WTF. So far I CANT SEE ANY overclocking test going IN DEPTH,

File: 1486787143697.jpg (383KB, 935x1402px)

383KB, 935x1402px

>over clocking any 8 core processor ever made above 4.5gHz
>Ever

No thankyou

>>59302858
too early, wait a month
it's been 3 work days, why do you expect results so early?

>>59302878
What the fuck are you on about
1800x already hit 5.8GHz on LN2.
Demonstrates the chip is way capable of hitting super high frequencies, the only limit being temps. SO WHY THE FUCK LINUS OR ANY OTHER MANCHILD IS MAKING A VIDEO TRYING TO OC THE 1800X WITH EXTREME COOLING SOLUTIONS ? ughh

>>59302914
What I despise is why the first cooling solutions employed are utter, total shit. Either stock cooler / shitty AIO. How the fuck someone popular enough to receive ryzen kits for reviews can't spend 3 more hours testing the cpu with premium cooling solutions. That really misleads the customer into thinking "yeah it wont OC", when all the reviews use pajeet cooling solutions

File: image.jpg (31KB, 494x309px) Image search: [Google]

31KB, 494x309px

https://www.reddit.com/r/overclocking/comments/5xvu8y/ryzen_r7_1700_binning_data_from_10_cpus_article/

>>59302992
Why don't we see the chart going to 1.5 ?
Will the chip become fucking molten lava at 1.5 ?

>>59302858
It's not because of heat, it's due to voltage required to go faster than that is above safe for the silicon value.

>>59303038
14nm transistors are just gonna burn up at voltage any higher.

>>59302929
On 1.6 voltage

i think it only worked for hours before burning up and taking MOBOs power circuit with it.

>>59302878
The /pol/ wells in me

>>59303072
You can easily push a 6700K over 1.4V on air. The 7700K runs far hotter at that sort of voltage though.

>>59303098

That's just the common sense firing off to curl your fingers in to a fist.

>>59303115
1.4 is the recommended safe value for 14nm FinFet transistors.

If you want to be a tech hero you can set up an experiment, buy a 1000 CPUs put them on 1.425, 1.450 and higher etc, run them for two years and provide us with statistical data on how long they lived.

>>59302929
With a super cherry picked chip by amd god knows if the same production

>>59303115
No kidding some mobos set 6700k vcore up to 1.45.
It seen Intel set 1.45 as upper safe value, but I usually don't want to go beyond 1.35. above that there are a lot of waste heat for small margin.

The problem with 7700k is the shit silicone rubber used. Since I dellid mine it works at maximum 65c 1.33v 5ghz

>>59303038
not recommended. only casual overclocking territory.

>>59302929
>5.8GHz on LN2
How is that a good thing? A 7700k can do 7ghz on all cores on ln2. You're only gimping your own system by having more cores.

>>59303269
more threads: more cores are better
less threads: higher clockspeed is better

>>59303189
t. assblasted 5960x user, previous holder of the 8core cinebench wr

>>59302858

4GHz is pretty much the hard cap of the Ryzen architecture and the 14nm LPP process combination.

I wonder if AMD is going to release updated enthuasist CPUs on 16nm FF+ process, that could result in 300-500 MHz gain.

File: Untitled.jpg (16KB, 620x330px) Image search: [Google]

16KB, 620x330px

>>59302858
|||||||||||||||||||||||

>>59303269
That is pretty common knowledge you need to increase voltage and reduce frequency upon multiple dies on same package.

>>59303433
>>59303433
>is going to release updated enthuasist CPUs on 16nm FF+ process
I think they are going to spend their next year studying why the fuck is it so and doing something about it and release a faster Ryzen gen 2

>>59303448
|||||||||||||||||/|||||

>>59303635
kek

>>59303448
>>59303635
wew

>>59303635
ʅʅʅʅʅʅʅʅʅʅʅʅʅʅʅʅʅʅʅʅʕʅʅʅʅʅʅʅʅ

>>59303876
#rekt

>>59302929
i'm 2 years now with a Phenom II running at 4.3 GHz at 1.6
never had problems and I'm on air

>>59303999
Well thats because Phenom IIs are goddamn tanks. These newer chips not so much, mainly due to the far smaller process node.

