RYZEN SUPPORTS ECC MEMORY

File: 4d6.png (102KB, 200x254px) Image search: [Google]

102KB, 200x254px

IT'S OVER
SHILLTEL BTFO

>>59024805
>googleusercontent.com
>googleusercontent.com

looks legit

>>59024831
It's google's cache of Asus's own website, so yes, it is very legit.

What's the big deal with ECC?

>>59024805
Cool.

I don't think I've ever encountered a situation where I wish I had ECC, but it's nice nonetheless.

Ryzen is a forced meme.

>>59024862
Its something Intel locks out and the only real reason is really out of spite to drive Xeon and workstation chipset sales

>>59024862
Higher density modules, high precision workstation applications.

Theoretically, if the board supports it, you could put ~512GB of DDR4 RAM into a board with Ryzen in it (4x128GB ECC modules).

>>59024925
Does this mean we can use registered ECC DDR4 modules when they start selling them off cheap? You can get 4gb reg ddr3 as low as $5 vs $25 for the normal stuff

Why is non-ECC listed twice, I don understand?

>>59024973
Buffered vs Unbuffered, i'd wager.

>>59025011
meant for
>>59024989

File: chrome_2017-02-19_22-37-43.png (5KB, 559x83px) Image search: [Google]

5KB, 559x83px

>>59024805

Windows 7/8 32 and 64bits and Win10 32bits BTFO!

>>59025033
Annnnnnddd dropped

>>59025033
:/

>>59025104
>>59025108
As much as I would like Windows 7 to stay this shit was coming. Leave it to Microsoft to just not give a fuck to put out a new service pack to support this shit

>>59024831
filtered

Is ECC actually useful for consumer use or is it only worth it for people who are running servers on their machine?

>>59025480
Only really useful in certain use cases, the big thing here is the fact AMD is not telling you to fuck off and buy there server/workstation parts like Intel does

>>59025480
It's good for prosumers that run servers on their machines.

>>59025480
Basically AMD tells you not to fuck off for Opterons if you want consumer CPU-based workstation.

>>59025537
>>59025545
>>59025602
Right, but as for use cases in which ECC makes sense, are they only server-related applications or could it be useful for things such as intensive audio/video editing/rendering? I'm guessing ECC would be redundant for games too.

>>59025644
ECC would make sense for NAS/pfsense which do not need server grade CPUs or anything over 4 cores really

>>59024925
That's only registered you moron
It's noticeably slower and never in the past been supported by consumer boards/chips.

>>59025033
K V M
V
M

It's not like every AM4 mobo is going to support ECC, though. I think ASUS would support it for sure.

Hasn't AMD always supported ECC on everything?

>>59027800
IIRC those APUs lack ECC.

>>59024805
i know xeons use ecc memory but....what use will i have from it?

>>59027866
24/7 server use

>>59025033
Actually that should be a asus issue because biostar listed their X370 with drivers from w7 x86 onward, at least that's for the X370GT3

>uses the word ECC
>not explaining what ECC is in the title

why do you shit post? trying to lower my productivity by making me google what the fuck ECC is?

>>59028789
>Not knowing what ECC is in the first place
>>>/v/

File: 1481763072137.jpg (56KB, 907x661px) Image search: [Google]

56KB, 907x661px

>>59024831
>Intel
>retard nigger

what are the chances to see core boot on zen

>>59028789
If you don't know what ECC is, firstly your time is worthless so 'productivity' being lowered is meaningless, and second you shouldn't be on /g/. Fuck off.

File: 1438815048851.png (99KB, 236x176px) Image search: [Google]

99KB, 236x176px

>>59024805
Welp that settles it.
Going Ryzen.

>>59028969

Good, but slow.

https://www.phoronix.com/scan.php?page=news_item&px=AMD-Zen-Will-It-Coreboot

They have said all future CPUs will support it in the past. But who the fuck knows.

File: 1432264671094.jpg (21KB, 640x480px) Image search: [Google]

21KB, 640x480px

>>59027797
>Linux kernel module: On.
>ECC active

>>59025644
>>59028789
Due to the extremely volatile nature of RAM, you invariably get errors every once in a blue moon, although it's so rare and limited in scope that for consumer systems it's not worth considering. No one will care if an image you left open in Photoshop overnight will have the color of one pixel change slightly, and a random system crash can be chalked up to bad programming or something you did.

