Ryzen - Return of the Jedi

[aGeoM]

New AMD 32m PI WR.

Attachment 132387

Hi

Link not working for me at least.

I suppose it is yours, in that case, congratulations, you deserved it. you have been working very hard for this moment, I don't post much but every day a came here .



Keep it on.

[chew*]

Thats odd opens on pc but not on phone.

Fixed.

[os2wiz]

What is the fastest memory speed recorded with Ryzen? I felt stymied with my dual rank 32 GB memory kit (B-die) ddr4 3200 running at 3066nhz. I decided to buy a Gskill single rank 16GB kit at 4266mhz with timings 19-19-19-36 . I plan on downclocking it at 3733mhz at 15-15-15-32 or 15-15-15-30 if possible with CR T1. Is it very crucial for performance to disable geardown, as you wind up with a T2 command rate, which has to impair performance somewhat? I hope my plan is feasible.

[Hell Hound]

Would love to see 2400 cas 9-10-10

Closing in on ddr3 cas /2000c7 vs 2666cs10

http://imgur.com/a/zzKD8

[chew*]

I have done 9-8-8 2666 but...it was sketchy.
Speeds i run are dictated by cpu when sub 0. I tried higher and looser. Would not boot.
Once I know speed it likes i just make due and dial in performance @ that given speed.

Os2wiz think someone broke 4000 on single stick on a gigabyte. I am more of a performance oriented mind then a suicide run kind of guy however.

[Hell Hound]

Was that 16 or 32gb(all banks filled) @ 2666c9

3200c10-9-9 would be a great improvement over my 1800c8-10-10

Quad channel would be the cherry on top

32gb sticks that can do that = Dreamland

[BeepBeep2]

What is the fastest memory speed recorded with Ryzen? I felt stymied with my dual rank 32 GB memory kit (B-die) ddr4 3200 running at 3066nhz. I decided to buy a Gskill single rank 16GB kit at 4266mhz with timings 19-19-19-36 . I plan on downclocking it at 3733mhz at 15-15-15-32 or 15-15-15-30 if possible with CR T1. Is it very crucial for performance to disable geardown, as you wind up with a T2 command rate, which has to impair performance somewhat? I hope my plan is feasible.

Unless you are extremely lucky with your IMC and RAM, there is probably no way you'll be able to run 3733 MHz at CAS 15. 3600 CAS 16 might be feasible on single rank.

I've seen a few people have 3600 stable and have seen AIDA / boot at 4000 on one CPU so far but most people are between 3000-3200 or 3466.

I have done 9-8-8 2666 but...it was sketchy.
Speeds i run are dictated by cpu when sub 0. I tried higher and looser. Would not boot.
Once I know speed it likes i just make due and dial in performance @ that given speed.

Os2wiz think someone broke 4000 on single stick on a gigabyte. I am more of a performance oriented mind then a suicide run kind of guy however.

How bad is your IMC on LN2? 2666 max? I was having issues booting above 2400 on B-die below -60 all the way down to full pot over the weekend. Boot up to 3600, not stable, 3200 runs 32M CL12. CLR CMOS / ASUS Safe Boot don't help with the CBB. Code 55 or code 8 no ram / no cpu. Tried diff proc ODT as well. By the time I got 2666 booting I was getting multiplier locked at 5.5x

[os2wiz]

I have done 9-8-8 2666 but...it was sketchy.
Speeds i run are dictated by cpu when sub 0. I tried higher and looser. Would not boot.
Once I know speed it likes i just make due and dial in performance @ that given speed.

Os2wiz think someone broke 4000 on single stick on a gigabyte. I am more of a performance oriented mind then a suicide run kind of guy however.

Yes, I am not looking at breaking any record in speed , just want to get a higher speed with very tight timings even 3600mhz at 14-14-14-32 would please me. I want my frame rates higher so I can maximize the RX Vega card I plan on buying next month. If my IMC limits me to 3466mhz I will live with it. The ram I bought is the most highly binned B-die available right now, so I am hoping for a little more juice. Thanks every one.

[os2wiz]

Thanks for the heads up.

[chew*]

How bad is your IMC on LN2? 2666 max? I was having issues booting above 2400 on B-die below -60 all the way down to full pot over the weekend. Boot up to 3600, not stable, 3200 runs 32M CL12. CLR CMOS / ASUS Safe Boot don't help with the CBB. Code 55 or code 8 no ram / no cpu. Tried diff proc ODT as well. By the time I got 2666 booting I was getting multiplier locked at 5.5x

This chip is really bad. Needs gen 1 to boot -80 needs to be -120 to reboot. Dead stop @ 2666. Valids just under 5.3. Benches 32m at least 5.1...but must downclock to save in paint lol.

But i dissagree with wizerty....that mentality of just grab another cpu ruined this hobby.

Doing what you can with the cards you were dealt is grass roots and only increases your experience level....so when you do find gold...you can make magic happen or if you do a live competition you don't look like an idiot whos skill lies only in bin to win.

