Ryzen - Return of the Jedi

[chew*]

Still think 1700 is best. 93% do 100 MHz over rated turbo. that's a preety solid # not to mention lower leakage chips are more resilient to voltage and willing to accept voltage vs higher leakage. My 1700 can run stable at higher voltage.

The 1700x and 1800x reach a certain voltage and say FU after that point.

Of course this may change as I test more chips but for now.......

As we can see by that the % of higher models doing 100 MHz over there rated turbo drops significantly

Doing 4gig on a 1800x that has Xfr to 4.1 and turbo @4.0 is not spectacular. Not to mention the heat that will need to be dealt with.

Sux to be ln2 guys though. binning gonna be expensive.

[chew*]

Think all Core, but don't know
And don't know How Silicon Lottery test them
Your Chip is preselected from AMD and the Chip's from Silicon Lottery are normal one
I think your can not compare with normal one

Long Time ago i bought 5 used Brisbane
Don't know before that they were Review samples
All of them scales better on v-Core than other Brisbane on hwbot

I've proved in the past and will prove it again. the chips amd sends me are nothing special.

Ill dig up an old thread to prove this and link it here, on my quest for 7 gig plus 955's I found them not in chips sent but in boxed OEM retails.

Myself and mad22 found our golden chips in retail.

http://www.xtremesystems.org/forums/...he-house/page2

http://www.xtremesystems.org/forums/...he-right-track

[chew*]

Quote silicon lottery

Passed the ROG RealBench stress test for one hour with these settings:
40x CPU Multiplier
1.44VCPU VCORE (Or less)
LLC Level 3 (Asus Crosshair VI Hero)
Test equipment:
MotherboardAsus Crosshair VI Hero BIOS 0702
Memory: 2 X 8GB 2400MHz 15-15-15-35
Cooler: Corsair H105 AIO
Thermal Paste: Arctic MX-4
Ambient temperature: 22C

I'm going to DL this and correlate how it lines up with what I am doing for stress testing. In fact right now i'm retesting 1700 with smt off.

I know what is not stable so far so after I find what is stable I will start undervolting and see how low I can go with this "method" if at all.

[BeepBeep2]

I've proved in the past and will prove it again. the chips amd sends me are nothing special.

Ill dig up an old thread to prove this and link it here, on my quest for 7 gig plus 955's I found them not in chips sent but in boxed OEM retails.

Myself and mad22 found our golden chips in retail.

http://www.xtremesystems.org/forums/...he-house/page2

http://www.xtremesystems.org/forums/...he-right-track

Got me feeling awfully nostalgic, never got to finish benching mine :o
http://hwbot.org/submission/2371807_..._be_9sec_875ms
http://hwbot.org/submission/2371818_...in_15sec_547ms

Spring break in 2 weeks, might make a trip to airgas to finish what I started 4 years ago.

Ryzen has to wait for me, but I'm wondering how the 1700X and 1800X compare on LN2, considering 1700X is probably higher leakage bin(?)

[cdawall]

Got me feeling awfully nostalgic, never got to finish benching mine :o
http://hwbot.org/submission/2371807_..._be_9sec_875ms
http://hwbot.org/submission/2371818_...in_15sec_547ms

Spring break in 2 weeks, might make a trip to airgas to finish what I started 4 years ago.

Ryzen has to wait for me, but I'm wondering how the 1700X and 1800X compare on LN2, considering 1700X is probably higher leakage bin(?)

Both are high leak bins. The 1700X would be a weak clocking high leak bin, the 1800X a good clocking high leak bin and the 1700 being a low leakage part.

[BeepBeep2]

Both are high leak bins. The 1700X would be a weak clocking high leak bin, the 1800X a good clocking high leak bin and the 1700 being a low leakage part.

You think so?

The Stilt ran tests with voltage scaling and it seems like 1800X are running higher than optimal voltage to reach those frequencies, so I'm not really sure if we can assume higher or lower leakage based on the clockspeed, TDP, and stock voltage differences. For example, I'm thinking an 1800X undervolted @ 1700 clocks could possibly run around the same power consumption as a 1700(?)



If the 1700X is just a part that can't meet the TDP target but scales with cold/higher voltage that's one thing, if it just clocks like s*it all around then I suppose that is another

[chew*]

1700 has been doing very poor on ln2. Regardless of what testing on air proves it literally means nothing cold. Those who have had acess to bin alot of chips are using 1800x....not the 1700. Since they had acess to bin and they have been running cold. Id say they are better informed than those who have not run cold.