However as you scale up in memory, from a few to several hundred GB of RAM, and have it running 24/7, this multiplies the chance of an error happening considerably. And in a server room of many terabytes of memory you basically have errors happen every second. If you're running a simulation, this could be catastrophic for the results, or if it's a database, you will have wrong data be recorded to disk, even if it's one digit off it could still mean a bank loosing a million bucks worth of customer cash for no reason.

>>59024862

Servers, reliability in general if you don't want extremely fast performance.

>>59024862
cheap af

>>59025033
vga passthrough.

File: ௵.jpg (21KB, 372x260px) Image search: [Google]

21KB, 372x260px

there is so much hype for Ryzen I expect people will legit kill themselves if AMD disappoints yet again

>>59029526

>Due to the extremely volatile nature of RAM

That is easily the dumbest fucking thing that's been said on this board in the last oh, 12 seconds.

>stupid fucking people will be the death of us all

>>59029526
>If you're running a simulation, this could be catastrophic for the results,
To expand, chaos theory expands the results of errors massively. ECC ram is used in weather simulations, because a computation error right near the start would compound on itself over and over again and produce completely inaccurate weather predictions.

>>59031819
Well if Ryzen fails, AMD is dead and intel will strangle x86 for the rest of time.
Basically meaning that ARM takes over and there's no actual reason to continue living.

>>59024805
Didn't the FX line support ECC too? I'd be surprised if Ryzen didn't.

>>59024831
Filtered

>>59024805
what do I do with my 32gb of 1333ghz ram

>>59032193
Use it. Ryzen supports ECC; it doesn't require it.

>>59032164
They did support ECC, for the same reason. Consumer CPUs are just chips that didn't quite make the cut for the enterprise Opterons. The memory controller has the additional bits for ECC, it just normally isn't used by consumer boards.

>>59032201
Hes not sticking DDR3 in a socket AM4 board.

AMD enables everything on every chip pretty much. They're not greedy like Intel where they strictly segregate features into different segments of the market and make it impossible to get what you want.

>>59032223
He never said it was DDR3

>>59032229
Use your brain. Try. Just once.
He doesn't have 1333mhz DDR4.

>>59032241
Quit being a fucking dolt. Nearly nobody knows all of the possible DDR4 frequencies off the top of their heads.

>>59031819
Since I'm still on Core 2, even if it's a massive disappointment it'll still be an upgrade for me.

>>59032270
The minimum spec is 2133mhz. Its painfully obvious that he was talking about DDR3. Seriously get the fuck out of here and don't come back.

>>59025033
very good, they shouldn't support obsolete inferior trash used by only loud minority autists, called win7 but no win 8.1 support is kinda shit

>>59032291
Again, see >>59032270
>Nearly nobody knows all of the possible DDR4 frequencies off the top of their heads
Eat a dick, sperglord.

after seeing regular ddr4 prices the other week i dont even want to look at ecc prices

>>59024862
you can leave a machine on all day, every day and not worry that bad RAM or random bitflips won't go undetected/uncorrected.

ECC modules are more expensive than non-ECC and generally aren't sold in the higher speed grades, but it's a very nice feature in workstation chips for those who want to built their own file servers, etc., without paying the Xeon tax for support.

>>59032320
This is unbuffered, so it won't correct - but the machine will know the memory is wrong and can act accordingly rather than writing out faulty bits.

Technology exists to prevent silent data corruption.
Do you want it?

>it's not """needed"""
>something something servers
>muh games
t. /v/ cr3w -=[MLG]=-

>>59032308
ECC is usually always just about 10% more expensive.

File: daaamn.gif (601KB, 450x253px) Image search: [Google]

601KB, 450x253px

>>59024805

>>59029526
>once in a blue moon
>so rare
Meanwhile Google released data showing about 5 single bit errors in 8 Gigabytes of RAM per hour on the high end. If you're willing to risk that with the assumption that "one pixel in an image can change slightly" (never mind the fact that if a bit changes there's a good probability that it's going to be in the high bits and cause a greater change than "slight") then you must not give a fuck about what your computer is writing to your disks. Bit flips in the kernel memory will probably lead to system crashes too, which are pretty fucking annoying unless you're autistic enough to save all the time.
>limited in scope
An arbitrary bit in your RAM can change its value. It's basically unlimited in scope.

