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

moar CB15 points!

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

 

https://www.guru3d.com/articles-pages/amd-radeon-vii-16-gb-review,29.html
Guru3D had a hard time overclocking the Radeon VII, they've updated their previous review with new overclocking results, they got about 6% performance increase compared to stock, not very good gains. They tried undervolting but didn't get any performance increase from doing that. I'm not sure they know what they're doing as much as some other sites (like GamersNexus) when it comes to overclocking and taking a 'scientific' approach to it. I've not looked for other overclocking reviews, so this might not be the best that's been achieved so far.

 

Interesting they did not mention the temps after undervolting. Undervolting is not really a performance enhancing technique. It's a temperature reducing technique. That suggests they may not know that much about overclocking, as you indicated. They totally missed the point of undervolting.

 

Like how Intel does with its CPUs, AMD uses a lot of voltage on their GPUs. If your card can handle it at a stable undervolt, the core will automatically overclock itself due to less heat. This is with the stock cooler though.

If you are looking to go to 2100+, you will need more voltage and major cooling. Unlike Nvidia, AMD locks their voltage behind software. Soft powermods are used often to unlock AMD GPU'S for greater overclocks.

 

I remember back in the day unlocking my Radeon card to the next level SKU. There were quite a few years that it could be done. Cant remember which cards they were, but it was quite fun paying for a lower tier card, and being able to mod it to the more expensive one.

 

Although, like @Rage Set said, undervolting AMD GPUs can produce more performance, see Vega 56 & 64 - mainly because they can easily bounce of the power & temperature limits at stock voltages which limits performance. I'm curious to see if the same thing applies to Radeon VII.

 

That Ice Storm kept crashing at anything but stock core clocks.

 

Cool, but one question, why do you load up GPUz after you've completed the benchmark as opposed to keeping GPUz running while you do the benchmark - we'd get a more realistic idea of what's happening to your GPU during the benchmark? Although, you probably get slightly higher benchmark scores with GPUz not running during benchmark, and maybe you only loaded up GPUz after the benchmark to show us the clocks your GPU was running at - because the clocks don't decrease during the benchmark anyway?

 

Point taken.

 

The modded 1080 Ti he purchased from me does NOT change clock speeds during the benchmark. It stayed locked 100% wherever I set it. Or, at least it never fluctuated under loads running it in my systems.

 

No, it's ok, I was just trying to understand as much as I could.

 

This will be interesting. OpenSUSE...

 

Not particularly impressive, something like a 14% increase in performance from stock Radeon VII compared to water cooled / overclocked & powerplay table modded:
https://www.gamersnexus.net/guides/...rplay-overclocking-results-liquid-cooling-mod

 

Nice. Dark mode in Linux is miles better than OS X and Win X.
Run cinebench using Q4Wine and you'll see the CPU performance should be better than Windows.

 

Kind of a hack job water cooling mod. The block only covers the GPU, and nothing else. I will wait for a true water block to be released, which EK is getting ready to release one. https://www.reddit.com/r/Amd/comments/au5v4x/ek_appears_to_be_preparing_an_amd_radeon_vii/
Just ordered my Radeon Vii from AMD yesterday, as I sold the RTX 2070. I will have my first all AMD system since the Phenom 2.

 

Quick firestrike benchmark on the 2070, with underclocked system ram 20,249

https://hwbot.org/submission/4089993_

 

I was gonna say, is there something wrong with that Firestrike GPU score, but then I compared it against guru3d's review of RTX 2070, and it's the same. It's only 11% higher score than my overclocked GTX 1070, surprised by that small difference.

 

Ya, not really impressed with it. It was a quick run, and placed 10th out of 17 on HWbot. Guru was probably running an Intel cpu also, which gave them a better physics score. Looking forward to the Radeon VII, I think it will be much more fun to bench. I know @Rage Set is having a blast with it. After all, isnt that what its all about

 

Whatcha doing over here Uncle Fox

Are you going to compare Linux vs Windows performance? If so, I'd check out the CPU difference as well if there's a crossplatform benchmark like geekbench you can use.

