 2080, to beat Navi but nothing on top. We will see a successor to the 2080 TI, but it won't be this year.

 

Some of those timings would hurt performance on Zen.

For example, here is my RTC (ignore the settings at the top, RTC doesn't read TR correctly for those after a couple AGESA and BIOS updates happened):


So, in his example, he should get tRAS as 28, tRC to 42. Now, for tRRDS and tRRDL, you need to move those in tandum, keeping 2-3 difference on them.

For Ryzen, you really want to try to keep tFAW and tRFC as low as possible, but that can take some hunting to find stable at higher frequencies.

For tCWL (CAS Write Latency), you need to keep this at parity with the CL latency most times, unlike on Intel where you can run it tighter than the CL.

For, tRDRD SCL and tWRWR SCL, having those tighter will complete mem testing faster, but effects on benchmarks are minimal, so if you are having trouble with getting stability, that is a place to loosen by 1-2, depending on your speed used. But keep them together and change them in unison when changing timings.

The timings for tRTP to tRDRD DD kind of work in unison. You can sort of play with tRTP, tRDWR, tWRRD and tWRWRSC one at a time in that, but they are connected, so that changing one means you may need to adjust the others to find stability on Zen. tWRWR SC and SD and tRDRD SD and DD are pairs. Treat them as such.

For tCKE, if you cannot do 1, it is OK. It can make it easier to get stable with some ram types using 4, 6, or 9 (6 & 9 are more common).

Then, if needed, dropping to CR2 helps. If really having trouble, you can enable Gear Down Mode, but try to keep it off if you can. For Bank Group Swap and Alt, using disabled and enabled, respectively, helps with performance/stability. If you turn the Alt off, it will be harder to get stable.

For RTT and drive settings for resistance, that varies by memory and WILL take a LOT of testing to find optimal for higher speeds. The ProcODT resistance also varies with the ram and SR versus DR on what will work to drive your sticks.

Mem OC on Zen is a different beast, so to speak. And, as I mentioned, this program isn't really reading my RTT or DRV strength settings correctly from the BIOS after a specific update. So do not copy those anyone! Also, RTT and drive strength are different on DR than SR, and I have B-die SR DIMMs. So that wouldn't quite help anyways.

After a BIOS update, I couldn't get 3600 stable anymore (I spent a week or two trying). But, my 3466 timings were enough to get about the same performance, roughly (before that update, I couldn't get 3466 stable if I tried, which didn't matter because I had 3600MHz).

Also, the DRAM calculator is more guidance and doesn't work for everybody. You can use it as a starting point, but then trying to manipulate it afterwords for tuning is recommended. Here is a techpowerup article from the calculator's author that can help.
https://www.techpowerup.com/reviews/AMD/Ryzen_Memory_Tweaking_Overclocking_Guide/

Edit: Also, for benchmarking with Win 10 on any Zen, you need to use HPET. But, even in your normal use, the RTC bug makes the 102 BCLK read as if the CPU is running slower.

As such, it would be best to get the biggest OC on multiplier and ram through changing the timings from the XMP settings to get better performance on the platform. That is one reason I don't use BCLK on the Zen platform. But, since you have more fine control of the multiplier (0.25 increments), you can roughly optimize it without needing the BCLK which can effect your I/O devices a bit (it doesn't separate as clearly as Intel does and above a certain BCLK, I think 104, your PCIe gen 3 switches down to gen 2 or 1, which there goes your GPU performance and NVMe drive performance).

Now, some of this is changed with Zen 2. I'll be keeping an eye out, but cannot play with it until the new threadripper cpus drop.

 

"The Road to 16K" is a rocky one.

On the way we have just beaten the fastest 9900K 1080 SLI Desktop Combo in TS with a P870:

https://www.3dmark.com/compare/spy/7438595/spy/6333105

It's rather what they did not do, being releasing the GT76 when they still had a chance to say most ODMs favorite "we are the fastest" catchphrase.

