I.nfraR.ed

Still struggling with Win7. Same bios settings (slacker timings close to auto atm, except primaries and tWCL). Installing same stripped windows 7, fresh boot, no tweaks, everything should be 1:1 and I get 3 seconds difference with the different installations. Can't figure out why. And it was consistent, not just some random run. Latest beta bios has drive strengths, but haven't played with them yet. Trains the memory first try though.

Got back to Ryzen, but I'm still 10sec behind. Currently only trying 4GHz, 100bclk and 3200 divider for good measure.

Don't believe that's possible with water cooling and current Zen revision. As Alex said, it should be some very powerful chiller (SS territory).

Why don't you fix the parser on hwbot to accept both http and https. Some people are using https-everywhere extension or similar Plus the validation message is not really helpful. It's like saying "there's an error with the value in this field" while you're trying to register for some service/site, without providing any details Sounds fun, eh? If I, as a developer, did similar thing with the validation for any of the products I'm working on, a QA would assign me a bug.

I will try, but I'm testing some Hynix M-die (2x16 and 2x8) right now. Need to get this done, then will be back to B-Die for benches and 32M.

So I thought it might be good to add my tests with the same model on Ryzen. Test system was 1800X on air, running on Asus Crosshair VI Hero with bios 0082. First I started with XMP. Worked without issues and I was able to lower the voltage down to 1.325V. Next, I simulated a "standard" 3600C16 rated kit, which also worked flawless. Last thing I tried was tightening the timings to XMP values on the same 3600MHz. Obviously a bump in voltage was required, but I still consider this on the safe side for 24/7 operation. This kit is weaker than my best B-Die, but still pretty decent. More than 3600MHz on Ryzen are not really viable for 24/7 at the moment, even this is pushing it, because you need rather high bclk to achieve such speeds and might have problems with various devices attached to the PCI-E lanes. I've also tried C12 and C11 - all worked with voltage increase. Haven't checked them on Apex yet and don't know voltage tolerance and max Pi32M with wazza. Overall it's a good memory, however if you own other good B-Die kit, there's no point changing it. You can also buy a higher bin B-die, they will still work at same timings on Ryzen. Another thing is my IMC seems to be above average, so you might not be able to get same results at 3600, but 3200 should be no problem on 100MHz bclk with 3200 ratio or with higher bclk and lower ratio.

I take it.

I could not get higher multipliers with my 1GHz T-Bird when I tried last time. My goldfinger is a different brand. Perhaps the limitation comes from the motherboards.

You have it in your thread: Should do the trick.

That's 32M

Common, even at LLC5 it is going slightly above what set in bios. It's not like going to 1.8V when 1.4V is set. I don't see what danger you're talking about. Also, what voltage limit on each LLC? You can measure it under socket for actual voltage.

Currently, we don't have control over any memory timings, except the main 5. AMD's AGESA code provides "tables" with different sets of timings depending on the divider used. Whenever you use e.g. 2666 divider, you always get the same set of timings and you can only change main 5 in bios - all other subtimings are what AMD thought suitable. Higher divider - slacker timings - higher latency. If you want the best possible results, you have to increase bclk and use lower divider (with tighter subtimings). There's another thing. I have tried 4x8 single rank Samsung B-die and it is impossible to boot at anything lower than CAS18 on 3200 divider, so if I want to run 3200C14 with four sticks and decent timings, I have to use lower divider in combination with bclk. There's something wrong with higher dividers. Same thing applies to 2x16 dual rank, I think, but have no such sticks to verify. 3200 with CAS lower than 18 basically only works with 1 sngle rank dimm per channel (2x8/2x4). Many people with high density kits have problems booting with higher dividers at all, so this is another reason to use lower dividers. PCI-E drops to GEN2, but I don't think this is a big issue even for 24/7. For benching you can go much higher in bclk, but for 24/7 it is not really recommended.

I will wait 2 more years, then will kill all rankings with some new DDR4 and cheaper CPUs PS: If XTU is still alive, lol.

Gold digger

TIM layer looks very thick. Next time you might want to try what I do: pre-heat the pot with a hot air gun or even a hairdryer and apply the paste on the pot, rather than on the cpu. Mount it, but twist a little before tightening the bolts. I know Roman shows how "easy" it is to spread the Kryonaut with the supplied applicator, but in reality my Kryonaut looks much thicker and harder to spread compared to his video. It falls apart and the pieces stick to the applicator, rather than the IHS, so pre-heating the surface and the syringe helps to make it a little more liquid than it is. I'd say I'm rather disappointed by Kryonaut.

I have 2 kits: One is older (Jan 2016) 3600C17 (17-18-18-38 XMP) and the other one is more recent 3733C17 (17-17-17-37 XMP). The 3200C14 test was done with both kits in a 4x8GB combination. Both kits do 4000+ 12-11-11 on Apex motherboard.

Just to make things more clear for you. You only receive points for your best submission with a given hardware. Benchmarks are divided into CPU, GPU, Motherboard and RAM. CPU-Z is a CPU benchmark, so the motherboard doesn't matter for the points. If you submit the exact same CPU model with 10 different motherboards, you will only get hw points for the best submission out of these 10. Motherboards are only ranked by FSB frequency. You use the cpu-z validation again, but this time you enter the motherboard reference clock. This time you have the opposite - the ranked hw is this specific motherboard model, so it doesn't matter if you submit 10 different results with 10 different CPUs using this motherboard. You still see your old submission with this motherboard in the rankings, but it does not add to your total points, because you have now a better score in the same cpu-z benchmark with the same cpu model. PS: More or less what George said

I have one dead mosfet on mine. Can be replaced. Hopefuuly you have your cpus alive.

Wow, this is beautiful! Excellent score, guys.

Lol, why do you think it's not on air? You can reach the same clock with the stock intel box cooler.

Well, that's the AMD "optimization", I guess Most important is the chip - B-die. Both kits I have (old 3600C17 and 3733C17) work without a problem on Ryzen. I even tested for stability 4x8GB at 3200 14-14-14-34 1T (trying to mimic that G.Skill screenshot from some time ago) at 1.35V. Also managed 3800+ at C12 bench "stable". But I guess it is good for people who just plug and play. It will offer a little better performance compared to 2133 JEDEC. And, this FlareX doesn't seem too bad, considering many recent high bins fail on the same test. It could be much worse, plus voltage tolerance is higher than expected.

0082 is the best so far for me. Mem could not go so high before. Also don't see the Win10 bugs, everything seems fine in terms of scores.

Setting maxmem is the way all top guys use to bench at these insane frequencies under x64 OS.

Use the tool "EZ Installer" provided on Asus page: https://www.asus.com/Motherboards/ROG-MAXIMUS-IX-APEX/HelpDesk_Download/

Yes, I've noticed mega now wants to make account and pay when the quota is exceeded or wait several hours until the download is unblocked again. This SP1 image is rather old though and does not include updates, but I didn't want to get a custom image as a base, you never know what is inside, plus I guess it is illegal. That's why I've opted for the official MSDN SP1 image, no cracks, hacks, activation, etc. It would be better to provide a simple guide how to integrate drivers, although there are tools for this already - e.g. on Asus site. Asrock might have one as well. But I will find some time to write a guide for NTLite, because you might want to integrate some extra stuff. This image was just to get things going for people who have no clue and/or time what is required and haven't researched on the topic.