>>59303610
Well let's see
Intel spend a bazillion coins to develop it own 14nm process
They took a lot more of time than expected and are sticking with it another generation now they matured fabrication allowing low voltage high frequency like 7700k
Meanwhile ryzen is the first big 14nm chip global foundries made
And design chips. Who designed and does the fab process is global.
Big Intel CPUs can strike up to 4.3-4.4 a tad more than amd probably due to better fab processes.

>>59302858
because Intel told all the well known shills to not do that.

>>59304122
>due to better fab processes.
nah, it was designed to be low power for servers
arch is made for big bux first in mind, current launch basically beta test

>>59302955
As much as I hate JayZ, he did install his Ryzen chip on a custom loop straight out of the box.

http://hwbot.org/newsflash/4335_ryzen_platform_affected_by_rtc_bias_w88.110_not_allowed_on_select_benchmarks

does it mean non of the results matter now?

I have a 2500k running at 1.45v for like 2.5 years. Somehow even 4.6g isnt so stable with a z68 chip on a good gigabyte mobo. Ill drop voltage to 1.25 and keep oc at 4ghz. The gains are at most 5-6 fps. Ill try to underclock the chip to 3.1 boost and 1v so gf can have this pc when i get ryzen.

>>59304122
While kaby lake really has an excellent OC headroom, when comparing out of the box frequencies and turbo implementations, Ryzen doesn't lose much face against Kaby lake. Especially when your own considerations are taken into account (new arch, new proccess, "new" fab partner - glofo x tsmc)

>>59304300
no, it means that there will extra steps for validation. And this bug also affected everything but old AMD lines and anything older than Kaby/skylake.

>>59304300
>does it mean non of the results matter now?
For starters Windows doesnt work with SMT properly so right now all results are without SMT or with it DECREASING performance.
Also the RAM bugs that dont allow the fastest ram to be used

Fixing that shit alone has massive potential for improvement.
AMD massively fumbled with the launch.

It's reasonable to expect significant improvements once those issues are fixed. 10%, maybe 20%? Why not, considering that right now multi threading yields negative benefits and is supposed to do positive work.

>>59304093
>built like a tank
that made my eyes a bit moist. Kinda comparing them to old Ladas or 80's volvos. Shitty, ugly as fuck, but beloved and almost eternal.

>>59304472
Or to an old warship.

>>59304472
of course all your neighbors will never fail to throw snide and derisive comments at you when driving by atop their brand new luxury SUV's, but you lightly tap at your steering wheel saying we don't mind them, right girl, we'll just keep on goin' like we always do.

>>59304300

It depends on how tests were completed.

But it largely throws a monkey wrench in anything good or bad.

>>59304472
a supra would be a lot more appropriate of an example. I've seen 2000hp supras, you can basically throw whatever you want at them until the chassis torques in half

>>59302858
Everyone is OCing at 8/16. How about 4/8 4.5ghz+

When would you run into a TDP limit or a limitation of the architecture

I finally got my 1800x, MB and AM4 mounting bracket for my water block. I'm excited to see what I can get with my custom loop.

I don't really care about chip degradation and my build won't be for 24/7 use so I'm fine with pumping 1.45v+ vcore into it as long as I can keep temps under control.

Is there even a point to overclocking cpu's anymore? Its been like 10 years since my system was bottlenecked by my cpu. And even with gaming, video editing/rendering and model editing/photoshop/etc, I still barely go above 50% cpu usage. It seems like stability is much more worthwhile than raw speed, which with software today (save an extremely small percentage of pc users) makes next to no difference

>>59302858
Most chips hit 4ghz on all cores on average without LN2, even the 1700

>>59304572
so, 90% of youtube we can write off with 99% certainty

>>59302858
You realize the AIO many reviewers were using were the EK 280MM solution right? That's about as high end as it gets for AIO.