[BeepBeep2]

This chip is really bad. Needs gen 1 to boot -80 needs to be -120 to reboot. Dead stop @ 2666. Valids just under 5.3. Benches 32m at least 5.1...but must downclock to save in paint lol.

But i dissagree with wizerty....that mentality of just grab another cpu ruined this hobby.

Doing what you can with the cards you were dealt is grass roots and only increases your experience level....so when you do find gold...you can make magic happen or if you do a live competition you don't look like an idiot whos skill lies only in bin to win.

I like it. Looks like I'm gonna have to put in work to get 2666+ working and stable. I should have clarified, 3200 CL12 is 32M stable on air @ 4G, CL11 was a no-go and high BCLK was a no go, just gets worse on LN2. "Safe boot" mode that "should boot" w/ LN2 mode = 105 BCLK, Gen2 and whatever else tricks obviously doesn't help, only the torch did >-60c. Couldn't reboot 2666 90% of the time even at full pot.

Code 8 surprised me, first time I quit getting code 55 I started getting code 8 until warmed up to -55c again. CBB code 8 stayed even with CMOS reset @ 2133 ram.
Anywho, point is Ryzen does not make it easy to bench on LN2 with a bad IMC and the board gimmick on the CH VI H didn't save me. Will have to revisit.

You ever get multi locked at 5.5x? Acted like extreme thermal throttling, but only with LN2 mode on. Not sure if you've had any ASUS boards cold yet but definitely made me feel stupid for not installing TurboV. Didn't get a single result or learn much from the session except that I should install the GPU driver before benching Cinebench.

[Hell Hound]

It seems that 2933 with low cas might be limit for gen1 Zen,

[chew*]

Yah i have had my C6H under phase more times than i can remember. It recently dropped a channel but i sorted it out.

Working fine well sort of i fooled with it last night on a 3200 bin c14 set known good for 3400.

Refused to boot 3333 turned power off went to bed as i was not in the mood.

It was probably the SPD on the memory. One of my boards rewrote SPD on one stick to stupid settings.....no clue how.

My taichi does not seem to care about the broken SPD. It is not as finnicky. Oddly enough the fatality is more picky...will not boot with the set on Msimax's board. Its the same damn board lol.

[os2wiz]

It seems that 2933 with low cas might be limit for gen1 Zen,

Well I installed my new 16 GB memory kit. G.Skill F4-4266C19D-16GTZKW. I found out quickly how mediocre my IMC is. leaving all timings at auto. I was barely able to boot into 3300mhz with total instability with very high auto timings. I backed down to 3200mhz and was able to tighten timings to 14-13-13-13-30 CR T1. I had set timings at 13-13-13-30 CR T1 and it automatically changes the first value to 14. I did the appropriate SOC voltage and dram voltage stabilizing . I passed Ycrunch memory stresser. One thing with Y Crunch I have seen over and over again is if you pass the Fast Fourier subtest you wil pass all the others. I guess it is the most stressful on the memory controller of all the exams. I am looking forward to Zen 2 to solve this memory controller problem. Am I deluded to believe it will be a substantially better IMC or is there some rational reason for optimism??

[Hell Hound]

Well I installed my new 16 GB memory kit. G.Skill F4-4266C19D-16GTZKW. I found out quickly how mediocre my IMC is. leaving all timings at auto. I was barely able to boot into 3300mhz with total instability with very high auto timings. I backed down to 3200mhz and was able to tighten timings to 14-13-13-13-30 CR T1. I had set timings at 13-13-13-30 CR T1 and it automatically changes the first value to 14. I did the appropriate SOC voltage and dram voltage stabilizing . I passed Ycrunch memory stresser. One thing with Y Crunch I have seen over and over again is if you pass the Fast Fourier subtest you wil pass all the others. I guess it is the most stressful on the memory controller of all the exams. I am looking forward to Zen 2 to solve this memory controller problem. Am I deluded to believe it will be a substantially better IMC or is there some rational reason for optimism??

Deneb to Thuban gave us 200 mhz @ same cas

Maybe 3800c16 stable , 4ghz max w/ cas 18-18-18(best case) 19-19-19 if your lucky

[BeepBeep2]

Well I installed my new 16 GB memory kit. G.Skill F4-4266C19D-16GTZKW. I found out quickly how mediocre my IMC is. leaving all timings at auto. I was barely able to boot into 3300mhz with total instability with very high auto timings. I backed down to 3200mhz and was able to tighten timings to 14-13-13-13-30 CR T1. I had set timings at 13-13-13-30 CR T1 and it automatically changes the first value to 14. I did the appropriate SOC voltage and dram voltage stabilizing . I passed Ycrunch memory stresser. One thing with Y Crunch I have seen over and over again is if you pass the Fast Fourier subtest you wil pass all the others. I guess it is the most stressful on the memory controller of all the exams. I am looking forward to Zen 2 to solve this memory controller problem. Am I deluded to believe it will be a substantially better IMC or is there some rational reason for optimism??