My 2c

High asic ( air water ) = low leakage

Low asic quality ( ln2 ) = high leakage

1800x choice chip used on ln2 thus far......

simple math 2+2 = 4

[cdawall]

To be fair they are all low leak chips the 1800x should however be more on the high leakage side than a 1700. I mean that is how binning works...

If you really look at the graph kicked out by stilt these chips shouldn't be clocked this high from the factory anyway. The vast majority of 1800X's would be equal to an Intel bin in the 3.3-3.4 range which is why they clock terribly.

[chew*]

OK well I got a baseline.........will be interesting. I ran prime and this. still running prime.

Regardless of possibly innacurate temp readings and all things aside......1700x and 1800x make my heatsink way hotter.

They make my pwm scorching hot.....the 1700 does not.......PWM cooking = power pulled is higher.....more power pulled is higher tdp....higher tdp at equal voltage .....high leakage parts

I am literally beating the crap out of the system and cpus way more than is recommended.

I clearly am not relying on one stress test I am combining them clearly to really piss these chips off so I am likely to be seeing more heat than most.

[BeepBeep2]

To be fair they are all low leak chips the 1800x should however be more on the high leakage side than a 1700. I mean that is how binning works...

If you really look at the graph kicked out by stilt these chips shouldn't be clocked this high from the factory anyway. The vast majority of 1800X's would be equal to an Intel bin in the 3.3-3.4 range which is why they clock terribly.

Well, I think you know what I meant. 3.0-3.6 is a large difference, especially with the voltage scaling, so I don't think these necessarily have to be low-leakage binned parts to reach the 3.0 GHz base clock @ 65w TDP vs. the 3.6 @ 95w TDP, because those parts need to be rather efficient in the first place too to hit the 95w TDP target, and part of the reason they run hot is because the voltage is jacked sky high to hit those frequencies. What I was speculating was maybe that some of the 1700X can hit the same frequency, but higher leakage and can't meet TDP target or temperatures are too high on air and water, but might be gems under LN2(?)

Brian I'd be really interested in seeing 1700/1700X/1800X min voltage for 3.8 stable and temps at say, 3.8 1.35v if you have the time in the future.
I also heard about the 1700's not clocking well on LN2.

How bad is the VRM heat? I remember hitting ~100c on the VRM/PWMs with Bulldozer on water and 1.45v

Also, that is a bit of an odd trend with voltage tolerance, typically on Phenom II and Bulldozer parts the higher leakage CPUs were more tolerant to scaling with higher voltage whereas the low leakage chips would hit a wall and stop scaling sooner, at least on LN2. Maybe there is a temperature factor on air/water vs LN2? Like, my 6.9G 955 doesn't scale at all above 1.82v on LN2 but my 7G chip eats 1.9v easy.

[cdawall]

Only time will tell, but with the clocks we are seeing on LN2 I don't know if any of the 1700X's are worth a .

[chew*]

Heat is bad enough to burn me with certain cpus installed.

Like i said my comparison is not by default vid. 1700 is way lower...im running all chips at identical volts...well +.0750mv on 1700 vs 1700x and 1800x.... Add 20c to my 1700 temps and those are my x chip temps equal volts.

Im not running a test at 3.8...time consuming i already ran 4.0 add 20c. Done.

Default vid is 1.2x on 1800x default vid is 1.0xx on 1700...

This just further supports my findings.

1700 scaling with voltage on air/water and likes it way more. Heat output despite this is less from heatsink...screw the temp monitor software..

1800x wants to be cold and refuses to scale otherwise. Yes i got to 4.2 at who knows how cold..for 40c idle with voltage highly advised against.

Between what im seeing and ln2 guys are seeing plus 1700 not liking voltage on ln2...the data is pretty compelling evidence.

The old way will no longer work. Turbo, dyanamic tdp, xfr and amd running a tight binning scheme has changed that method drastically.

Even BD was different...the way i had to bin changed and even then it was still somewhat of a crapshoot, turbo was most likely playing a factor in that...lowest vid sucked.

Old way worked for deneb thuban...

Ive played with enough silicon to understand it. Its a gift.

[chew*]

Anyone want to question the validity of my idea of stable? speak now or forever hold your peace. Prime is running blend fwiw to stress cpu, imc and memory.

[zanzabar]

do you think a micro code update can get you any more to play with or do we need a new stepping?

[chew*]

Time with the process might see 100-200 but gonna need like a c3 type respin i'm thinking for anything significant without binning and luck.