 

https://www.userbenchmark.com/UserRun/15181765
https://hwbot.org/submission/4091249_joet44_cinebench___r15_core_i9_7980xe_5132_cb?recalculate=true

 

I will check out Cinebench and Geekbench on Linux when I get home next week. I am out of town on a business trip until Saturday afternoon.

 

Be safe my Brother. Will be praying for you. Let's talk when you get back.

 

Well yeah, but I'm more into using my GPU for gaming purposes, but I just like the DIY modding & tweaking that goes on here to get the most out of the GPUs/CPU/RAM - the way I see it's free performance plus a curiosity/hobby in it's own right. I enjoy the process of researching/choosing/understanding hardware from a performance/value perspective, then I enjoy the tweaking of it, then I enjoy the using of it, the gift that keeps on giving!

 

You're the second man to ditch RTX in favor of Radeon 7. Any compelling reason?

 

For one, the Radeon VII pretty much smokes the 2070 I sold, and second its going to be fun to benchmark. If I had a 2080 or something to begin with, I probably wouldnt have switched.
I want to put it under water, and play

 

@Rage Set
https://www.primochill.com/products...coverage-gpu-water-block-clear-a-radeon-vii-x

 

I have a Bykski block for my Asus 2080 TI and it isn't bad. So I'm going to pre-order this one. I'm curious about the one from EK though. Thanks for the info.

 

And this is why I stopped competitive benchmarking. I have, after reconfiguring my loop and cleaning the gunked up dye that settled into the CPU microfins, beat @Johnksss score with a 7960X on multithreaded @4.8GHz ( https://hwbot.org/submission/3845237_johnksss_geekbench3___multi_core_core_i9_7960x_69156_points/). Due to the rules, because of the timer RTC issue, even though I am using HPET, I cannot submit this score. I'm now running [email protected]. I might see if disabling spread spectrum requires more than 1.375V and check temps (right now, I don't go above 69C with this config and VRM peak at 84C in only the most stressful tests, otherwise being in the mid-70s or lower).

This wasn't even using a bench optimized OS, as I got rid of those since I can't submit any Win 10 runs outside of PCMark/3DMark, even though spread spectrum was recently added to my bios, my bios doesn't allow for going below 100 BCLK (which below 100 skews the RTC clock to perceive that it ran faster, above 100 causes the RTC to think it ran slower, so BCLK increases would actually harm your score instead of straight 100 with spread spectrum disabled).

 

You're kind of late with that bclk situation. SniperOZ, Q56_Monster and Myself were the first to push the bclk down to around 90 and were getting outrages scores. Then they came and changed it because of the RTC clock and blah blah blah complaint after complaint. Ah the good old days.

 

My complaint is more that certain benches, now with better AMD optimizations in the software and OS, do give higher scores even with the sandbag of HPET on Win 10, but even if doing it just right to make the score proper for comparison to Intel chips on Win 10, you are not allowed to submit them for points.

Now, if AMD does pull a rabbit out of a hat and hits high 4GHz to 5GHz on chips this next round, but the RTC problem isn't fixed, I feel others, like me, will just give up on competitive OC as a hobby because you can't submit them (not to say we won't bench, just there is no way to submit them without them getting tossed).

But I do hear ya on the fun that could be had with it!

Also, you see Der8auer's vid on the 9990XE 14-core chip? That chip is a BEAST! Sucks power like a drunken sailor on shore leave, but is a beast!

 

I think I am going to wait to see what the EK block looks like. .....Maybe..... I may get a hair up my you know what, and just pre order it also.

 

When it comes to bykski, it was one of the better performing blocks on threadripper, like 1C difference from xspc raystorm neo. EK TR blocks sucked. With that said, I use an ek block currently on my GPU and it works well. Point is don't necessarily write off one of the largest block manufacturers in China just because you don't know the brand as well. It might surprise you.