Any details on the chipsets? Intel has been jewing us for years on the fucking chipsets.

>>59033283
Dude they announced chipsets back on CES. Just google it.

>>59033255
Ah, the old 'cosmic ray' attack.
Rating: Severe
Description: Vulnerability exists in the case where attackers can control cosmic rays to arbitrarily flip bits in memory.

>>59033309
>Attack
No one said attack, everyone is implying the potential for massive data loss.

>>59033337
There a 0.000000000078125% possibility that one random bit will flip. That's not very massive.

>>59033410
B...but what if it's the 'delete all data' flag?

>>59032396
what are you smoking?
buffered/unbuffered is purely the difference between an extra array of latches right before the bus that enables more DIMMs per channel.

ECC is about using more SDRAM lanes/chips (9 or 18 chips with 72b total instead of 64b) to store redundancy information that the host memory controller verifies. Every write sends the extra bits, every read sends them back, and the memory controller can autonomously read the entire SDRAM array ("scrubbing") to preemptively detect and fix errors.

They're strictly orthogonal things, even if buffering is almost never used on non-ECC sticks.

>>59024862
The consumer use would be for a home built NAS

>>59033448
The most probable thing to be affected by random bit flips is the operating system. It's the only part that is up at all times while the computer is on. User applications are opened and closed on demand so their memory is constantly being refreshed. It's possible but the window for an error to occur and then not be corrected is much smaller.

>>59024862
>What's the big deal with ECC?
People think it's to prevent random cosmic ray bit flips, but it's really more about detecting and correcting for subtly defective SDRAMs long enough to replace them.

Various large scale trials have shown that the majority of errors in enterprise come from flaky bit cells or bit column sense amps that fuck up once a day or week, i.e., not enough to be guaranteed to be caught in manufacturer QA or an initial torture test but enough to lead to probably bugs/data corruption/crashing in a server.

>>59033410
Did you fucking bother to read what I posted?
https://research.google.com/pubs/pub35162.html
The actual, field error rate is orders of magnitude above the old, assumed error rate.
>0.000000000078125% possibility
That means a lot when you don't have an interval assigned to it. 0.000000000078125% per 0x10^-32 picoseconds?

>>59033609
5bits out of 64,000,000,000bits = 0.000000000078125%

Not sure if I did the math right but there's a very small chance of any one particular bit flipping.

File: cat.png (73KB, 1440x900px) Image search: [Google]

73KB, 1440x900px

>>59024805
>4 DIMM max 65GB
I can't run exchange server on this

>ryzen doesnt have igpu
>HAHAHA AMD BANKRUPT ETC
>ryzen comes with ecc support
>y-you dont even n-need it goy
>w-whats ecc????

>>59033651
No, 5/(64e9) = 78e-12 = 0.000000000078 = 0.0000000078%

But you calculation is nonsensical in any case.
Like the anon above said, you need to express errors in terms of rate per time or proportion of bits transferred. Beyond that, you also need to consider the nature of stored data and that a single bit flip will ruin an entire 64b pointer, etc.

>>59033651
>any one
But your system can use a lot of that RAM depending on what you're doing. Caching to minimize "wasted" RAM will also open up more data to memory errors.

>>59033800
don't worry famalam, you're allowed to stick 64GB DIMMs in the other 3 slots too.

>>59033800
didn't they charge you by the core? how umch did you pay for windows?

File: cores.png (117KB, 1920x1080px) Image search: [Google]

117KB, 1920x1080px

>>59033857
Its just Offendows 7 Enterprise. Max 2 physical CPUs and unlimited cores. (this image is a different server from other screenshot)

File: IMAG0755.jpg (889KB, 2048x1536px) Image search: [Google]

889KB, 2048x1536px

>>59033842
64 gigs per DIMM?

Things just got interesting.

>>59033894
>20c/40t x2
Noice

>>59033800
MORE CORES

>>59024862
meme

>>59031960
>don't worry guise, just wait for zen+!

>>59033296
>B350 "mainstream" chipset
>4 SATA ports total
Am I the only one who likes to have 6 ports minimum? The enthusiast boards will have 6 but I'm sure those will be $200+.