There will be no cross-flashing with 'Super' cards...all cores are different.

 

Awesome job on the TS score.

Well, if they release a 2080 Ti Super (which they may not) I may sell my 2080 Ti and use the money to buy three or four crappy old AMD GPUs to rack up hardware points and say goodbye to the Green Goblin for good. Kind of sucks because I have had nothing but bad experiences with AMD GPUs and it's really hard for me to get excited about their lackluster graphics cards. I don't think that Navi is going to offer anything compelling except in the low end and mid-range market.

 

Ah, so I see.

 

I am still amazed about the clocks we can run with a GTX in PR.
Here another PR in PR, yet still failing the 3500K mark:

https://www.3dmark.com/pr/102811

 

Way to put the smack down on the competition!

 

Well, or to look really stupid, because those 1080 SLI at 2150Mhz can't even come close to a single 2060:

https://www.3dmark.com/compare/pr/102811/pr/79888

 

So possibly something along these lines in performance.
1080 ti 17 fps.

 

Sounds more like another buggy 3DMark application rather than a hardware issue. Seems they (UL/Futuremark) just keep getting suckier and suckier. Their decline in product integrity start quite some time ago, when we were all still friends with folks at Tech|Inferno. An unfortunate sign of the times... it seems there are more things that suck than rock in PC world, especially where software is concerned.

 

Not sure, but something is severely underutilized when running Ray Tracing on a GTX.
These cards hit 2113-2126Mhz region max in all other benches, while running 2151Mhz only in Port Royal.

One would expect RT to generate a higher load on the core without the dedicated hardware of an RTX, but this shows quite the opposite.
Also not sure if they are even interested in showing a GTX real RT potential...

 

Yes, I think the new/added hardware NVIDIA included in the RTX cards for DLSS and ray tracing is very real and extremely useful under the right circumstances, even though folks put off by the absolutely ludicrous pricing don't want to admit it. Those features don't do anything special for software that doesn't use DLSS or ray tracing, so they have a point in that respect. A 2080 Ti is only a minor upgrade over 1080 Ti (mainly GDDR6) for anything that doesn't use the RTX features.

Probably something added to the hardware monitoring elements that artificially cripples results when it doesn't find the "special" stuff the benchmark was designed to showcase. They are probably paid an extra premium by Micro$lop and NVIDIA to block older OSes and cripple benchmark scores on select components of their benchmark suite.

 

There is one other benchmark that allows for high clocks but most dont run it.

That's what i'm seeing. If i wasn't benching I would not have bought a 2080 anything. As ray tracing really doesn't mean anything to me. All these games look cool on ultra high and all, but nowhere near perfect. People tend to over hype it because they bought into it.

 

For some reason I don't like Ice Storm.
I can't convince myself to enjoy a bench that has the same result for a 8/16 (core/thread) and a 1/1 run...even phones can do better than that these days.

 

Just a quick aside for any of you folks that run a high refresh rate TN panel monitor (e.g. 144Hz) - I've found that my contrast ratio doubles if I decrease the refresh rate from 144Hz to 100Hz, and I think it triples going from 180Hz down to 100Hz. So, as a result I run all normal desktop stuff (and games where I don't want more than 100fps) at 100Hz so I can get markedly better picture quality, and then run something like BF1 at 180Hz where speed is more important than image quality. Just a thought for you folks - it helps get the most out of your high refresh rate TN panel, I'm assuming the reduction in contrast at high refresh rates will be a thing for all TN panel monitors (I measured contrast ratio with my Spyder4Pro colour calibrator). Running my monitor any lower than 100Hz doesn't increase contrast ratio any further (well only by a tiny amount so not worth it).

 

Yeah, for anything other than benching and ray tracing, RTX cards are worthless. The ray tracing looks great and it might become the way of the future in gaming if NVIDIA can manipulate the game developer community enough to make it so, but honestly... we don't need it. Don't know why anyone should care. Gaming was already good enough without it. Photo-realism isn't required and doesn't make a boring game fun or a fun game boring. It's all just a creative mechanism to generate a new revenue stream to feed the greed machine.

