Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
EmpyreanFlux
Mar 1, 2013

The AUDACITY! The IMPUDENCE! The unabated NERVE!
No, Raja Kodouri. It's all been a long con since RTG got snubbed funds in 2014, and now as CEO of Intel he shall have his vengeance!

Adbot
ADBOT LOVES YOU

Otakufag
Aug 23, 2004
Jim Keller is gonna be Intel's next CEO actually.

SwissArmyDruid
Feb 14, 2014

by sebmojo
That would be the darkest timeline.

I imagine Keller being locked in a literal ivory tower, the CEO of the company, but unable to leave or work on chip design for fear of it getting into the hands of AMD.

Christobevii3
Jul 3, 2006

Measly Twerp posted:

I just moved back from Europe to Australia, but I managed to leave the part of my motherboard that the cooler screws onto behind. I've currently got my computer set up open air on top of the motherboard box with the cooler simply held down by gravity. The temperatures are actually quite good, hitting about 30C at idle.

Should I even bother asking ASRock if they can send me a replacement? What other options are there?

Which model? I just took one off my x470 and may work since I went with a corsair water cooler.

EmpyreanFlux
Mar 1, 2013

The AUDACITY! The IMPUDENCE! The unabated NERVE!
You can't contain Jim Keller, he'll just drag the entire company to his next gig like a bull dragging a hapless clown. Intel would just be his own personal fab for eleventy billion ARM designs.

Speaking of which I still want to see K17, wonder how it would compare to Cavium or Qualcomm designs.

Anarchist Mae
Nov 5, 2009

by Reene
Lipstick Apathy

Christobevii3 posted:

Which model? I just took one off my x470 and may work since I went with a corsair water cooler.

ASRock AB350 Pro4. What water cooler did you go with that doesn't require the back plate? I'd like to go back to water cooling as I'm sure in summer that'll be quieter than the stock cooler fan.

Christobevii3
Jul 3, 2006

Measly Twerp posted:

ASRock AB350 Pro4. What water cooler did you go with that doesn't require the back plate? I'd like to go back to water cooling as I'm sure in summer that'll be quieter than the stock cooler fan.

Corsair i80, it came with one. I'll try to remember to take a picture. Email my user name at gmail.com

GRINDCORE MEGGIDO
Feb 28, 1985


So who bought Ryzen+. and did you get 2700 or 2700x?

Canned Sunshine
Nov 20, 2005

CAUTION: POST QUALITY UNDER CONSTRUCTION



GRINDCORE MEGGIDO posted:

So who bought Ryzen+. and did you get 2700 or 2700x?

I bought a 2700X, paired with a ASUS Crosshair VII Hero.

GRINDCORE MEGGIDO
Feb 28, 1985


SourKraut posted:

I bought a 2700X, paired with a ASUS Crosshair VII Hero.

Nice board. Have you had success with ram speeds?

Kinda hoping I can crank the ram speed up when my parts come next week. I have some good ram ready to use.

CaptainSarcastic
Jul 6, 2013



ufarn posted:

Any last-minute advice before I attempt to fix a few crooked pins on my Ryzen CPU? I’ll be very careful, but more than that is needed when a 100% success rate is required.

I’ll be wiggling a toiny flat iFixit screwdriver to bring them upright, then I’ll use a rolled out boxcutter to align them properly. Still don’t know whether, when nor how they might break though.

That sounds a lot like the procedure I used to use salvaging CPUs at a recycling place - I used a probe and a standard razorblade. I was usually successful, too - hope you had good luck with this.

Canned Sunshine
Nov 20, 2005

CAUTION: POST QUALITY UNDER CONSTRUCTION



GRINDCORE MEGGIDO posted:

Nice board. Have you had success with ram speeds?

Kinda hoping I can crank the ram speed up when my parts come next week. I have some good ram ready to use.

Honestly I just bought a set of G.Skill 3200 MHz 2x 8 GB CAS 14 modules that have Samsung B-die, set ASUS’ equivalent to XMP to 3200 mhz, and let it be. Has been working great and haven’t had any issues so far.

GRINDCORE MEGGIDO
Feb 28, 1985


SourKraut posted:

Honestly I just bought a set of G.Skill 3200 MHz 2x 8 GB CAS 14 modules that have Samsung B-die, set ASUS’ equivalent to XMP to 3200 mhz, and let it be. Has been working great and haven’t had any issues so far.

I have some b die too. I'm really looking forward to this upgrade.

Klyith
Aug 3, 2007

GBS Pledge Week

ufarn posted:

Any last-minute advice before I attempt to fix a few crooked pins on my Ryzen CPU?

