ROG Crosshairs day - Ryzen and Thuban day

[flanker]

Most liked platform for me was AMD Dragon. I was surprised at 2010 after the launch AMD Phenom II X6. Yes, its 7 years ago, memories...Today we have new Crosshair motherboard. Sixth generation named Crosshair VI Hero.

Everything what you need to know about Crosshair VI Clock generator is great info for us!

 

Motherboard - new Crosshair

It was more clear, I must have some Ryzens and this motherboard. I was in silence, first BIOSes was not ideal and BIOS chip at my board was broked. Ryzen 7 1800X is my AIO rig. But I bought also Ryzen 5 1500X and Ryzen 7 1700X. And have borrowed 1700 non X + 1600. I tested quickly only 1800X and 1700X, a little 1500X.

 

The main benchtable PC - 1800X inside and Crosshair with Alphacool Eisbaer cooling. Type of DRAM Im using HyperX Savage, Corsair Dominators Vengeance or TridentZ from GSkill. Unfortunatelly, all kits are not B-die. My favourite BIOS for this board are 1001 and 1002.

 

Whats inside the box?

Many good stickers , SATA cabels, ROG SLI HB bridge, coaster for drink etc., discount for sleeve PSU cables, RGB extension cable, installing DVD, paper manual and IO cover

 

This is perfect inovation. Crosshair VI Hero has holes for AM3 and AM4 brackets .

 

The board has solid digital power circuit, which is including phases for CPU=vcore+SOC part. In blue is it digi controlled 2 phases for DRAMs. Remember, if you are use Ryzen, it has 16x PCIe 3.0 links for GPUs and extra 4 directly for M.2 NVMe discs. Others PCIe are 2.0 and interconected with Promontory FCH chipset.

 

SupremeFX with latest ALC1120A codec, some Japanese cappacitors, ampfs and separated lines in PCB

 

Nice designed pasive of PCH. The board looks like really nice, similar as Maximus IX Hero sister. M.2 port is almost bottom, its smart, because its not between SLI/CF slots for GPUs. NVMe M.2 discs are a bit hot, specially in 512GB or higher capacity

 

Ryzen is supporting up to 64 GB DDR4 memory (A2+B2 are primary slots). Ussually you can use 3200 MHz, but at this Crosshair board its possible to get around 3600 MHz or higher for some benchmarks! But you must change BCLK and 2400/2666 MHz in BIOS Near the 24-ATX pin on the left you can see modern front USB3.1. In this area are measuring points of voltage. And post display.

 

OC part is at the bottom. Its change against Common opinion. I think, its better to have it on upper the right corner. Reason? If I will use for extreme OC last PCIe, after I have not good access to the buttons. At photo you can see classic START, small reset, safe and retry buttons, LN2 jumper and slow mode switch. Everything what the extreme overlcockers need to be happy. The board has support for ROG OC panels. And for casemoders there are also two RGB connectors.

 

IO part, so many USB ports, including 2.0, 3.0 and 3.1 :-D I like it. But one PS2 could be here also. Function buttons are BIOS flashback and clear CMOS.

 

Stock vs OC AIO cooled benchmarks

 

I tested only few benchmarks. But you can see big impact in IPC for AMD. Compared it with FX-8370 in table. I did also quick stable OC for 30min in AIDA. I got 4074 MHz with Ryzen 7 1800X. Benchmarks are doable at 4100 MHz also.

 

Crosshair IV Formula + AMD Phenom II X6 1090T BE and LN2

Few days ago I was at dealers day as ROG overclocker. I wanted test my favourite platform AMD Phenom II and new Ryzen. I started with old one, with Thuban. I must say, it was really fun and I used non tweaked Windows 7 for it.

 

Kingpins thermal paste for LN2. Its better than I thought.

 

Isolation with old classic method and many papers. I used also OC panel for read the temperatures and PCI post card into the motherboard.

 

Temperatures minus 186C was the max on sensor via OC Panel.