Have you also noticed that running a high refresh rate contributes to 3DMark instability with a high GPU overclock? Sometimes when I cannot get a Time Spy or Fire Strike run to finish without crashing, dropping my refresh rate from 165Hz to 60Hz stops it from exiting to the desktop with the stupid "Ooops an error occurred" message.

Yes, Ice Storm is kind of hokey. I've never really cared for it. I'm just about to the point that I don't like anything in the 3DMark suite other than Sky Diver any more. It's all buggy garbage. 3DMark 11 and Vantage are better products. Sky Diver is the only truly reliable and stable component in the 3DMark suite. Fire Strike is fairly decent. The rest of it is rubbish.

 

Ice storm is a cpu benchmark more than it would be considered a gpu one... So the faster the cpu the better the score...
No gpu over clock... Different version of 3dmark also.
https://www.3dmark.com/compare/is/4477137/is/4495648#

 

Ray Tracing is great for the future of VR/AR.

Yes if we bump the single core Turbo as it doesn't utilize many cores/threads properly at all. Ran it twice once and moved on...

Edit: Yes, Epic Fail on that one:

https://www.3dmark.com/compare/is/4477137/is/4495648#

 

I've not noticed that, but I've not tried it, useful to know for benchers - interesting phenomenon!

 

This rubbish has no place in a PC benchmark. They shouldn't be lacing their software with this kind of worthless feces, and I have lost a lot of respect for them because they have. Separate it and put that garbage up for download on Google Play. Cloud Gate is also iffy, and I don't know why anyone would care about running benchmarks on the products Cloud Gate is made for. Ice Storm just needs to be removed... it's cah-cah.

 

No, not an epic fail at all. Just a different tactic for a different benchmark. As Brother @Mr. Fox mentions.... Those benchmarks are really for mobile devices which run mainly on cpu, with the help of a gpu. Not the other way around.

 

Of course an Epic Fail for me bro. Running those high clocks and getting wiped by your stock score. It's hilarious.

 

If it was a legit benchmark it would not say your HTC U11 is "too powerful" for this test. It would just dutifully crunch the numbers without complaining and spit out a proper score based on the numbers it crunched.

 

I really wish EVGA included a single slot bracket for the Hydro Copper blocks. The blocks are installed and waiting on the last parts to arrive for the loop.

 

Question, why are you using 37.75 multiplier for CPU? Limitation of your chip? Also, I'm not fully familiar with PBO since gen 1 didn't have it. But isn't there where you can take the limiter off of it and let it boost to your cooling solution?

If you want to get your latency down and raise the cache rates on that test, you are going to have to get your core up. Now, if you have your core OC already but are using lower to try for higher mem OC, then plan on combining them when you find the limits of your ram, that's cool. Step by step and very common way of doing it.

Either way, you def got the mem performance up! Great job there! Can't wait to see your 3466, 3600, and 3733 attempts (you get that last one, you'll be in an elite group on Zen 2, and 3866 is like the top cream and a LOT of luck on your IMC being good). But sometimes you have to loosen timings so much for 3600 and above to be stable on that platform that it is worse than 3466 and tighter timings.

I just hate seeing a zen CPU at under 3.9GHz.

 

16K TS check:

https://www.3dmark.com/spy/7438595

I hope the FS and TS sample runs we provided to the hwbot team are enough to convince them to let everyone bench their mobile cards against Desktops and get some additional points.

 

I knew non-X didn't have XFR, but no PBO. ...

In any case, the CPU is unlocked. Let me know if you change your mind on core OC if you need any assistance.

 

@Mr. Fox @Johnksss @Prema @Rage Set @Papusan @electrosoft

Regarding the NVIDIA "Super" cards... It's crazy to think how well a shunted 2080Ti FTW3 can do in being within striking range of the KP cards... now if the "supers," get the same quality Samsung mem and a stronger core, then it'll be even closer to the KP's ... The "super," cards are suppose to have more cuda core as well eh?