Performance wise I'm preety certain an update can unlock hidden potential.

[StAndrew]

do you think a micro code update can get you any more to play with or do we need a new stepping?

I've been thinking this as well. Overclocking Gulftown, I needed 1.45v to get 4.1 (HT on). Increasing clock amplitude and adjusting clock skew I was able to get to 4.2 on 1.37v

I'm guessin when mobaord partners figure out the finner tuning of these CPU's we might see some marginal increase in clocks...

[chew*]

Ok bored as hell stressing chip so figured i'd delve into this simple not really a big deal but otherwise annoying bug.

Seems it does not just effect gigabyte with smt off.

Anyway not sure if this is cpu-z issue or vendor issue but as you can see if effects all but normal cpu-z. Upon tabbing the core tab it shows only 2c/4t

obviously the normal cpu-z is reporting correctly.

[charged3800z24]

HMM, so it actually shows it as a two socket 4core setup i nthe other CPU-Z's?

[chew*]

Ok so since this thing is almost done and we seem to have dialed in "preety" stable settings I say pretty stable as once again I find it important to point out that Prime 95 needs an update cache is reported wrong therefore may not be 100% reliable in testing yet.

But anyway lets call it pretty stable. Red apples to green apples pretty stable OC air vs 5 year stable AIO water OC ( its aio so seriously its quieter realistically a few degrees cooler ).

Specs wise looks like max realistic OC head to head is going to be 4.0 8 core versus 4.5 4c/8t. While seemingly not fair its not AMDS fault that sandy is not ddr 4 and likewise its not Intels fault AMD can't clock high.

Why not newer? Simple I don't have it and tbh I think many are still on sandy well at least the smart ones did not repeatedly upgrade for a couple frames per second....and lastly I want to see if this is an upgrade and if I will lose much in this configuration vs what I gain being able to swap to workstation mode with simply loading a bios profile.

Why 8 v 8? because in the games I play 16 is just dumb regardless of reviews I had full intention of gaming this way, when not gaming ill load a profile equivalent of the task needed...

Why not 4 v 8? well simple I use my PC for more than just gaming I run background apps don't need to close them it runs smoother this way. Want the choppy experience that's your choice.
I ran 5 gig 4 core for about a year I did not like how it ran.

So lets talk about the system.

ASUS M4E ( with awesome USB drop issue to this day )
Binned 2600K ( 56xx very very hot chip ) 4.5 @1.3v
Gskill trident 1866 8-9-9-34 2x8g ( its faster than the looser timing higher speed trident, want to debate go beat my sandy efficiency on hwbot with less clocks then we can debate my systems performance and my knowledge of system performance )
1200 watt quality PSU. if its quality it really wont matter, I'm not pulling the corsair just to match PSU's.
Fury X for both systems, top slot on M4E to avoid nf200 latency issues. Move on to other cards later.
Yes its going to be on AIO water h220X straight distilled, norprene, kill coil recently serviced for a fluid swap.
Monitor ACER KG240A
HD WD 160g per OS fresh installs per system.

[chew*]

HMM, so it actually shows it as a two socket 4core setup i nthe other CPU-Z's?

Shows if you tick the cores 2 physical cores not cpu's.

[chew*]

Last chance to question my idea of stability.

[chew*]

For those that asked.

The above Hard crashed @1.4250 v in less than 1hour.

It desktop crashed prime 95 @1.4375 in less than 3 hours.

It passed 24h plus kickers @1.4500 Peak temps sustained for periods of time 77c

Heres same settings 1.4250v silicon lottery test. Peak temp 66c

[chew*]

Reliable testing method we have yet to find > prime95 with kickers > prime 95 by itself> realbench.

I see some guys claiming what stilt and myself said about prime 95 on various forums.

Then they go on to say realbench is the best method.

I myself have definitely not said that realbench is reliable.

Clearly the above shows that is not the case.

We are concerned prime 95 is not enough stress...and the cache is wrong so it may not be using it all.

Without enough stress its still capable of crashing way before "realbench".

Do with this info as you will. Im not challenging silicon lottery. If they bin by this method it will still weed out the good from the bad.

If you run at the voltage they state your cpu passed at don't be surprised while playing bf1 and bragging about your new ryzen rig that you bsod and embarress yourself.

[VietOZ]

Put it on WCG for a week, if it's not restart then it's stable

[chew*]

Why do you think i test like this

You guys build systems to and most overlook that 24/7 mission critical stability.

Dave and i were preety close so always have kept crunching guys in mind.