Author |
Message |
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
This is error I got
Iter: 51/245559, ERROR: ROUND OFF (2.812417676e+013) > 0.4
Continuing from last save file.
Disregard last error. Result is reproducible and thus NOT A HARDWARE PROBLEM.
For added safety, redoing iteration using a slower, more reliable method.
Continuing from last save file.
Iter: 51/245559, ERROR: ROUND OFF (6.796424524e+013) > 0.4
Continuing from last save file.
Unrecoverable error, Restarting with next larger FFT length...
Fatal error at setup : Number sent to gwsetup is too large for the FFTs to handle.
I try few combinations
fresh copy windows 7 without SP1 (x64) - llravx works without problem ( time is same as "normal LLR")
fresh copy windows 7 WITH SP1 (x64) - got error listed above.
Also I have to say that this processor is identified as Core2 ( by avx app)
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
Title will be AMD FX X8 8120 NOT working with AVX
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
|
My FX 4100 doesn't even get that far. It fails instantly with that last error. I had the same issue happen while testing the AVX version of Prime95. |
|
|
|
I can confirm that the FX-8150 doesn't work the apps either. I killed 8 Cullens in about a minute |
|
|
rroonnaalldd Volunteer developer Volunteer tester
 Send message
Joined: 3 Jul 09 Posts: 1213 ID: 42893 Credit: 34,634,263 RAC: 0
                 
|
Iter: 51/245559, ERROR: ROUND OFF (2.812417676e+013) > 0.4
Continuing from last save file.
Disregard last error. Result is reproducible and thus NOT A HARDWARE PROBLEM.
For added safety, redoing iteration using a slower, more reliable method.
Continuing from last save file.
Iter: 51/245559, ERROR: ROUND OFF (6.796424524e+013) > 0.4
Continuing from last save file.
Unrecoverable error, Restarting with next larger FFT length...
Fatal error at setup : Number sent to gwsetup is too large for the FFTs to handle.
Thanks for pointing to. I get the same message with all apps that were compiled against the new gwnum27.2.
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
Maybe "new" gwnum27.2. have some error inside code? Can someone recompile app but using older gwnum?
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
I can now confirm that llravx NOT working even under Linux ( Ubuntu 11.10 64 bit)
Error I got
NORMAL LLR
7745*2^245547+1 is not prime. Proth RES64: 69C444D105D03865 Time : 100.067 sec.
LLRAVX
Iter: 51/245559, ERROR: ROUND OFF (2.812417676e+13) > 0.4
Continuing from last save file.
Disregard last error. Result is reproducible and thus not a hardware problem.
For added safety, redoing iteration using a slower, more reliable method.
Continuing from last save file.
Iter: 51/245559, ERROR: ROUND OFF (6.796424524e+13) > 0.4
Continuing from last save file.
Unrecoverable error, Restarting with next larger FFT length...
Fatal error at setup : Number sent to gwsetup is too large for the FFTs to handle.
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
rroonnaalldd Volunteer developer Volunteer tester
 Send message
Joined: 3 Jul 09 Posts: 1213 ID: 42893 Credit: 34,634,263 RAC: 0
                 
|
Maybe "new" gwnum27.2. have some error inside code? Can someone recompile app but using older gwnum?
I will do this in the next minutes...
[edit]
Here they are. Both static linked.
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
Can you make build with AVX for AMD processor ( with older library)?
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
rroonnaalldd Volunteer developer Volunteer tester
 Send message
Joined: 3 Jul 09 Posts: 1213 ID: 42893 Credit: 34,634,263 RAC: 0
                 
|
Can you make build with AVX for AMD processor ( with older library)?
This would be useless because the older lib doesn't know anything about AVX.
You will get no improvement, if you compile the app with the switches "-mAVX" or "-mtune=core-avx-i".
The assembler part of the gwnum-lib makes the cpu/feature-detection and math itself.
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
So, for now there is no option to build AVX version of LLR for AMD , right?
So we must wait :))
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
Any news about AVX version of LLR for AMD?
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
rroonnaalldd Volunteer developer Volunteer tester
 Send message
Joined: 3 Jul 09 Posts: 1213 ID: 42893 Credit: 34,634,263 RAC: 0
                 