The fact is, Ryzen doesn't over lock as well. Get over it. It's a pretty great arch otherwise.

>>59305096
I'm kinda tired of tinkering. I hope they iron out xfr real good and i'll just let it pickup my slack. At least i'll stil have a justification to search for better cooling solutions.

>>59305096
I've kinda followed this route. Still got a 2500K but never been bottlenecked by it enough to need to overclock. I do have a large custom cooler on it though so temps under load never go above 55c.

>>59302858
kek someone expected a product of amd to not be a premium home heater

>>59305550
>>59305096
same here, who even cares

>>59302992
You have to go back.

>>59305550
Why didn't you buy the non-k part then you massive faggot?
It's not lambda calculus, it's just upping the frequency and voltage until you get stability. Jesus fucking Christ I'm mad.

Here's what I want to know - how come XFR is so incredibly shitty, and how come AMD can't OC single cores like Intel can to have substantially higher ST scores.

On top if that, how come no reviewer has disabled components and downclocked to simulate R5/R3 components, and how come no OC attempts on fewer than ALL COARS have been made public?

What the actual fuck, this is all basic bitch testing, guys

>>59305970
>how come XFR is so incredibly shitty
it's not "incredibly" shitty. It's that tech first iteration and is probably serving as a proof of concept, not to mention that double dipping on the boost freqs (it's what does, for 2 cores only though) seems very interesting on paper. Later reviews will possibly allow for bigger boosts when the rough edges are dealt with.

>>59305970

>how come XFR is so incredibly shitty, and how come AMD can't OC single cores like Intel can to have substantially higher ST scores.

Not all architectures or processes are equal. Hell you can call Intel's a shitty low clocking design when put up against monsters like POWER and bulldozer.

>On top if that, how come no reviewer has disabled components and downclocked to simulate R5/R3 components

A few have. IIRC computerbase.de is one of the sites that did.

>how come no OC attempts on fewer than ALL COARS have been made public?

Ryzen is seemingly temperature limited - pulling heat off the core to the IHS is a real limitation.

>>59305970
>>59306208
I'm wondering if the xfr defaults to core 0 and 1 on the ccx or if one could try to find the strongest cores to have them running on them.

and if it runs exclusively on the same ccx or if it can utilize cores from different ccx's.

there could be some relatively tweaks that end users could try to have them perform better.

>>59303087
Nope. 5.8ghz run was at 1.97v. Only a cunt hair away from 2v.

>>59303170
There is no such thing as a safe voltage level for a node size by its name. It depends entirely on the process characteristics.
Ryzen is pulling 1.475v for its 4.1ghz XFR state. It can handle 1.475v.
The only thing limiting system stability is temps. Too much localized heat at high voltages.

>>59303269
The 7ghz runs with Kaby Lake that I'm familiar with were only 1 or 2 cores, not the full chip, and absolutely not comparable to an 8 core chip with all 8 cores and cache active.
Before the NDA was even up we heard Ryzen was hitting 6ghz on one core, but generally no one is interested in those figures.

>>59303433
>I wonder if AMD is going to release updated enthuasist CPUs on 16nm FF+ process, that could result in 300-500 MHz gain.
No, 16nm FF+ does not show a ridiculously higher fmax than 14nm LPP, and there is absolute 0% chance of this ever happening. They're not going to spend millions of dollars porting Zeppelin to another foundry for no benefit.
The die and process are both tailored to the clock range of a enterprise chip.

>muhhh overclocker's dream

Fuck off you sorry cunt.

The 8 cores don't overclock well.
They explode in temperatures and thus also in energy consumption.