Sent from my SM-G900P using Tapatalk

 

Bykski's quality has surprised me. My Strix 2080 TI performs very well, all things considered with their GPU block. Their terminal design on this block sucks though, as you have to alter the majority of NVLink bridges to work with it in SLI.

 

CB20 is out, and for download at Guru3d as a standalone package: https://www.guru3d.com/news-story/download-maxon-cinebench-r20-benchmark.html
Here's mine with a 4.65Ghz CPU:

 

I got you! LOL... 8168

 

Yeah, 4 times more cores should do that

 

Yep, and the single core score I got was 435 (the stock one they include is 400). But the stock 7700K score for single core is 476, meaning still getting handed my butt on IPC and single core frequency! That is why I'm looking forward to the 11-15% IPC gains for Zen 2! Also readily admit once others with 14-core+ Intel chips start sharing their scores, mine is going to look a bit less... appealing. But, still a nice score! Especially because I don't have a bench OS sitting around for benching anymore.

 

Yeah, We should see near the 11000 territory then.

 

Ha, true! I'd still rather have my 6700K in comparison to your Threadripper though - it suits my needs better, single threaded performance is a lot better, as well as high refresh rate gaming. Here's my single threaded CPU score, 495, pipped you by 13%! Probably snappier general PC use (office apps & web browsing) than your Threadripper too due to that single core performance, and considering that's what I use my PC for as well as gaming, then it's pretty ideal at the moment.

 

Yeah, wish my 6700K wasn't in another system ATM. I used to do 4.8GHz daily driver on it.

As to snappiness, it isn't as noticeable as you think, although the frame rate argument is definitely there if primarily gaming.

But in the past month or two, a couple companies put out multi thread improvements that really drastically improved CPU side performance of my TR. One game that did is Civ VI on AI turn benchmark. But hopefully AMD does more with the GDC on March 20 when more Zen 2 info is released.

I'm trying to convince myself to wait to upgrade until 2020 for DDR5 (will be supported for server chips at least, but 2021-22 for mainstream/HEDT most likely). But the IPC and frequency increase should be nice.

Sent from my SM-G900P using Tapatalk

 

4.7Ghz @1.4V used to be my daily driver for mine, but I brought it down to 4.65Ghz @1.35V - I was reasonably certain I was seeing a small amount of silicon degradation after 1 yr of use at that 4.7ghz (it lost 10mv of stability on previous stress testing). Since then I've also relaxed the severity of the overclock stability testing (in terms of what I deem stable), before it was OCCT, now it's just x264 on a loop all night (plus 2hrs of Prime95 non-avx), so I saved something like another 10 or 15mv by relaxing my testing. Can't imagine I'll see anymore degradation now it's at 1.35V.

 

Geez look at the voltage I had to use to complete the test

 

I thought the same thing, but it was running hotter from the AIO on the 6700K being dirty. Cleaned that, lowered the temps, and the voltage was stable again.

The only chip I've seen true degradation on was my 4790K, which was only 10mV or so and stabilized after that.

What I found more interesting is the voltage needed on my 1950X. When the CPU block was clogged with die, etc., ran way hotter and I needed 1.2875 to support 4.1GHz (100mV less for 3.95). Now I can do [email protected] with no problem, including the financial bench on sisoft Sandra and the blender bench. I messed up one hose at first and it wasn't cooling as well and took 1.375 for stability at 4.2GHz before. So sometimes it is heat at issue, not true degradation (hotter=more voltage leak, etc.).

But the TR is really the most pronounced example, save the FIVR haswell design....

Sent from my SM-G900P using Tapatalk

 

I don't think it was increased heat in my situation, although I did blow all the dust out recently with compressed air and it lowered temperatures, but only a couple degrees, so don't think it was that.

 

You're slowly making me want to take the plunge on that cooler, bro.

 

The Hailea chiller has been the best investment I've made yet.

 

Ran it this morning before I left, A little over 4000 score. Didnt have time to run single thread.

 

4.9 Ghz