|
There are new AVX-executables on the PG-server and i got the same rounding error on my Core2-host for: ./primegrid_sllr_3.8.6_dev20120111_i686-pc-linux-gnu__avx -d -q"30448908048555*2^666666-1"
The win-version is here and produces the same error.
Our problems will be unsolved until George finds the mistake in his gwnum27.2-lib.
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
Ken_g6 Volunteer developer
 Send message
Joined: 4 Jul 06 Posts: 929 ID: 3110 Credit: 237,002,205 RAC: 39,933
                           
|
Our problems will be unsolved until George finds the mistake in his gwnum27.2-lib.
Unless George finds a mistake in the Bulldozer processors first!
It wouldn't be the first time...
____________
|
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
I think it is compiler error.
Why AMD Buldozder was show as CORE2Type processor? Is CORE2Type Intel mark?
And also on "ordinary LLR application" also AMD processor are shown as CORE2Type ...
It looks like compiler cannot recognized new AMD processor, and that is at least part of error.
Has anyone compiler optimized for AMD buldozder processors?
Make exe file just for one type of proc is maybe extra work, but if that exe works that it will be great news!
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
rroonnaalldd Volunteer developer Volunteer tester
 Send message
Joined: 3 Jul 09 Posts: 1213 ID: 42893 Credit: 34,634,263 RAC: 0
                 
|
I think it is compiler error.
Why AMD Buldozder was show as CORE2Type processor? Is CORE2Type Intel mark?
And also on "ordinary LLR application" also AMD processor are shown as CORE2Type ...
It looks like compiler cannot recognized new AMD processor, and that is at least part of error.
Has anyone compiler optimized for AMD buldozder processors?
Make exe file just for one type of proc is maybe extra work, but if that exe works that it will be great news!
Why should it be a compiler error?
The CPU-detection and calculation is done in the assembler part of the gwnum-lib and only 27.2 has support for AVX.
Here the complete Makefile for the gwnumXX.X-lib: # Makefile for Linux and FreeBSD 32-bit gwnum library
#
CC = gcc
CFLAGS = -I.. -I../qd -O2 -march=i486 -malign-double -Wno-unused-result -Wno-format
CPP = g++
CPPFLAGS = -I.. -I../qd -O2 -march=i486 -malign-double -Wno-unused-result -Wno-format
OBJS = cpuid.o gwnum.o gwthread.o gwutil.o gwdbldbl.o giants.o ecmstag1.o // gwnum26.6 without AVX
OBJS = cpuid.o gwnum.o gwthread.o gwutil.o gwdbldbl.o giants.o ecmstag1.o gwtables.o // only needed in gwnum27.2 with AVX
LIB = gwnum.a
#########################################################################
$(LIB): $(OBJS)
cp linux/gwnum.a .
ar -rs $(LIB) $(OBJS)
clean:
rm -f $(OBJS) gwnum.a
.c.o:
$(CC) $(CFLAGS) -c $<
.cpp.o:
$(CPP) $(CPPFLAGS) -c $<
-march=cpu-type: Generate instructions for the machine type cpu-type. The choices for cpu-type are the same as for -mtune. Moreover, specifying -march=cpu-type implies -mtune=cpu-type."
I tried to compile the AVX-app without the file 'gwtables.c' and got it not compiled like all others.
Gary pointed me to this file and then i got the app compiled. In the header of the file 'gwtables.c' you find the information: "This file contains the C routines to build sin/cos and weights tables that the FFT assembly code needs."
[add]
Whether you take the llr3.8.6- or the llr3.8.6dev-source makes no difference, only the gwnum-lib decides whether you have AVX-support or not.
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
Neo Volunteer tester
 Send message
Joined: 28 Oct 10 Posts: 710 ID: 71509 Credit: 91,178,992 RAC: 0
                   