At 2666 ratio and up, CL automatically rounds to the next nearest value. 3200 CL12 looks like it would be the fastest option for you.

You may also find a DRAM hole where it will not post on a certain ratio at a given speed but will at a higher ratio or higher speed. ie. You may have 3200-3300 working, but 3300-3466 do not work, and 3466-3600 works.

[Hell Hound]

Sometimes you gotta massage the chip.

Whats the best you seen those sticks do ?

3200c10-13-13 Is still smoking + Quad channel can up make for bandwidth loss

I would love to see that

[Hell Hound]

Sometimes you gotta massage the chip.

Whats the best you seen those sticks do ?

3200c10-13-13 Is still smoking + Quad channel can up make for bandwidth loss

I would love to see that

RAM Timing vs Frequency Tabley.xlsx

https://drive.google.com/file/d/0B5e...5TLUZMYzg/view

[I.nfraR.ed]

At 2666 ratio and up, CL automatically rounds to the next nearest value. 3200 CL12 looks like it would be the fastest option for you.

You may also find a DRAM hole where it will not post on a certain ratio at a given speed but will at a higher ratio or higher speed. ie. You may have 3200-3300 working, but 3300-3466 do not work, and 3466-3600 works.

On my board (only have CH6), to get odd CAS at higher speeds, GearDown needs to be disabled. When Enabled (or AUTO), odd CAS is rounded to next even value and you can't control CR.

[os2wiz]

Chew I remember you recently talking about noise in the memory ecosystem and how that could affect performance just as much as speed and settings. My new super binned DDR4 4266 dimms are a case in point. Yes my IMC is lousy on my 1800X, so I can only do 3200mhz stable at 14-14-14-34. But my cinebench 15.038 Open GL score is consistently at 129.45 fps as opposed to my Flare-X B-die at 123 fps before I sold them off. The settings are identical. So this super binned B-die must be producing less noise interference than the Flare X B-die did at the same speed. Thanks for your tutorial. I now feel somewhat vindicated in my decision to go for this higher-binned ram.

[chew*]

It could also be a lower or a higher sub you still have no access to or a timing that is more stable. ddr 4 is picky and so is ryzen and it reflects in the results.

[chew*]

Was that 16 or 32gb(all banks filled) @ 2666c9

3200c10-9-9 would be a great improvement over my 1800c8-10-10

Quad channel would be the cherry on top

32gb sticks that can do that = Dreamland

I have not had much luck sucking in tight with DR.

I am actually running 64gb now......also lack the cahones to shove volts into $700 worth of 64gb b-die lol.

64gb 3066 1T 14-14-14-28 is good

64gb 3200 2T 14-14-14-28 is good

32gb 3200 1T 14-14-14-28 is good

All DR

At least on C6H.

Of course I do not have 64gb on hand for R7 just happened to be curious. Its for something else

[BeepBeep2]

Chew I remember you recently talking about noise in the memory ecosystem and how that could affect performance just as much as speed and settings. My new super binned DDR4 4266 dimms are a case in point. Yes my IMC is lousy on my 1800X, so I can only do 3200mhz stable at 14-14-14-34. But my cinebench 15.038 Open GL score is consistently at 129.45 fps as opposed to my Flare-X B-die at 123 fps before I sold them off. The settings are identical. So this super binned B-die must be producing less noise interference than the Flare X B-die did at the same speed. Thanks for your tutorial. I now feel somewhat vindicated in my decision to go for this higher-binned ram.

All subtimings identical? A lot of subtimings are set by the motherboard or SPD / XMP profile if left to "AUTO" and can have quite an impact on performance. Something as simple as leaving geardown mode on vs. running 1T with it off could account for that difference. chew* I believe was talking about power delivery, that's only going to help or hurt on the edge of stability to push an overclock and should have no impact on performance

[vario]

...

Of course I do not have 64gb on hand for R7 just happened to be curious. Its for something else

...

So, we can expect day one TR review from you Chew ? :-)
Also, have you tested 32Gb on ryzen with something other than b-die at 2933+Mhz ?
So far from what im reading 32Gb/3200mhz kits are only for high end boards with high end B-die sticks.

[chew*]

I would call it more of a report than a review and i really hate doing videos but there is so much to cover and some of it you just need to see to understand. Conveying it through a forum post will be hard. It will be off the cuff no editing no rgb disco lights humour just raw and uncut straight to the point as always and most likely a bunch of small clips covering various aspects.

I am not really aware of to many big density kits that are not bdie.

Really went above my means on this one financially and capability? wise. Originally had no intentions of working on it as I find myself at a loss to fully utilize it (TR and 64gb). I had discussed this with AMD already.

Was asked to by a vendor or shall i say an overclocker i respect who works for one.

I think i have the real world use or shall I say user that can actually show real world use covered now. I will do my report. Pretune it then pass it off and see what he comes up with.

This is the best I can really come up with for the OC community/workstation crowd which I think its marketed more towards.