I fixed a bent pin with a mechanical pencil -- remove the lead and then the tip was the perfect tool to slide onto an individual pin and bend it back.

This was more than a decade ago so pins were bigger and with more space between them. But it might still work if you had a .5mm ultra fine rotring or something like that.

FaustianQ posted:

I really hope AM5 has a better retention mechanism. Nothing wrong with PGA, just there really shouldn't be a risk of damage when just removing the cooler.

Use cheap / old-school thermal paste and the old-school minimal quantity application style. The newfangled pastes that 'cure' after heat have a number of good properties and advantages, but easy release is not one of them. If you're changing stuff out often enough that this is a concern, better to just accept the 1 degree extra heat and make removal easy on yourself.

(Personally I dislike curing pastes just because they're not shelf-stable.)


The common story was that mobo manufacturers like the PGA because it's cheaper and has less returns on their side, but couldn't you spec a socket to have pins but also a bracket retainer like the LGA types? Though that's also 25c extra cost just to prevent the occasional user fuckup.

ufarn
May 30, 2009
I’d already watched that which, while largely great and reassuring, still ends with him breaking two pins. :smith:

He doesn’t mention how bent those two were either, which is a shame.

Same thing with the 5mm pencil; sounds good, but I’m not super sure anout the succes rate.

I’ll stick to the wiggle+align strat for now. I’ll keep the pencil as backup; the pins are bent so far down I’ll have to bring them upright first,.

ufarn fucked around with this message at 11:12 on Jun 25, 2018

Palladium
May 8, 2012

Very Good
✔️✔️✔️✔️

Klyith posted:

I fixed a bent pin with a mechanical pencil -- remove the lead and then the tip was the perfect tool to slide onto an individual pin and bend it back.

This was more than a decade ago so pins were bigger and with more space between them. But it might still work if you had a .5mm ultra fine rotring or something like that.


Use cheap / old-school thermal paste and the old-school minimal quantity application style. The newfangled pastes that 'cure' after heat have a number of good properties and advantages, but easy release is not one of them. If you're changing stuff out often enough that this is a concern, better to just accept the 1 degree extra heat and make removal easy on yourself.

(Personally I dislike curing pastes just because they're not shelf-stable.)


The common story was that mobo manufacturers like the PGA because it's cheaper and has less returns on their side, but couldn't you spec a socket to have pins but also a bracket retainer like the LGA types? Though that's also 25c extra cost just to prevent the occasional user fuckup.


I think the trick to avoid ripping the pinned CPUs out of the socket is to twist the HSF to unglue the TIM before pulling it away.

The LGA = more expensive story gotta be a bullshit excuse though. AMD should had used it for AM4 for the get go.

Palladium fucked around with this message at 12:55 on Jun 25, 2018

Munkeymon
Aug 14, 2003

Motherfucker's got an
armor-piercing crowbar! Rigoddamndicu𝜆ous.



Klyith posted:

I fixed a bent pin with a mechanical pencil -- remove the lead and then the tip was the perfect tool to slide onto an individual pin and bend it back.

This was more than a decade ago so pins were bigger and with more space between them. But it might still work if you had a .5mm ultra fine rotring or something like that.


Use cheap / old-school thermal paste and the old-school minimal quantity application style. The newfangled pastes that 'cure' after heat have a number of good properties and advantages, but easy release is not one of them. If you're changing stuff out often enough that this is a concern, better to just accept the 1 degree extra heat and make removal easy on yourself.

(Personally I dislike curing pastes just because they're not shelf-stable.)


The common story was that mobo manufacturers like the PGA because it's cheaper and has less returns on their side, but couldn't you spec a socket to have pins but also a bracket retainer like the LGA types? Though that's also 25c extra cost just to prevent the occasional user fuckup.

I'm going to try https://amazon.com/gp/product/B07CK9SHZG/ next time I swap out my CPU, which will probably be when Zen 2 comes out.

GRINDCORE MEGGIDO
Feb 28, 1985


Mechanical pencils used to be good for this, but I don't know if modem pins are too smol.

ufarn
May 30, 2009
It’s kind of crazy to me how thin they are - and how easily they bend. You can fit paper in between then, but credit/plastic cards are too thick.

At least you’d think they had something set up to repair broken pin.

Stanley Pain
Jun 16, 2001

by Fluffdaddy

:goleft:

SwissArmyDruid
Feb 14, 2014

by sebmojo

Palladium posted:

I think the trick to avoid ripping the pinned CPUs out of the socket is to twist the HSF to unglue the TIM before pulling it away.

The LGA = more expensive story gotta be a bullshit excuse though. AMD should had used it for AM4 for the get go.

