768k Skylake Problem/Bug - Page 29

tha · 2016-01-07, 11:04

This is a very good result. I will continue to work on a case where we can have prime95 test the CPU for the update having been implemented or not.

I have a version of mprime working that uses no random offset when filling the FFT's. It is working on 8 exponents in the 14M range in 8 threads on 4 cores. When it finishes I will rerun the test and see how the results compare.

I tried this first with 8 exponents in the 3M range, enforcing mprime to use FFT2=768K, but a bit unexpected to me, all 8 results matched the ones in the GIMPS database.

chalsall · 2016-01-07, 19:16

Just to put on the table... I have found it helps if I post a message here which is "Currently being moderated." over on the Intel Forum for it to be released.

Quote:

Hello Cesar Badilla

Certainly we will "allow some time" for a response. But is there any chance you could provide an estimate as to how long we should expect to wait? After all, if you (read: Intel) have a fix which is now being rolled out, you should have at least a new MCU.

Please do also note that while it was Prime95 (and _very_ observant testers) which found this bug, there are many who use compute for scientific (and financial et al) purposes which might also exhibit "certain complex workload conditions".

It would be a shame if a lot of previous computing had to be redone, just to make sure the results were correct....

tha · 2016-01-08, 08:54

Quote:

Originally Posted by chalsall

Just to put on the table... I have found it helps if I post a message here which is "Currently being moderated." over on the Intel Forum for it to be released.

Legitimate questions, but it would be a bigger shame if the info would be out in the wild for everyone to exploit, and the fix would be unavailable. If the advisory matches our expectations I am perfectly OK if that comes out when the BIOS updates are available too.

ldesnogu · 2016-01-08, 09:07

I wonder if the fix will impact performance.

chalsall · 2016-01-08, 16:51

Quote:

Originally Posted by tha

If the advisory matches our expectations I am perfectly OK if that comes out when the BIOS updates are available too.

This was posted on the Intel Forum last night:

Quote:

Hello All,

Original equipment manufacturers are aware of issue. In this case please contact the system manufacturer for the latest system BIOS updates.

Regards,

Caesar B_Intel.

So it's time for Skylake owners to ask their Mother Board manufacturers for the required BIOS update.

dh1 · 2016-01-09, 00:18

interesting <http://www.ginjfo.com/actualites/composants/processeurs/processeur-skylake-intel-admet-un-bug-et-apporte-une-solution-20160108>

hoping for update soon here: <https://www-ssl.intel.com/content/www/us/en/processors/core/desktop-6th-gen-core-family-spec-update.html>

ewmayer · 2016-01-09, 01:34

Quote:

Originally Posted by Madpoo

And remember that half the people you meet have below average intelligence.

No, half the people you meet have below median intelligence. ;)

Glad Intel appears to have found the problem - I find the whole 'using ever-increasing amounts of microcode to effect post-hoc fixes for hardware bugs' subtheme very interesting, the once-clear line between field-programmable and fixed logic has gotten evermore blurry as chip complexity balloons in accordance with Mr. Moore's famous rule of thumb.

chalsall · 2016-01-09, 20:14

Quote:

Originally Posted by ewmayer

...the once-clear line between field-programmable and fixed logic has gotten evermore blurry as chip complexity balloons in accordance with Mr. Moore's famous rule of thumb.

This is to be expected.

Imagine writing a few _billion_ lines of assembly code, and then not being able to release an update after spending a few billion dollars to release the software....

Dresdenboy · 2016-01-10, 16:35

Quote:

Originally Posted by ewmayer

No, half the people you meet have below median intelligence. ;)

That depends on the typically visited locations. ;)

chalsall · 2016-01-10, 20:31

Based on the number of "Guest Viewers" it appears this is beginning to be "Out"...

Google shows a couple of popular forums have picked this up, along with two technical press sites (one French, one German).

Only a matter of time now before large news sites start covering this; it's a good thing Intel already have a fix out to their partners.

0PolarBearsHere · 2016-01-11, 00:53

Quote:

Originally Posted by chalsall

Based on the number of "Guest Viewers" it appears this is beginning to be "Out"...

Google shows a couple of popular forums have picked this up, along with two technical press sites (one French, one German).

Only a matter of time now before large news sites start covering this; it's a good thing Intel already have a fix out to their partners.

Yep. Hitting the tech blogs.
http://www.lifehacker.com.au/2016/01...lex-workloads/

2016-01-10, 20:31	#318
chalsall If I May "Chris Halsall" Sep 2002 Barbados 2×5×7×139 Posts	Beginning to hit the forums and foreign press... Based on the number of "Guest Viewers" it appears this is beginning to be "Out"... Google shows a couple of popular forums have picked this up, along with two technical press sites (one French, one German). Only a matter of time now before large news sites start covering this; it's a good thing Intel already have a fix out to their partners. Last fiddled with by chalsall on 2016-01-10 at 20:39 Reason: title: s/hit for forums/hit the forums/

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Skylake vs Kabylake	ET_	Hardware	17	2017-05-24 16:19
Skylake and RAM scaling	mackerel	Hardware	34	2016-03-03 19:14
So does skylake-nonXeon actually get us anything?	fivemack	Hardware	36	2015-09-08 01:42
Skylake processor	tha	Hardware	7	2015-03-05 23:49
Skylake AVX-512	clarke	Software	15	2015-03-04 21:48

2016-01-07, 11:04	#309
tha Dec 2002 2×11×37 Posts	This is a very good result. I will continue to work on a case where we can have prime95 test the CPU for the update having been implemented or not. I have a version of mprime working that uses no random offset when filling the FFT's. It is working on 8 exponents in the 14M range in 8 threads on 4 cores. When it finishes I will rerun the test and see how the results compare. I tried this first with 8 exponents in the 3M range, enforcing mprime to use FFT2=768K, but a bit unexpected to me, all 8 results matched the ones in the GIMPS database.

2016-01-08, 09:07	#312
ldesnogu Jan 2008 France 2·5²·11 Posts	I wonder if the fix will impact performance.

2016-01-09, 00:18	#314
dh1 "D" Sep 2015 23₁₆ Posts	interesting <http://www.ginjfo.com/actualites/composants/processeurs/processeur-skylake-intel-admet-un-bug-et-apporte-une-solution-20160108> hoping for update soon here: <https://www-ssl.intel.com/content/www/us/en/processors/core/desktop-6th-gen-core-family-spec-update.html>