>>59033981
Nah, X370 starts at very honest 129 burgers.

>>59032306
>nobody knows
gtfo /v/shit!

>>59033981
There's two included with the Ryzen CPU and Stoney Ridge APUs themselves.

>>59033800
>>59033894
>>59033908
32 GB DIMMs are bad enough, but 64 GB is in pure "money is no object territory". You would be literally better off paying more for a Naples MCM with cores you didn't need just for more memory channels/slots.

Honestly, get a Broadwell (-E or Xeon) just for the extra channels if that's your main concern.

>>59033991
Memorizing a bunch of easily googleable numbers isn't a requirement to post on /g/. And I didn't say nobody knows, dumbass. I said most don't, which is definitely a fact.

>no iGPU
FINISHED & BANKRUPT!

>>59033981
> t. no-NAS club charter member

>>59034017
>basic tech literacy isn't required for taking about technology
>most retards such as myself just don't know

>>59034013
>but 64 GB is in pure "money is no object territory".
Power use of a few TBs of memory is huge, the less DIMMs the better.

>>59033981
SSD + 3x 10 TB HDDs not enough?

File: 1485965107937.jpg (69KB, 611x548px) Image search: [Google]

69KB, 611x548px

>>59033939
>tfw its not E7-8890 v4 x2

>>59025104
It does work on w7 though.

>>59034020
let's everybody laugh at the basement dwelling retard

>>59024805
>based ryzen supports win7
>no wait, JUST A PRANK BRAH

>>59034048
RAID.

>>59034074
Everything fucking support win7, it's just there will be no official drivers for chipset and whatnot.

>>59032308

32x2 sticks 950US$

For a 330 dollar processor... 120 dollar mother board 550 dollar GPU.

>>59034020
AMD has a line with iGPUs, they're called APUs. I'm sure like previous generations some motherboards will have iGPUs as well. The fact that these don't have them on the die means literally nothing.

>>59034045
Quit being a fucking twat.

>>59034054
>24c/48t
Not at 7 fucking grand a chip though.

>>59034048
My setup is a bit convoluted

SSD for Linux
SSD for Windows
HDD in NTFS for Windows
2 HDDs in ext4 for Linux
BDR drive

>>59034118
X370 then, or a PCIe SATA controller

>>59033981
just buy a fucking SAS card or something jesus

>>59025033
>>59025104
>>59025108
How come I can still install Windows 95 on a modern motherboard when the motherboard doesn't support Win'95? There simply is no company support for any troubleshooting problems, but it works okay otherwise.
Everyone will be using Win7 without problems with Ryzen. Don't fall for fake news.

>>59034092
More like less than $300, drama queen

>PRIME PRO
When will this gamer shit end. Old mid-high end hardware was pretty plain, and they were beautiful. Heatsinks shouldn't look like flames or guns or whatever it is gamers think is cool.

>>59034118
Be a big boy, get/build a NAS, and get a m.2/NVMe/PCIe SSD.

File: g1.assassin.jpg (498KB, 1600x1012px) Image search: [Google]

498KB, 1600x1012px

>>59034196

>>59034217
Holy fuck.

ECC RAM prevents the inevitable future where an actually good version of Windows 10 is written to RAM by random errors

Why would you try to prevent that, anon?

File: grenade_mouse.jpg (38KB, 520x442px) Image search: [Google]

38KB, 520x442px

>>59034232

>>59034244
That's retarded, not cringy.

>>59034244
what in the fuck is this shit

>>59034270
A mouse

File: 1432000762827.jpg (42KB, 417x515px) Image search: [Google]

42KB, 417x515px

>>59034244

>>59034274
makes me want to kys myself

>>59033255
>saving manually

Why do you use shit tier software that doesn't autosave for you ever n seconds?

It's literally a one time job of setting it up and then you're safe for hte rest of time.

>>59034162
Except you run out of PCIe lanes if you do that. AM4 will have AT MOST 8 lanes of PCIe 2.0 split between several PCIe 1x and PCIe 16x (4x electrically). And even then, most motherboards are going to dedicate those PCIe lanes to other chips for added "features". If you want to use two discreet cards, you literally won't have enough room to add an HBA card that isn't a chink-shit SATA port multiplier.