 

Time for some results. The cores of CPU was similar, the best cores was first, third and fifth. So for some benchamrks I could use mix of 5900 and 5800 MHz , example for Cinebenh R11.5. I was a bit disappointed from Superpi, I thought, this CPU could be 6500+ MHz chip, but it not happens...

 

Wprime 1024M and 32M

 

Cinebench R11 and R15

 

Superpi

 

Geekbench

 

Max clock ( I had almost 6500 Mhz, but file is corupted)

 

 

That was my Thuban And Chew* has the right in this case. This sample is really great on AIR, but only good at LN2...My 1100T BE is with LN2 better extra 300 MHz. Still, it was fun and most of liters was consumed by Thuban ¨

 

AMD Ryzen 7 1700X and LN2

 

For Ryzen, I have not much LN2. Next time I will start with Ryzen and not with oldschool platform Ryzen setup was almost the same, I changed only OC panel for OC Panel II

 

 

And I started with Cinebench R11.5. I found some interesting things. The CPU scaling very well with low voltage. I dont know, if is it normal, or abnormal with Ryzens. Example -120C, 1.44V and 4800 MHz Cinebench R11.5. I was surprised with this. Of course, the best way is start with lower BCLK and 2666 MHz RAMs, specially with low temperatures at CPU/SOC part.

But I tried step by step temps down and the lowest CPU/SOC voltage. Few times I got something as coldbootbug. If I was at -180 C at sensor and working in WIndows, everything was perfect. But if crashed it, I got code for CPU. If I warmed pot up to -150C, CPU was alive and booted But sometimes crashed CPU at the same temperatures around -180C and rebooted at -180C also...I dont know what exactly happens. It was my first experience with Ryzen and LN2...

Step by step I jumped up to 1.488V and last one nitrogen in thermos. 4950 MHz Cinebench R11.5 passed After I was without LN2 and temps goes up...So time for Superpi.

 

 

-srry without better DRAM clocks and DRAM timings - it was dry run without LN2

 

 

Ryzen was fun also, after few first experience and issues at start I was much smarter. Repetation of Ryzen in next 2-3 weeks guys! Hold my fingers! Il need it.

[GeorgeStorm]

Looks like fun, hope you manage some more next time!

[flanker]

I hope so, I believe 5100 MHz could be easy. Maybe with 1.55-1.6V?

[Alex@ro]

Push 1.9 and call it a day

 

Sent from my SM-N910F using Tapatalk

[flanker]

Of course, I hope 5100+ MHz will done with Cinebench glory

[xoqolatl]

Good job

My 1700X also seemed pretty good at lower than expected voltage. This might be a pattern.

[chew*]

The good chips like volts. The good air chips usually do not and that is why good air does not always mean good ln2.

 

Just like i told you on xs flanker about your great air thuban chip

 

Case in point my 1400 does 4150 cb air. -40 Cold it does 4400 cb same volts. More voltage = nothing.

 

These are just like llano...scale with cold hardly any scaling with voltage.

 

If anything binning for voltage tolerance may prove to be usefull as well as binning for -40 cold imc that does not drop threads over 2666 in prime 95...

 

Performance wise its not about how high the core can go....its about balance...

 

-40c 4.4 w/ tight as a virgin 2666 mem CB lost to air @4.150 w/3600 mem in my testing...

 

You can see the same thing happening in the 32m pi low clock challenge...

 

Zen is raping 4.6 + ocs @ 4.2 on air...except mine because im chasing a balance...@ 4.5 it would be very close between us..

 

Of course the balance will most likely limit your chips top possible OC...but if higher speed cpu/low ram is slower....why bother..

Edited May 5, 2017 by chew*

[flanker]

Next week again. But now only the Ryzen (and maybe Kabylake). Target is 5+ GHz for Cinebenchs, hope for 5100 MHz,maybe with some luck could be higher.