Ding ding ding ding ding! One of those old-rear end tricks that people have slowly forgotten because of the ubiquity of Intel LGA sockets.

Also, no, LGA = more expensive is not bullshit. It really is more expensive, in that it shifts the cost over onto the motherboard manufacturer, as opposed to the chip manufacturer.

AMD, not being in a position to dictate terms after the long and many disastrous years of construction cores, probably did it as a concession to their motherboard partners.

Based on their improved market positioning, I would expect LGA-type sockets for AM5 going forward in 2020.

And yes, mechanical pencils are still good for fixing bent PGA socket pins. Just be SUPER loving slow, speed is your enemy and will lead to shearing off pins going back in the opposite direction.

SwissArmyDruid fucked around with this message at 22:07 on Jun 25, 2018

Cygni
Nov 12, 2005

raring to post

I think intel has/had some patents about the socket too which made some people scared.

In the end, having dealt with bent pins in an LGA socket which is a pain in the rear end, and still having to deal with crazy mounting brackets for big coolers, I don't feel like it's really that superior. Other than its cool to open it up, looks like a spaceship and poo poo aw yeah.

ufarn
May 30, 2009
Happy to report that the pins were unbent. :unsmith:

I have yet to install everything so I’m not going to jinx it by saying everything will be fine, but I literally couldn’t tell the bent pins anymore by the end.

I also quintuple-tested the mounting fit before I locked it in.

Here is my mad-professor lab for anyone interested:



I had the .5mm mechanical pencil ready didn’t use it much. But having it there was nice.

Thanks to everyone for the reassurances and help.

I’m gonna go start breathing now and rehydrate after all the surgery flop sweat.

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE
So regarding Ryzen steppings... 1000-series Ryzen and Threadripper are B1, Epyc is B2. One theory at the time was that B2 might be the "mirror" die that Epyc uses 2 of.



I checked today and Ryzen 2000 also shows as B2 stepping.

Since they didn't change the library, is it feasible to just take the 14nm die and run it on 12nm? 12nm is really a 14+ after all. And if so, can they use the package substrate to "flip" the die to the proper pinout?

Anime Schoolgirl
Nov 28, 2002

LGA also gives you ~15% more pins for the space before things start getting unmanageable on pin/land or pin/hole placement.

EmpyreanFlux
Mar 1, 2013

The AUDACITY! The IMPUDENCE! The unabated NERVE!

Paul MaudDib posted:

So regarding Ryzen steppings... 1000-series Ryzen and Threadripper are B1, Epyc is B2. One theory at the time was that B2 might be the "mirror" die that Epyc uses 2 of.



I checked today and Ryzen 2000 also shows as B2 stepping.

Since they didn't change the library, is it feasible to just take the 14nm die and run it on 12nm? 12nm is really a 14+ after all. And if so, can they use the package substrate to "flip" the die to the proper pinout?

Yes, that's what AMD said they did, Ryzen 2K has gains the benefits of of the optical improvements only and that results in a bunch of dark silicon. Don't know about the second part.

Is you're theory that Epyc has been manufactured on 14nm+/12nm this entire time, and Ryzen 2K is just cast offs?

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

FaustianQ posted:

Yes, that's what AMD said they did, Ryzen 2K has gains the benefits of of the optical improvements only and that results in a bunch of dark silicon. Don't know about the second part.

Is you're theory that Epyc has been manufactured on 14nm+/12nm this entire time, and Ryzen 2K is just cast offs?

I made that purely as a statement of (relatively strong) fact. I wasn't aware of what stepping Ryzen 2000 was on at all. Wikichips just listed it as stepping "2".

Optical improvements on B1 != optical improvements on B2 though. And you would think that would be like, a different stepping or something, if it was actually different (like, C1 and C2). And I'm not even sure that you can actually just run a 14nm die on 14+, even without the library changes (AMD has said they did not change the library).

But yes, I had that thought when I was taking the pupper for a walk today. What if the reason Epyc didn't really ramp until Q1 2018 was they were waiting for 14+/12nm? Maybe they announced in Q2/Q3 2017 when they were just getting engineering samples. It also explains why Epyc isn't getting the Ryzen 2000 treatment... it already is a Ryzen 2000.

I'm not going to die on that hill, too many unknowns there, but it sure is a thought.

The alternate theory for the delay that I consider plausible is the segfault bug, and Ryzen 2000 doesn't seem to fix that. There are people in the AMD Community thread reporting segfaults on Ryzen 2000 processors, with ECC RAM. Not enough that I am willing to declare it definitively still a thing, but if AMD taped out in Q1/Q2 2017 they probably wouldn't have known about it. Given that there has been no additional stepping for Epyc, that says they weren't waiting for a new stepping.