File: 1487893938157.png (277KB, 359x431px) Image search: [Google]

277KB, 359x431px

>>59304540
>tfw iowa and north carolina will never be reactivated and refitted

>overclocking
Gross

>>59302858
OP here.
Simple question. Has anyone stabilized at 4.2+GHz on custom loop / extreme AIO ?

>>59302955
>>59304281
i was gonna say that too, give it a couple weeks, Jay's got a twin rad ryzen loop built

although on the other hand, I...... rather like Jay

File: peak power.png (45KB, 960x540px) Image search: [Google]

45KB, 960x540px

>>59309171
>3.9ghz all cores static frequency
>less than 125w
If your chip could hit 4ghz on all cores stable then it'd still be under 135w if you knew what you were doing.
2.1ghz to 3.3ghz is where voltage/clock scaling is ideal, but power per core still isn't unreasonable at 4ghz.
With uncore pulling roughly 15w at stock settings in the 1800X, the power per core for sustaining a 3.6ghz base clock, and slightly higher call core turbo, is about 10w per core. Under a 100% AVX load the chip is only pulling 110w, which works out to be just under 12w per core. At 4ghz we see roughly 15w per core. Power per core wouldn't get unreasonable until you went higher than that.

For Kaby Lake the i7 7700K pulls 21.5w per core at 4.2ghz.
The locked i7 7700 SKU with a 3.6ghz base clock is 12.5w per core.

All things considered the clock scaling on Zen is beyond incredible. Really the only limiting factor is its temp ceiling brought on by relatively high required voltages. Though even at these voltages we aren't seeing extremely high power draw, and to understand that you need a basic knowledge of current, amperage, and how stepping works.

>>59307914
wew lad what do they have an inch of silicone to hold that fuckin 2volts in

>>59303170
pls do
I am not rich

>>59309592
Chipworks has an article looking at Samsung's 14nm LPE process in the Exynos 7420 that launched a couple years back. The fundamental difference between LPE and LPP is a refined higher intensity light source, they're still comparable in other metrics. Physical examination of a gate's structures can tell you a lot about the process without having to make any measurements. IE a low power transistor and a high performance transistor don't look the same. Higher performance parts you'll see thicker oxides deposited in some places, sometimes different insulation around source and drain wells, etc.
I think Global Foundries took the 14nm LPP process and make a Vt just for AMD's CPUs. I think the high efficiency at low locks, and the tame power draw at high voltages are indicative of certain characteristics that would be glaringly apparent at a glance.
High density, thick gates.

File: 1486132794570.jpg (15KB, 284x413px) Image search: [Google]

15KB, 284x413px

>tfw 6700k nice and cool at 4.5ghz with a 212 Evo

Could push it further probably.

>>59302858
>buys amshit
>thinks he has money for anything but stock cooler
Yeah nah.

Sorry, anon. This time, I can't give (You) a (You).

>>59302992
So the 1700 is the lowest of three no-cut binnings?

I wonder how much variability the 1700X and 1800X have at present.

>>59302929
Your CPU would degrade pretty fast with the amount of power you'd need to feed it. 14nm LPP FinFet wasn't designed for high frequencies. It's literally called Low Power Plus and was designed for <3 ghz. Past about 3.2 the power curve goes from a steady ramp and turns into a cliff.

>>59303999
Those were built on a much larger process. There's more space between the transistors, and a larger surface area to dissipate waste heat.

>>59310075
Low Power Early and Low Power Plus are marketing names.
LPP internally stands for laser produced plasma. LPE only existed because the light source wasn't yielding high enough output when the rest of the node was ready for risk production, so they substituted it.

Insinuating that a process is only good at lower frequency because of its name is just retarded. Keep your retarded assumptions to yourself.

>>59305796
The non k versions are usually clocked slightly lower.

>>59302992
3.8ghz at 1.25v is what TheSilt saw with his 1800X sample.
Seeing a 1700 hit over 3.9ghz at the same voltage is impressive.