|
I think it is compiler error.
Why AMD Buldozder was show as CORE2Type processor? Is CORE2Type Intel mark?
And also on "ordinary LLR application" also AMD processor are shown as CORE2Type ...
It looks like compiler cannot recognized new AMD processor, and that is at least part of error.
Has anyone compiler optimized for AMD buldozder processors?
Make exe file just for one type of proc is maybe extra work, but if that exe works that it will be great news!
Crun-Chi,
I had the same problem and I also (somewhere here in the threads) asked the same question you just posed.
This is clearly some type of error, of which, I have not seen a solution nor explanation other than my own solution...
Return the AMD Bulldozer chip and replace it with a Phenom II X6 BE.
Neo
AtP
|
|
|
|
I think it is compiler error.
Why AMD Buldozder was show as CORE2Type processor? Is CORE2Type Intel mark?
And also on "ordinary LLR application" also AMD processor are shown as CORE2Type ...
It looks like compiler cannot recognized new AMD processor, and that is at least part of error.
Has anyone compiler optimized for AMD buldozder processors?
Make exe file just for one type of proc is maybe extra work, but if that exe works that it will be great news!
Gwnum-lib has several different code paths and picks the fastest one. My Sandy Bridge shows up as Core2 or Pentium4. This is normal, if a bit confusing. |
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3114 ID: 50683 Credit: 76,797,694 RAC: 4,051
                       
|
I think it is compiler error.
Why AMD Buldozder was show as CORE2Type processor? Is CORE2Type Intel mark?
And also on "ordinary LLR application" also AMD processor are shown as CORE2Type ...
It looks like compiler cannot recognized new AMD processor, and that is at least part of error.
Has anyone compiler optimized for AMD buldozder processors?
Make exe file just for one type of proc is maybe extra work, but if that exe works that it will be great news!
Gwnum-lib has several different code paths and picks the fastest one. My Sandy Bridge shows up as Core2 or Pentium4. This is normal, if a bit confusing.
Your Sandy Bridge is Core 2 or Pentium4: my AMD Buldozder is definitely NO Core 2 processor :))
____________
92*10^1439761-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
314187728^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
|
my AMD Buldozder is definitely NO Core 2 processor :))
It is also not like any previous AMD processors. |
|
|
|
-
____________
|
|
|
|
All silent for over 1 month? No AMD AVX love? |
|
|
rroonnaalldd Volunteer developer Volunteer tester
 Send message
Joined: 3 Jul 09 Posts: 1213 ID: 42893 Credit: 34,634,263 RAC: 0
                 
|
An AVX-version for BD can't be faster than their SSE-counterpart.
You need one integer-part of a cpu-modul for SSE (128bit) but both parts of a modul for AVX (256bit).
A BD with its different architecture can only be faster, if George adds support for FMA (Fused Multiply Add) in its gwnum-lib and this needs time...
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
|
Both PFGW and LLR have new versions out today, incorporating Buldozer friendly gwnum 2.75.
https://sourceforge.net/projects/openpfgw/
http://jpenne.free.fr/Development/
Give 'em a try! |
|
|
|
Please do try these new versions - if you find any problems please post here or PM me. Expect these coming to BOINC and PRPnet soon.
Cheers
- Iain
____________
Twitter: IainBethune
Proud member of team "Aggie The Pew". Go Aggie!
3073428256125*2^1290000-1 is Prime! |
|
|
|
I found only 32-bit Linux builds for LLR 3.8.8 / 27.5 on Jean Penne's site. I downloaded and ran them (there were two: one static-linked, one dynamic). Their speed was comparable to the other 27.x 32-bit builds, meaning, about 10% slower than the recent 64-bit builds. At least, all residues matched existing stock apps.
All of this was on a 2600K (i.e. AVX capable) and HT off.
--Gary |
|
|