Paul MaudDib fucked around with this message at 05:06 on Jun 26, 2018

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!
The page you've linked says that the kill-ryzen.sh shell script was broken under the Linux distro he's used, not the Ryzen 2xxx CPU itself. The guy mentions after fixing the script, everything was dandy.

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

Combat Pretzel posted:

The page you've linked says that the kill-ryzen.sh shell script was broken under the Linux distro he's used, not the Ryzen 2xxx CPU itself. The guy mentions after fixing the script, everything was dandy.

The guy mentions that even after his patch he had two lockups from USB stick, one on kill-ryzen and one just on memtest (but not one that his ECC RAM caught - implying a problem from his cache/memory controller, which is the classic segfault problem).

quote:

+Ubuntu 17.04 is discontinued and Ryzen 2000 series CPUs do not seem to work without adjustments. Same goes for Arch Linux. I have simply substituted gcc7.1 with gcc8.1 and manual dependency installation. This way the script can be run with gcc8.1. Unless Ubuntu 17.04 is not working for you it is *not* recommended to use the updated version for the sake of reproducibility.

quote:

After fixing the script to work again on ubuntu-18.04 (uncomment the section that installs the build tools in ubuntu). I was able to run it on my Ryzen 2700 / ASUS Prime PRO + 16GB of Crucial DDR4 Unbuffered ECC. After ~36 hours of testing (24 contiguous), all is well (at least when I test from installed ubuntu). I had two lockups when testing from a USB stick. One in kill-ryzen.sh one in memtest86. Not sure why for either.

If you can't build and run it, that doesn't count as a "test". Building it on a new version of GCC with different optimization pathways (moving from 7.1 to 8.1) also sounds pretty suspicious, i.e. this is something that might break the kill-ryzen test since it relies on specific micro-instruction behavior. Let alone some manual dependency installations. He even says as much - "not recommended to use this for the sake of reproduciblity".

On the proven version, run from a USB stick, he is having problems, no?

(I consider a sample size of 1 to be quite weak evidence anyway, and that's all I've seen so far on kill-ryzen. But I've seen a few people crashing with Civ5/BF1/Rainbow 6 Siege, that is not replicatable on Intel chips (even of equivalent core count), in a similar way to some problems that were resolved by segfault RMAs on Ryzen 1000 (often ameliorated by disabling SMT and/or opcache, just as with segfaults), so I think there is mild anecdotal evidence that they are still susceptible. If they were really taped out in Q1/Q2 2017... yeah, AMD probably didn't know.)

Paul MaudDib fucked around with this message at 09:15 on Jun 26, 2018

repiv
Aug 13, 2009

:drat:

Truga
May 4, 2014
Lipstick Apathy
:five:

EmpyreanFlux
Mar 1, 2013

The AUDACITY! The IMPUDENCE! The unabated NERVE!
"I just got promoted!"
"Nice, what's you're new position?"
"Senior Manager of the Sick Burns Department"

Mister Facetious
Apr 21, 2007

I think I died and woke up in L.A.,
I don't know how I wound up in this place...

:canada:
Didn't know they were recruiting from the 'POS... :eyepop:

SwissArmyDruid
Feb 14, 2014

by sebmojo

Holy poo poo, YOSPOS is leaking.

e:fb

Stanley Pain
Jun 16, 2001

by Fluffdaddy
Epyc in ways I cannot even comprehend yet.

HalloKitty
Sep 30, 2005

Adjust the bass and let the Alpine blast

AMD's fuckin' going for it, no shame. Why the hell not?

BangersInMyKnickers
Nov 3, 2004

I have a thing for courageous dongles

Is anyone actually using those in a large production environment? Every time I go to look at AMD server processors when they make a big push in to the market it turns out there is some horrible, crippling flaw in the chip or chipset and nobody has the nerve.

Subjunctive
Sep 12, 2006

✨sparkle and shine✨

Microsoft is using them for Azure, and Baidu for their cloud stuff, but I don't think the exact scale of those deployments have been made public.

BangersInMyKnickers
Nov 3, 2004

I have a thing for courageous dongles

I'm more curious about medium sized enterprise's experiences with them. Small scale has loose requirements and are shopping on price so they probably won't care, and the huge providers have such purchasing power that AMD will throw engineers at a problem to help them figure stuff out. Everyone in the middle of those tends to get screwed when they opt for an AMD PowerEdge or whatever when a nasty problem surfaces.

Adbot
ADBOT LOVES YOU

Subjunctive
Sep 12, 2006

✨sparkle and shine✨

Medium-sized deployments tend not to make news, so I don’t know that the information is really available.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply