AMD CPU and Platfrom Discussion Episode IV: A New Hope is Ryzen

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > AMD CPU and Platfrom Discussion Episode IV: A New Hope is Ryzen

«‹›2 »

Rastor: Jun 2, 2001

Even if Ryzen becomes a huge success it will take time to build up momentum. Some of the pieces still aren't even released. Just look at the posters in this thread who won't even consider it until it's available in ITX form factor.

# ¿ May 2, 2017 03:28

Adbot: ADBOT LOVES YOU

# ¿ May 6, 2024 00:58

Rastor: Jun 2, 2001

Today as part of their Financial Analyst Day 2017 AMD has shown off Naples chips, which will be marketed as EPYC.

http://wccftech.com/amd-unveils-epyc-cpus-32-cores-64-threads-datacenter/

# ¿ May 16, 2017 21:59

Rastor: Jun 2, 2001

Oh hey, I didn't notice that this week AMD released the AMD Optimizing C/C++ Compiler, AOCC. The Intel compiler is pretty well known for turning out better performance on their platform, so a similar move could be good for AMD.

quote:

https://www.phoronix.com/scan.php?page=article&item=amd-ryzen-aocc&num=5

While we were very excited when first hearing of AMD's AOCC compiler for potentially better performance on Ryzen, with this 1.0 release we found very little change in performance. In most workloads of the dozens of tests ran, the AOCC 1.0 performance ended up being right in line with upstream Clang 4.0. In a few cases AOCC was faster, but it was generally about three percent or less. In some of the cases where AOCC was faster, it was only to then lose out to the GCC compiler being faster yet.

Dammit, AMD.

Kazinsal posted:

We need an :amd: like we have a

# ¿ May 19, 2017 21:04

Rastor: Jun 2, 2001

NewFatMike posted:

Hot deets on that AMD RoG notebook.

Well, 80s nostalgia is big right now after all...

# ¿ May 30, 2017 15:39

Rastor: Jun 2, 2001

Malcolm XML posted:

Nah just create an ocp box and then the clouds will eat that poo poo up

cloud providers are the new drivers here

Cloud providers buy in the biggest quantities but at the same time for that reason they want to be sure of the performance and stability before signing the purchase order. Some design wins with HP/Dell/Cisco/IBM/etc. will no doubt be high priority for AMD right now.

# ¿ Jun 4, 2017 20:03

Rastor: Jun 2, 2001

Malcolm XML posted:

Lol cloud providers are like 95% driven by cost

Failure is built into the design dude. They just migrate loads around.

Azure used to use poo poo opterons since they were cheaper than xeons
Low margins ultra high volumes > than ultra high margin ultra low volumes for amd here. Sure the hospital datacenter will buy E7 but aws will buy 100x the epyc 32 core proc

Yeah, they're driven by cost, and if you've got nodes failing left and right your costs shoot up.

The big cloud providers are big movers, I'm just saying don't expect them to necessarily be first movers. Not on a brand new platform, anyway. Proven reliable opterons you can get for cheap is a different beast.

# ¿ Jun 4, 2017 20:48

Rastor: Jun 2, 2001

Oh no doubt there are test machines in labs, I agree there. But stuff like being unstable under heavy compiling loads is gonna have to be ironed out before the production order is placed.

The cloud providers don't care about five-nines reliability on any given node but they do need the poo poo to actually work, in production conditions (where they don't baby systems at all), under heavy loads.

# ¿ Jun 4, 2017 21:07

Rastor: Jun 2, 2001

Malcolm XML posted:

Oh you sweet summer child

Without violating NDAs, I know what I'm talking about

# ¿ Jun 4, 2017 21:27

Rastor: Jun 2, 2001

Dell Alienware signs exclusivity deal, to be the only pre-built gaming PC vendor to sell desktops with 16-core Ryzen Threadripper processors until the end of 2017:

https://www.techpowerup.com/234477/alienware-inks-ryzen-threadripper-16-core-exclusivity-deal-with-amd

(this does not block DIY customers from buying retail processors)

# ¿ Jun 19, 2017 16:08

Rastor: Jun 2, 2001

Here come the white box EPYC servers

http://www.anandtech.com/show/11567/tyan-announces-amd-epyc-tn70ab8026-server-1p-16-dimms-26-ssds-oculink

# ¿ Jun 22, 2017 22:05

Rastor: Jun 2, 2001

Here's a Gigabyte mini-ITX board for the AMD Ryzen socket AM4 platfrom

# ¿ Jun 27, 2017 16:35

Rastor: Jun 2, 2001

Threadripper is so bad it's good.

# ¿ Jun 30, 2017 02:12

Rastor: Jun 2, 2001

wargames posted:

now one for THREADRIPPER.

Measly Twerp posted:

I hope your happy, a PCB layout expert just killed themselves over this.

Certainly an engineering challenge. Just the Socket TR4, IO Shield, PCIe socket, and RAM sockets on the top, every other component on the back?

# ¿ Jun 30, 2017 18:28

Rastor: Jun 2, 2001

Measly Twerp posted:

Or; mount the CPU on a separate PCB and then mount that PCB into a slot on the motherboard.

Ooh, that's a good one. What kind of connector can you trust between those two boards, though?
Also you'd need to be specific about which coolers will be supported.

# ¿ Jun 30, 2017 18:31

Rastor: Jun 2, 2001

Second Sun posted:

That's the most childish bullshit.

Welcome to sales and marketing.

# ¿ Jul 12, 2017 17:02

Rastor: Jun 2, 2001

Sidesaddle Cavalry posted:

Have we seen the inside of a Threadripper package yet? I keep seeing this shot of Epyc and it's making me wonder if people are getting the wrong idea of where the dies are located under the heat spreader.

16 cores = 4 CCXs = only 2 dies, right?

I think that's right, Ryzen = 1 die, Threadripper = 2 dies, Epyc = 4 dies

# ¿ Jul 20, 2017 15:23

Rastor: Jun 2, 2001

Every Threadripper is actually an Epyc with two dies lasered off, you can reconnect them with a graphite pencil

# ¿ Jul 20, 2017 15:50

Rastor: Jun 2, 2001

That packaging is only cool if you open it by depressing a hidden catch which causes the processor to rip and tear through the front window

# ¿ Jul 24, 2017 16:05

Rastor: Jun 2, 2001

I can't wait for the first project build of an ITX system in the Threadripper box

# ¿ Jul 24, 2017 20:19

Rastor: Jun 2, 2001

Subjunctive posted:

AMD needs to acquire Tesla to get Keller back and land a design win for their AI ambitions. Leverage, baby.

Jim Keller is a ronin, when AMD needs him again he will be there.

# ¿ Jul 25, 2017 01:37

Rastor: Jun 2, 2001

PC LOAD LETTER posted:

You'll probably need to wait for single socket Epyc CPU and platform to come out if you want a white box work horse mobo.

If you can't wait for that then just buy TR and turn off all the LED's on your mobo of choice. HEDT on either AMD and Intel is at least as much about doubling down on RGB blinkenlights now as it is about having moar corezzz.

Mr Shiny Pants posted:

This! I wish they would make some no-frills mobo's like Intel used to do. I don't want lights, I want a quiet mobo with some solid components.

eames posted:

The early Gigabyte X399 leaks looked fine.

but somehow turned into this thanks to their marketing department:

Gotta say though, that's an excellent LPC ratio (LEDs Per Core). LED enthusiasts should be pleased.

Hey there's always hope

https://twitter.com/BitsAndChipsEng/status/878575914996695040

# ¿ Jul 26, 2017 12:44

Rastor: Jun 2, 2001

Maxwell Adams posted:

Bitwit on unboxing:

Gamer's Nexus on unboxing:

OK that's a pretty funny contrast

# ¿ Aug 3, 2017 15:48

Rastor: Jun 2, 2001

There were always going to be more than 250 engineering samples / mistakes

# ¿ Aug 3, 2017 19:14

Rastor: Jun 2, 2001

All of the RGBs... ALL OF THEM... and put them in a Thermaltake View 28 RGB case...

# ¿ Aug 4, 2017 17:42

Rastor: Jun 2, 2001

isndl posted:

How long before we get CPUs that use tiny little LEDs instead of transistors? Billions of them, all in your computer where you can't see them. But they're there.

http://www.megaprocessor.com/

# ¿ Aug 4, 2017 18:32

Rastor: Jun 2, 2001

Look at that performance difference! Definitely use this slide during the marketing presentation.

# ¿ Aug 7, 2017 16:32

Rastor: Jun 2, 2001

ArgumentatumE.C.T. posted:

So are other CrYpTocurrencies around now that haven't been dominated by the Chinese disaster factories? Frisbees that aren't in danger of roofs?

I should ask this in a different thread.

Cryptocurrency thread here:
https://forums.somethingawful.com/showthread.php?threadid=3824394

Mining goons seem to use an app which automatically picks the most profitable currency

# ¿ Aug 10, 2017 01:20

Rastor: Jun 2, 2001

Now that Google is pushing for Android apps on Chromebooks there has been a recent push for ARM-powered Chromebooks, notably the Samsung Chromebook Plus and the ASUS Chromebook Flip C101 which are both powered by the Rockchip RK3399 "OP1" chip.

Previously the Rockchip RK3288 powered the ASUS Chromebook C201, ASUS Chromebook Flip C100, CTL Chromebook J2, CTL Chromebook J4, Haier Chromebook 11, Hisense Chromebook, Viglen Chromebook 11, and Xolo Chromebook.

Also noteworthy, the Acer Chromebook R13 is powered by a Mediatek M8173C "CorePilot" chip.

# ¿ Oct 21, 2017 03:36

Rastor: Jun 2, 2001

kujeger posted:

You need to be root to write to the efivars.

edit: which is why this is so ridiculous.

The EFI explicitly presents these vars as writable. So they get set up as writable by root. But (some) EFIs can't handle them being cleaned out.
...and for some insane reason people are personally blaming Poettering for this.

If I write to the boot drive as root, the operating system needs to be reinstalled

If I write to the EFIs as root, the computer has been turned into a brick

There's a difference

# ¿ Nov 30, 2017 18:30

Rastor: Jun 2, 2001

It's a stack of assholes saying "I just need to engineer this to work, that issue is someone else's problem" the whole way down

# ¿ Nov 30, 2017 20:51

Rastor: Jun 2, 2001

The fact that AMD does have a marketing department and they come up with these names is actually funnier than the parody

# ¿ Dec 5, 2017 23:26

Rastor: Jun 2, 2001

Hey better one person with two keyboards than two people with one keyboard

https://www.youtube.com/watch?v=u8qgehH3kEQ

# ¿ Dec 8, 2017 23:11

Rastor: Jun 2, 2001

https://www.fool.com/investing/2017/12/11/intel-corps-whiskey-lake-revealed.aspx

quote:

In the same conversation, "chrisdar" also said (per a translation of his comments from Chinese to English) that Intel's upcoming 10nm manufacturing technology is "really miserable"

...

Intel appears to have, once again, tried to lead investors astray as to the health of its 10nm manufacturing technology and the timing of the launches of products based on that technology and its derivatives (e.g., 10nm+ and 10nm++). Considering that Intel has routinely claimed that it has a multiyear leadership in chip manufacturing technology -- something that, frankly, rings hollow in light of the company's product launches over the last several years -- this isn't a good look for the company.

# ¿ Dec 14, 2017 18:25

Rastor: Jun 2, 2001

Combat Pretzel posted:

Bleh, apparently if the CPU supports PCID, the performance loss can be reduced to next to nothing. Haswell and newer does support it.

These database benchmarks were posted in the Intel thread (WHY HAVEN'T WE MERGED THE THREADS YET)

https://www.postgresql.org/message-id/20180102222354.qikjmf7dvnjgbkxe@alap3.anarazel.de

code:

readonly pgbench (tpch-like), 16 clients, i7-6820HQ CPU (skylake):

pti=off:
tps = 236629.778328

pti=on:
tps = 220791.228297 (~0.93x)

pti=on, nopcid:
tps = 198959.801459 (~0.84x)


To get closer to the worst case, I've also measured:

pgbench SELECT 1, 16 clients, i7-6820HQ CPU (skylake):

pti=off:
tps = 420490.162391

pti=on:
tps = 350746.065039 (~0.83x)

pti=on, nopcid:
tps = 324269.903152 (~0.77x)

So with PCID, a 7% hit

Without PCID, a 16% hit

Now that's a database benchmark, which is one of the worst affected workloads, but it doesn't sound like PCID reduced to next to nothing

# ¿ Jan 3, 2018 16:50

Rastor: Jun 2, 2001

Paul MaudDib posted:

The mitigations available for an erratum and their impacts matter.

The numbers being thrown around are pathological test-cases and don't seem to be borne out in general instances. IO-heavy workloads are going to take a moderate dent here, but even then real-world workloads do not consist solely of running 'du' and loopback SELECT 1; operations 24/7 so this is still overstating the problem in general.

This will have less than a 1% impact on most applications and the real-world worst case will be probably a 5-10% impact on databases and compilation workloads. That's embarrassing and AMD will be nipping even closer on their heels in a few tasks, but it's far from the apocalyptic "50% performance loss!111!" numbers that are getting breathlessly thrown around.

That may go some ways to explain why the Linux kernel developers set the pti=on to be the default for all x86 processors in their patch rather than having an exception for AMD.

So it may become 'make your database faster on AMD processors with ONE WEIRD TRICK'

# ¿ Jan 3, 2018 18:12

Rastor: Jun 2, 2001

Cygni posted:

so the current Linux fix has the performance hit regardless of CPU manufacturer, do we know if thats the case with the NT kernel fix yet?

Paul MaudDib posted:

The really funny thing is that 50% number actually comes from running the patch on Epyc processors, not Intel (where the pathological case is 30%). In other words, a rough guess might be that this patch hits AMD twice as hard as it hits Intel. If they switch it on by default, Intel could actually gain relative performance

10 years from now we'll still be hearing about how Bernie Would Have Won if only Intel wasn't cheating, release the nopti benchmarks shintel!

It's really not the world's worst idea to just do it by default if the performance impact is that minimal, though.

Latest news is that AMD's request to turn off the PTI on their chips *will* be accepted into the Linux kernel, though it may go into 4.16 instead of 4.15
https://www.phoronix.com/scan.php?page=news_item&px=Linux-Tip-Git-Disable-x86-PTI

# ¿ Jan 3, 2018 19:01

Rastor: Jun 2, 2001

More fat for the AMD fire.

https://www.theverge.com/2018/1/9/16867068/microsoft-meltdown-spectre-security-updates-amd-pcs-issues

quote:

�Microsoft has reports of customers with some AMD devices getting into an unbootable state after installing recent Windows operating system security updates,� says a Microsoft spokesperson. �After investigating, Microsoft has determined that some AMD chipsets do not conform to the documentation previously provided to Microsoft to develop the Windows operating system mitigations to protect against the chipset vulnerabilities known as Spectre and Meltdown.�

# ¿ Jan 9, 2018 13:23

Rastor: Jun 2, 2001

a medical mystery posted:

So if I got a FX-4130 it won't have the boot problems with this patch? Windows Update says KB4056892 is gonna install on my next restart so I'm a little nervous as someone who's hardware-illiterate.

Microsoft already halted updates for the affected systems, so if you are being offered updates, it should be OK to proceed with them.

You keep everything important backed up anyway, right?

# ¿ Jan 9, 2018 16:26

Rastor: Jun 2, 2001

Rexxed posted:

Does this affect my 1.4ghz thunderbird with the screaming fan from 2000? I used that at some point to write an image to a compact flash card. It may not have been in this decade, but it's definitely mission critical!

Better switch to a 486, just to be safe
http://yeokhengmeng.com/2018/01/make-the-486-great-again/

# ¿ Jan 9, 2018 19:48

Adbot: ADBOT LOVES YOU

# ¿ May 6, 2024 00:58

Rastor: Jun 2, 2001

PerrineClostermann posted:

Thread 2: Ripper Boogaloo

Threadripper 3: Rip Threads with a Vengeance

# ¿ Feb 19, 2018 21:15

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > AMD CPU and Platfrom Discussion Episode IV: A New Hope is Ryzen

«‹›2 »