[flanker]

Next LN2 try...Ryzen and LN2, its fun, but issues with retraining of RAM, coldbootbug IMC etc. Not always easy. If you want stay at 1002 BIOS, be carefull to use TurboV in Windows, sometimes some settings changed at strange values :-) I had issues with boot to WIndows at Vcore 1.7 or more. So I kept 1.625V

 

validation 5202 MHz

 

PiFast at AMD

 

Wprime+Ryzen is right choice. But with WinXP could be results better with the same frequency

 

Cinebench R15

 

Superpi 32M, I wanted under 7minutes at this clock :-(...

 

Superpi 1M

 

GPUPi with CPU (100M and 1B)

[GeorgeStorm]

Some nice results, reckon there's much more in it?

[chew*]

Veeeeeeesta for wprime multithreaded.

 

Xp for single core.

[flanker]

I thought, XP is best for both, wow. New info for me after many years . Thx.

[zeneffect]

For 32m definitely. Just found more efficiency l.

 

Very nice clocks regardless, that's a good CPU you have there

[flanker]

As AMD enthusiast, Im interesting for change in BIOS/agesa. I thought, the best for me as benchmarker will be 1002 BIOS and I was opponent of new BIOSes. Why? Because at Superpi depends J. But after relase new beta with full unlock all first and secondary timings and drivings I was more and more curious...

 

So, what about compare old good classic BIOS 1002 with agesa 1004 vs new BETA BIOS 9943 with agesa 1006? Lets do itJ. I did comparison in HWBOT benchmarks, which is most interesting for me and maybe for few other guys (Im sure, for real apllications will be new BIOS better).

 

 

I tested at Crosshair VI Formula, Windows 7 64-bit with CPU Ryzen 5 1500X and Gskill TridentZ 2x 4 GB memory. These kits are E-die. I set CPU at 3986 MHz (fluctuation to 3987 MHz) and memory at 2932 MHz effective with timings 12-16-16 in first case (BIOS 1002). Others doesnt change, its automatic. Also I tried BIOS settings „performance BIAS“.

 

In second I cant post with same memory divider, so clock in second testing was 3990 MHz and memory 2946 MHz timings 12-16-16 + auto settings also. The small difference in CPU/memory clock is 4 MHz, so in theory results could be inaccurately in tenth second in Superpi.

 

In all cases the process of benchmark was set to very high. And for Superpi I did two results, because with Ryzen and prediction function seems second run always much better.

 

As you can see, performance BIAS working solid for AIDA/ Geekbench with both BIOSes. „Cinebench R11.5 BIAS“ works best with both Cinebenchs R11+R15 and for wprime 1024M also. But is all the times not good with Superpi runs. „Cinebench R15 BIAS“ seems very good for Superpi 32M and not bad also for Cinebench. Performance „AUTO BIAS“ seems good only for Superpi 32M.

 

My recommendation are:

 

Superpi 32M = use Cinebench R15 BIAS or AUTO settings of performance BIAS

 

Wprime 1024M = use settings „Cinebench R11.5 BIAS“ , difference are not big in others cases

 

Cinebench R11.5 = use settings „Cinebench R11.5 performance BIAS“

 

Cinebench R15 = use settings „Cinebench R11.5 performance BIAS“, u can more play with „Cinebench R15 performance BIAS“, if you will be more lucky than me...

 

Geekbench = use settinngs „AIDA/Geekbench performance BIAS“

 

AIDA 64 = in average the best is again „AIDA/Geekbench performance BIAS“

 

 

Now back to the table with all results. New BIOS 9943 brings small impact of performance, include also Superpi 32M. So dont worry and if you are HWBOT overlcocker, flash your BIOS :-) And of course, now you have more manually tweakings to improve again the score with the same settings.

 

example of best Cinebench with agesa 1004, BIOS 1002

 

And with beta 9943+1006 agesa

 

And Superpi 32M from table of results:

 

PS:now you can easy to do tighter tRAS, TRC subtimings and many more in BIOS!

[zeneffect]

Superpi is still lacking efficiency on latest Asus beta in comparison to other boards. Even manually tuning in all timings it's still not possible to go sub 8:20 @ 4ghz from my experience