>>59034072
>waiting evolved

>>59034347
Then buy a server motherboard.

>>59034379
But my X99 motherboard can support that and overclock my 5960X, so go suck a dick

>>59034408
Why are you this upset? If the product doesn't suit your needs don't fucking buy it.

>>59034408
I guess enjoy your 970 SLI instead of a proper Titan XP setup and the accompanying stutter.

>>59034093
>APUs
>high performance
>I don't have to know anything about tech
lad...

>>59034447
APUs have better GPU performance than the IGPUs that Intel has though.

>>59034408
Your chip you paid 3x for? Ok I guess you could be smug about a few extra lanes for $1k

>>59034408
That's the difference between consumer and prosumer/server/workstation hardware tho. AM4 will be consumer stuff that competes with intels LGA1151 consumer stuff.

It could cover the bases for people that simply went LGA2011v3 for more cores, but for those that need more expansion cards.

Will probably have to wait until they release workstation/server grade stuff if you're interested in a sidegrade.

>>59034552
>they are shit CPUs tho
ayyyyyyy lmao

>>59034408
>i-its not good enough
>HAHAHA MY SHIT IS BETTER WAT R U POOR??? HAHAHAHA

fucking intel fags kek

>>59034591
Just wait for Raven Ridge. Then you can neck yourself.

>>59024805
Fuuuuck. I'm still waiting for the 16c/32t platforms to be announced.

> 4 DDR4 channels
> 8+ DIMM slots for up to 512 GB
> 64 PCIe lanes

Hell, I'm still praying that there are Opteron SKUs that expose the rumored 10 GbE controllers.

>>59034713
Dont forget there's the rumor of the 32c/64t CPUs floating around.

ECC support is useless for a gaming computer, its only useful on Servers and Scientific computers something that Intel has 99% of the market cornered with their XEON line. Intel is smart enough to not force idiots into buying useless features.

>>59034752
That's not rumors, top Naples is certainly 4-die MCM.

>>59034767
>Intel is smart enough to not force idiots into buying useless features.

reminder that Intel has to literally chops pieces off consumer CPUs with lasers to prevent their idiot consumers to not buy higher-end accessories.

File: 1482851277387.png (297KB, 448x468px) Image search: [Google]

297KB, 448x468px

>>59034767

>>59034347
>>59034408
>wahhhhhhh this new platform i'm not being forced to move to doesn't support my extreme edge case
topkek, intelfags mad they overpaid for 4 years and now need to validate their purchase

File: lure.jpg (304KB, 2824x1277px) Image search: [Google]

304KB, 2824x1277px

>>59034767

>>59034767
>wat r werkstations

Neck yourself

>>59024831
I'm just going to tell you to kill yourself before filtering you forever. Kill yourself.

>>59034767
>Y-yes goy, you don't need more features

>>59034827
>mad
Smug's more like it. AM4 is looking to be the very thing that ruins Ryzen's potential. AMD had to force compatibility with their shitty APUs left over from the Excavator days and that resulted in this tepid pile of shit before us.
LGA1151 is literally more advanced than AM4, and it's already two years old.

>>59034856
>LGA1151
1151 is also a shitty APU platform.

>>59034821
That's not true anymore. i7-6950X is the full Broadwell-E die, there are no more deactivated CPU cores. Now hang yourself.

>>59034767
>complains about features he doesn't need on a CPU at least half the price or more if compared to xeons
>muh gaymez

If you don't need it don't put fucking ecc ram in it you cocksniffing faggot

>>59034872
But it still has DMI3.0, which has far better expansion support and higher transfer speeds than that Infinity Fabric nonesense AMD shat out because their HyperTransport was too old to be recycled again.
I bet IF is still HT, but with a slightly wider bus

>>59034856
>it's already two years old

Time for an upgrade amirite?

File: 1485987312544.jpg (538KB, 1520x2167px) Image search: [Google]

538KB, 1520x2167px

>>59034856
How, exactly,?

>>59034882
ECC memory isn't even that expensive when compared to non-ECC. ECC memory can sometimes be cheaper in certain stores.

>>59034915
I know.

>>59034915
It's more expensive in general but if you shop around and compare different brands and makes then you can sometimes get cheaper ECC than non-ECC RAM.

