Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
1)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76651)
Posted 3419 days ago by  Sarge104
I find that if in doubt, as long as your memory isn't actually faulty, then you can usually fall back on the RAM's XMP settings, along with a CPU multiplier you know to be stable.
Took the time to run the Windows memory test on extensive mode(Mats+,Stride38,wschckr,wstride-6,chckr4,wchcker3,erand,stride6, and chcker8) for seven passes after bringing the tower back up. Went through the log and found that seven passes didn't check all the memory(pages tested 4167985/untested 1122) so will let it run longer tonight...More time the tower is not crunching :(.
As for memory settings with the cpu currently on a non-overclocked setting the memory voltage and its frequency has remained steady at 802.7, north bridge has also not dropped below freq 1605.
Current timings are 8,8,8,24(standard timings) and do not change when cpu is overclocked, voltage is kept at 1.5V since I really do not want to run the ram at 1.35V when I am relying on fast access.
In the end I was able to find out more about what the current hardware I'm running can do and its limitations, I do appreciate the assistance and hope everyone has a great weekend!
|
2)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76630)
Posted 3419 days ago by Sarge104
If it's not overclocking, my next guess would be faulty memory. I have no reasonable theory about why it may have worked in the past but isn't working now. Something changed, but it's not obvious what.
Took the tower down for maintenance last night. Will clean the screens and check on some Hardware configurations I'm working with to sort out some wire ugliness before I bring it back up with the overclocking settings and I really need to allocate some time to memory testing.
I've tried to go back and see the last set up I was running the LLR's with due to the fact I just upgraded hardware last June. The WU's history doesn't go back that far though. It may be I had the slower memory and cpu at that time when I ran those so that would bring me back to doing some memory testing.
Since last June the computer has run the gambit of projects and I've not seen any issues. I've ran some rather large WU's from other projects but that doesn't mean there is no issue with the random ram allocation that can go on.
|
3)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76626)
Posted 3419 days ago by Sarge104
Two questions: Is your CPU overclocked? Do you get validation errors on our other LLR projects, or is this the first time you've run that computer on PrimeGrid?
The second question is about your comment regarding "physical cores".
IRT Overclocking:
The first eight WU's were subjected to overclocking(4.8ghz) when ran, the rest were ran at non-turbo clocking(4.0ghz) to see if stability would be an issue. Temperatures stayed well within operational limits overclocked(40-50C, chip can be ran to 60-65C overclocked).
At non-turbo speeds the temperatures stayed 15-22C, I really do not think heat was an issue with the invalidation. This was my first testing for the Cullen LLR, I've ran (Riesel, Sierpinski / Riesel, PSP, Woodall, TPS, and 321 Prime Search) LLRs with no issue before on overclocked settings.
IRT Physical cores:
Guru3d FX-8350 review
Guru3d does a better explanation.
While the chip has eight physical cores there is four Piledriver modules driving them all. I was incorrect in previous post with the Core statement. The link above will elaborate. I believe with one core per module plus less cores communicating with the ram led to the decrease in times.
|
4)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76609)
Posted 3419 days ago by Sarge104
WU's worth 743 hours still pending, 464 hours worth have been invalidated. Nothing has been validated atm. No error-ed out WU's but purged the rest due to the long run times and another upcoming challenge .
The FX-8350 really struggled fully loaded at 111 hours per WU, with just the physical cores loaded (4) the times went down to 60 hours at the end. Ran out of time to finish any more WU's so I'll never know the best run times for the chip unfortunately.
Chip AMD FX-8350
Memory 2 x 8gig stick ddr3 1600 Crucial low profile (placed to support dual channel set up)
Motherboard M5A99X EVO R2.0
Was an interesting test to see what the capabilities were for the chip on these larger WU's, good luck to participants for the rest of the challenge!
|
5)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76562)
Posted 3421 days ago by Sarge104
Ouchie, seems quite a few of the tasks my computer processed seem to be on their 6-8th round though. Will be interesting to see if any validate before the challenge ends. :(
|
6)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76537)
Posted 3422 days ago by Sarge104
The term "AVX2" is frequently used incorrectly. AVX2 really applies to an improved set of integer instructions, which are not of much use to us.
FMA3 is one particular instruction (an enhancement to certain AVX instructions) that is not really part of AVX2, but was released at the same time with the same processors.
Therefore, when we talk about FMA or FMA3, we're talking about the same CPUs as AVX2, but it's the FMA instruction, not the AVX2 instructions, that we're using.
AMD's Piledriver supported both FMA3 and FMA4, so I guess I'm still wondering why/how the implementation is flawed I guess. If the support is there why isn't the instruction set used to allow a more efficient use of the processor. I must be missing something...
|
7)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76533)
Posted 3422 days ago by Sarge104
AMD's implementation of the AVX instruction set is so awful that the LLR program doesn't use AVX on AMD processors. There's simply no advantage to using it. (It's not just LLR. Most, if not all, similar programs, such as PFGW and Genefer only use AVX on Intel chips.)
On Intel processors, AVX (and the subsequent FMA3 with the Haswell CPUs) provides a massive speed improvement. On AMD -- nothing. As a result, Intel CPUs dominate the primality testing domain because they are much, much faster.
Just so we are on the same page we are talking AVX2 instruction sets, not general AVX correct? I noticed the reference to FMA3 so that is why I'm asking. Thanks!
|
8)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76531)
Posted 3422 days ago by Sarge104
On Intel processors, AVX (and the subsequent FMA3 with the Haswell CPUs) provides a massive speed improvement. On AMD -- nothing. As a result, Intel CPUs dominate the primality testing domain because they are much, much faster.
If you want fast, you don't need expensive -- but you do need Intel.
Appreciate the fast response, I assumed that the instruction set might be the issue, thanks for confirming it.
As for your statement regarding the price of chips I must disagree, with AMD its not the chip price but the power cost that will get you. Intel is just expensive regardless but has a lead on power cost.
|
9)
Message boards :
Number crunching :
Race to the New Moon Challenge
(Message 76523)
Posted 3422 days ago by Sarge104
Cullen LLR in about a day? I'm running an AMD FX8350 and I'm getting 112hr WU's. I did run the chip overclocked to 4.8ghz but currently do not have any overclocking to see if it makes any difference. Still seeing over 90hrs with the WU's being processed currently not completed.
Just wondering there is an instruction set that isn't being used because I'm not sure I can afford the 115.8Uberghz processor to get these WU's done in under a day ;) or do you guys just have something against AMD :D
|
10)
Message boards :
Number crunching :
The Leonids Challenge
(Message 28023)
Posted 4702 days ago by Sarge104
GAME ON! Good luck and gimme those taskes!
YaY! just got my first dump as I was typing!!!
|
Next 10 posts
|