Also, 9900KF in route...looking forward to testing that... the reviews are a mixed bag, but accordingly to some data it should be able to OC higher at a lower voltage while producing less heat vs the 9900K ... This should cater well to XOC stuff... who know's but we'll see...

 

FWIU it's the R0 stepping that's better than the P0 9900K's not the KF per se. R0 includes new stock of 9900K as well as KF and presumably the KS that is coming which will be all the top bins Intel have been harvesting for however long now.

Silicon Lottery show slightly worse stats according to their binning process for 9900KF against the K...... 31% vs 35% for 5.0ghz, and 4% vs 7% for 5.1ghz (at 1.300V AVX-2)
https://siliconlottery.com/collections/all

 

You mean R0 stepping is better due the inbuilt security patches? But what if you don't want them and rather want max performance?


http://forum.notebookreview.com/thr...ers-welcome-too.810490/page-581#post-10915447

http://forum.notebookreview.com/thr...50h-coffee-lake.810891/page-130#post-10913616

 

Hmm interesting. I'm aware of the KS variants, but contrary to SL's data, there are those who have reported the KF to OC better in terms of thermals at lower voltages according to their tests. It's all with a grain of salt until we can get more data, but even if the KF is the K with the igpu disabled, it is possible that there could be some benefits. I'm really looking forward to getting my hands on the KS bins..that should be interesting..

 

Slowly but surely... despite being busy with life, it's coming along. Seeing all you studs with those monster physics scores inspired me to just change course a bit.

I still plan to play ball in the 9900K class though. IMO it's one of the special classes due to it being a common cpu that the avg. person has (also marked as the best gaming cpu), so it's nice to showcase some respectable numbers with that chip...

 

I do like that increased core speed there! Definitely seeing it on latency reductions.

Now, with BGS enabled, although you may see a bit in peak like in AIDA64, you may want to check performance in programs. Here is an old chart published by AMD when trying to help customers optimize ram settings after Zen 1 dropped (things may have changed since then with AGESA and BIOS optimizations, but to understand where the recommendations came from):

https://www.extremetech.com/computi...os-updates-improves-performance-adds-features

 

Good morning fellas,

So I spent the better half of last night working on setting up the 2nd test bench for the Z390 Dark board, so that I don't have to swap mobo's in the future now that the X299 is in the picture... that'd be tedious. As much as I wanted to bench, the responsible thing to do was to chip away at the projects lol...

"Ain't nobody got time to be counting calories... I mean dB's ... bring the noise on!" LOL...


These things push some serious air...


Push/Pull Config on that monster Nemesis GTR 360 Rad...


I've been waiting to test out my new RAM 4x 8GB sticks (G.Skill 4600MHz Royal Z) which I had sitting on the shelf waiting to be installed on the Z390 and now X299 test bench...

For giggles I installed the 1080 Ti w/ a stock 8700K (not yet delidded and surprised to even get 52x out of it without trying...) + The 4600 sticks (2x) .. All I did was just load the XMP (1.45v I think) and ran some fun runs of Fire Strike...

4600 posted?! Crazy eh? Dang these Samsung B-Dies ... talk about strong. Now imagine these things tuned..


Here are some uploaded runs to show the RAM info at 4600MHz:

https://www.3dmark.com/fs/19600439
https://www.3dmark.com/fs/19600447

I'm really looking forward to getting some real benches in... it feels good to be at it again...

@Mr. Fox @Prema @Johnksss @JoeT44 @Papusan @jaybee83 @Robbo99999 @Rage Set @electrosoft @Convel

 

Nice, you definitely change hardware like underwear it feels, ha! Is the physics score a little low for 6 hyperthreaded cores at 5.2Ghz? Or Firestrike not make full use of 6 cores?