>>59034905
>multiple NVMe drive support without needing an external controller
>native USB3.1 controller inside the chipset
>more robust memory controller and VRM
>DMI3.0 >>> a rebranded HyperTransport
>PCIe Gen 3.1 support from the chipset rather than the 2008-era PCIe Gen 2.0 from AM4's SB
>supports at least 6 SATA ports and 1 SATA Express port rather than 4 SATA ports and 1 SATA Express port (that's shared with the M.2 slot)
>lower latencies between chipset-to-CPU transfers
Need I go on

>>59034965
And don't forget twice as expensive

>>59034965
You seem to be wrong on a few items, and massively exaggerated on others

>>59035001
That's his fucking job.

>>59035001
proofs?

>>59034893
>>59034965
>But it still has DMI3.0, which has far better expansion support and higher transfer speeds than that Infinity Fabric nonesense AMD shat out because their HyperTransport was too old to be recycled again.
>I bet IF is still HT, but with a slightly wider bus
>DMI3.0 >>> a rebranded HyperTransport
Make a claim, can't back that claim up. Classic Intelfag.

DMI 3.0 is absurdly slow, only clocking in at 4GB/s if you saturate all 4 data paths. AMD has already stated Infinity fabric scales from 30GB/s to 512GB/s. So that alone makes it 9x faster at the low end, and over 100x faster at the high end.

Please just neck yourself. The human race doesn't need a retarded little baby like you.

>>59035032
Actually, what makes this even more hilarious is that HT2.0 (2004) is already faster than DMI 3.0 (2015)

>>59035016
The 3.1 thing is total bullshit
VRMs??? Lololol
((((Memory controller)))) whateverthefuck
Sata express? Nothing uses it

And its him making fantabulous claims about lower latency, and pretty much everything he posted. Total bullshit.

>>59024862
It's slower

>>59034965
>>multiple NVMe drive support without needing an external controller
I'll be interested in seeing how you'll put two m2 SSDs into one M2 slot.
>>native USB3.1 controller inside the chipset
AM4 supports USB3.1 gen1 and gen2, without any off chip controller
>>more robust memory controller and VRM
You don't know how anything about the IMC besides the number of lanes, and Intel doesn't put VRM there, but motherboard jews, also more VRMs just means you have a more power hungrier chip if you forgot.
>>DMI3.0 >>> a rebranded HyperTransport
DMI3.0 has pathetic bandwidth, 6Gbit/s to the SB, two NVMe drives will choke and one NVMe drive with a fast USB drive will choke, also you know nothing about IF and apparently you know nothing about HT either.
>>PCIe Gen 3.1 support from the chipset rather than the 2008-era PCIe Gen 2.0 from AM4's SB
Supported by the CPU directly.
>>supports at least 6 SATA ports and 1 SATA Express port rather than 4 SATA ports and 1 SATA Express port (that's shared with the M.2 slot)
This is the only thing you didn't lie about, congrats.
>>lower latencies between chipset-to-CPU transfers
Intel's desktop southbridge are on board instead of integrated like AM4's, there is theoretically no way your 3-external bus hopping and back chipset has lower latencies.

>Need I go on
Please do

>>59035032
>DMI 3.0 is absurdly slow, only clocking in at 4GB/s if you saturate all 4 data paths
Nope, not even the non-Intel white papers show DMI dropping to half its bandwidth in any circumstance other than saturating all of the DMI's bus to a 2-to-1 ratio (and both ways, since DMI is full duplex).
> AMD has already stated Infinity fabric scales from 30GB/s to 512GB/s
Yes, and AMD also said that Bulldozer was totally going to beat an i7-980X in multi-threaded benchmarks

Kill yourself, Pajeet.

>>59035073
>The 3.1 thing is total bullshit
AMD does not have USB3.1 controllers inside their SB chipset and have to rely on an external chip from ASMedia. Try again.
>VRMs?
Yes, did you not know that decent Intel boards have had dedicated power and voltage regulators specifically for their DIMMs?
AM3+ and FM2+ had no such thing, and neither will AM4
>((((Memory controller))))
>>>/v/
Stupid AMDkiddie doesn't know what a memory controller is. I bet he doesn't even remember AM3+ CPUs shitting the bed if you even pushed the memory clock beyond its 933MHz stock clock.
>Sata express? Nothing uses it
But at least it doesn't rob you of bandwidth on any other ports on current LGA1151 boards.