>>59302858
>workstation CPU
>let's OC it out of existance

>>59310179
Fott from BitsandChips had a sample that scaled down even lower than Stilts from what I remember.
There does seem to be a decent degree of variability with every chip regardless of SKU, though they all have similar peak OC frequencies under conventional cooling. It does however show that the enterprise SKUs could have surprisingly high clocks.

File: BSOD RYZEN CRASH CPU EXPLODES (GONE NSFW) (HOUSEFIRE).webm (2MB, 640x360px) Image search: [Google]

2MB, 640x360px

>>59302992

>rypoo has a shitty overclocking

so sad...

>>59304395
Yep pretty much considering and budget restrictions they made a good chip.

I'm still hoping that Intel manages to produce a coffee lake x capable of holding even higher frequencies / IPC. Skylake x will probably stock will slightly increase upon broadwell. I just hope they include kaby efficiency into the production, the golden kabies manage to run 5ghz 1.18v

File: Power.png (20KB, 600x400px) Image search: [Google]

20KB, 600x400px

>>59311867
The 1800X pulls significantly less power than the i7 6900k. Even at a 3.9ghz all core OC it can pull less power. At every frequency it can realistically achieve under an air or water cooler Ryzen is more efficient than Broadwell where it counts. In rendering workloads even where the 6900k is ahead of top end Ryzen it isn't ahead by much, and its perf/watt is substantially lower.

It doesn't need to be a good overclocking chip beyond 4ghz. The 16 core Snowy Owl and 32 core Naples chips are going to displace Xeons in an awful lot of places.

>>59305230
I'm gaming/workstation it might hurt partners because boards can reap sales over oc features and there isn't much of a point in using custom coolers.

It just seems weird that new technology like Ryzen is pretty much evenly dead in the water with the ancient i7 platform. Sure the i7 is better at tasks that require fewer, stronger cores and the Ryzen is better at tasks that make use of all of it's cores.

But the margin is so insubstantial that it really doesn't matter in real life applications. I don't get /g/'s autism sometimes.

>>59311977
>But the margin is so insubstantial that it really doesn't matter in real life applications. I don't get /g/'s autism sometimes.
Ah, yes the "but it's so small, the margin doesn't even matter" narrative. Is that what's on your script this week shill?
"Just down play that amd is losing performance wise and down play the amount. Be sure to provide no specifics, numbers or charts, be vague and just downplay it with no proof"

>>59311977
I7 isn't older tech; Intel deliveries new architectures often, but they are largely incremental.

>>59311977
The Kaby Lake i7 7700k has a little less than a 17% clock speed advantage over the top Ryzen SKU going by base clocks, and IPC is somewhere between 5% and 10% higher nominally.
In some game benches we see the Kaby Lake part about 20% ahead of Ryzen. Thats pretty much exactly what you'd expect from something not explicitly well threaded. Its also with sometimes shoddy Windows load balancing, and all the random little issues from early BIOS versions on various boards.
A 20% performance delta isn't huge when you're looking at chips clocked that far apart. Some games the performance delta closes to 10% or less.

Everyone trying to spin this narrative of Ryzen being bad is just a shitposting fanboy. Its kind of miraculous that it performs this well over all.

>>59312036
>He made valid, positive points about both Rzyen and Intel
>This is considered choosing a side and shilling

Anon, I'm just a dude trying to build a new PC who can't make an informed decision on which CPU to get because of all the misinformation being regurgitated by BOTH parties.

Don't be such a turbo autist.

>>59312099
>Anon, I'm just a dude trying to build a new PC who can't make an informed decision on which CPU to get because of all the misinformation being regurgitated by BOTH parties.
Then go look at benchmarks from many different sites instead of listening to shills on /g/

>>59312074
Well said. Ryzen is a good cpu, it doesn't matter if it can't beat kabys on single thread, even big Intel's can't.

Once fixes are applied we need to check the performance on single thread against big Intel's.