>>59035130
My z170 board sureasfuck has an 3.1 chip on it. You didn't explain the memory controller, so I assume you have no idea either
So you really have nothing substantial ?

File: bentium.png (59KB, 256x256px) Image search: [Google]

59KB, 256x256px

>>59035130
>muh bulldozer
Should I unearth Intel's past failures and make baseless conjectures about the future, too?

>>59033908
nice rig anon
what did you get the tesla for?

File: single socket holocaust.jpg (28KB, 550x550px) Image search: [Google]

28KB, 550x550px

>>59035201

>>59035222
AKA Emergency Edition

>>59035097
>I'll be interested in seeing how you'll put two m2 SSDs into one M2 slot.
I'll be interested in seeing how many AM4 motherboards there will be that has more than one M.2 slot that supports the full PCIe3.0 4x specification and won't sap 8 lanes from the CPU itself
>AM4 supports USB3.1 gen1 and gen2, without any off chip controller
https://www.techpowerup.com/223583/amd-zen-processor-integrated-chipset-has-usb-3-1-issues-could-escalate-costs
Their solution was literally to outsource the problem and buy an external USB3.1 controller
>You don't know how anything about the IMC besides the number of lanes, and Intel doesn't put VRM there
First off, AMD has ALWAYS had shitty memory latency issues since the K6 days. Second, Intel motherboards support additional power and voltage regulators SPECIFICALLY FOR the memory controller for better stability at higher clock speeds. AM3+ had no such thing and suffered for it. Try using a pair of 2133MHz RAM and overclock the CPU at the same time. You'll bluescreen faster than you can say "POO"
>DMI3.0 has pathetic bandwidth, 6Gbit/s to the SB
8GT/s, pleb. It's equivalent to roughly 8 lanes of PCIe3.0
>Supported by the CPU directly
But not the SB. 2008 called, they want their 32nm chipsets back.
>Intel's desktop southbridge are on board instead of integrated like AM4's, there is theoretically no way your 3-external bus hopping and back chipset has lower latencies
Nigga what the fuck are you smoking?

>>59035226
And it still lost to Athlon64. To usual, consumer ones, FX series literally raped anything that Intel had to offer.

>>59035222
Oy vey
Looks like it's time for another socket change, goy

>>59035130
>>DMI 3.0 is absurdly slow, only clocking in at 4GB/s if you saturate all 4 data paths
>Nope, not even the non-Intel white papers show DMI dropping to half its bandwidth in any circumstance other than saturating all of the DMI's bus to a 2-to-1 ratio (and both ways, since DMI is full duplex).
Even if it's full duplex it's still slower than HT2.0 which runs at 22.4GB/s as compared to the theoretical 4GB/s both ways of DMI.
Even better, HT is also full duplex and could manage 44.8GB/s since apparently you think that it being full duplex means you add them up.

>> AMD has already stated Infinity fabric scales from 30GB/s to 512GB/s
>Yes, and AMD also said that Bulldozer was totally going to beat an i7-980X in multi-threaded benchmarks
It did though.

>Kill yourself, Pajeet.
After you.

>>59035267
is this linus

>>59033255
>Meanwhile Google released data showing about 5 single bit errors in 8 Gigabytes

That's some bull shit. For one the paper I found by google on the issue seems to show the data they have is all over the place and that bit flips appear to be occurring mostly in the same modules. This leads me to think they are shitty RAM and getting away with it because redundancy.

For another I decided to a similar claim I saw on /g/ this using ramfs. I used dd to grab 2GB of garbage from /dev/urandom and put it in a ramfs. I saved the sha512sum on my hard drive. I set up a script to dd the random file to another file in ramfs, wait 1 minute, and dd from that file back to the original, wait 1 minute, and repeat . I let it run for 3 days. I copied the file out of ramfs and on to my hard drive and compared the sha512sum with the original sha512sum. Exactly the same.

So with a 2GB target sitting there in memory for 72 hours I would have expected 90 bit flips by your rate. Then add to the potential for lossy copying that happened 4320 times. I was disappointed. I wanted to use my RAM as a cosmic ray detector. Turns out shit doesn't happen anywhere near as often as memers would like to suggest. That is unless cosmic rays know how to calculate hash collisions and only flip bits that go unnoticed by sha512.

If ECC RAM was the same price for the same performance yeah I'd buy it. Until I'm running programs with huge memory foot prints that I can't just rerun if something goes wrong because it's a simulation that takes a month to run or something I won't sacrifice for it.

File: Intel-Z170-chipset-block-diagram.jpg (212KB, 600x450px) Image search: [Google]

212KB, 600x450px

>>59035232
>I'll be interested in seeing how many AM4 motherboards there will be that has more than one M.2 slot that supports the full PCIe3.0 4x specification and won't sap 8 lanes from the CPU itself
Very likely most of those with M.2 slots, since AMD has no problems with PCIe 3.0 support with their previous architectures.

>https://www.techpowerup.com/223583/amd-zen-processor-integrated-chipset-has-usb-3-1-issues-could-escalate-costs
>however, is rumored
Ok.

>AMD has ALWAYS
Wow, remember how Intel ALWAYS has I/O bugs every gen needing new steppings and recalls, a borked lookup table and 30 stage housefires?
Like most of your claims, there's a lot of assumptions.
>8GT/s, pleb. It's equivalent to roughly 8 lanes of PCIe3.0
Looks very much 6GB/s to the southbridge from Intel's own slides.
>But not the SB. 2008 called, they want their 32nm chipsets back.
80% of the lanes are already in the CPU, some cheap motherboards won't even have a chipset for extra lanes.
>Nigga what the fuck are you smoking?
Nothing, you're apparently unaware that the entire southbridge is in the CPU instead of on the motherboard, for power saving and latency.
Ergo there's no way a Intel chipset that's still on a motherboard can have lower latency.

>>59034316
>kill yourself myself

What did he mean by this?

>>59035368
he meant that he's going to kill that person himself. please keep up with the class in reading.

File: 1479834857320.png (973KB, 801x1500px) Image search: [Google]

973KB, 801x1500px

>>59035222
make longer bipeline :DDDDDDDDDDDDDD

File: windows-intel-itanium-logo.jpg (26KB, 295x221px) Image search: [Google]

26KB, 295x221px

>>59035201
lest we forget Itanic...

>>59034447
Where did I say APUs were high performance?

File: o0700100012696567516.png (127KB, 700x1000px) Image search: [Google]

127KB, 700x1000px

>>59035201
Not even the worst arch Intel has done.
Remember Tejas and Jayhawk?
Was so bad it didn't even make it to market.
https://en.wikipedia.org/wiki/Tejas_and_Jayhawk
Itanium was also DOA.

File: 46549874.jpg (998KB, 1814x1022px) Image search: [Google]

998KB, 1814x1022px

>>59024831
begone

File: eJwNyMsNwyAMANBdGACDMR9nG0QQiZQEBO6pyu7tO76v-sxLbeoQGWsD2M9V-tz1kj5zq7r13q6ax7l06TdkkVyOuz6yAA2nFKMPzjM5QrKA5P_hrCfGyIkCgaWY0JjgXCLW42nq_QHBTCKO.UGGJuoTecQx7cRqNg5JehEooZIo.png (378KB, 552x543px) Image search: [Google]

378KB, 552x543px

>>59036686
51 STAGE PIPELINING

File: ae983dd69a1157ab7f7342a6037b38f10d652402_full.jpg (33KB, 600x600px) Image search: [Google]

33KB, 600x600px

>>59036714
They had 100 stage bipeline chips taped out.

>>59036776
Th-
WHAT
THE
FUCK.

>>59036776
>wow guys what do we do with all these leftover ALUs?
>put them in a new chip!

>>59035353
>almost anything in and around a computer has error correction, but ram doesnt need it because it costs a bit more

and so we keep buying ram with a design flaw that was a compromise when ram used to be really expensive.

>>59035353
>If ECC RAM was the same price for the same performance yeah I'd buy it.

But it is. The performance impact is pretty much zero and the price is almost the same. (Actually if you're buying second hand ECC RAM is cheaper.)

>>59024805
that's cool.
Is there any difference with the DIMMs that would make Motherboards more expensive?

>>59038608
...Do you not know how memory controllers work?