PrimeGrid Challenges 2020

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
Theoretically,
  • i9-9000K with its 16 MB unified last-level cache scales well to 4 simultaneous cpuGFN18 jobs,
  • Coffee Lake i5 with 9 MB unified LLC (in desktop variants) scales well to 2 simultaneous cpuGFN18 jobs,
and with more concurrent jobs than these, total throughput either increases marginally, or perhaps decreases, I don't know which.
 

lane42

Diamond Member
Sep 3, 2000
5,721
624
126
The 2 cpu's i just got 3800x and 9900k are the first one's ive had
with H/T. Didnt need it for Seti. A core i3 ran 4 gpu's in one computer,
and a core i5 ran 5 gpu's, well for awhile, in another.
So i fired up GFN18 on the 9900k and wow 16 workunit's, or is that what you call jobs.
are running but sloooow. So i turned off H/T and now 8 workunit's are
running, less 1 for the gpu. Progress bar in B/M really moving now :)
So 7 workunits running takes about 8 hours. if i cut the number cores
to 3 or 4 things speed up more. Why the speedup or slow down, these
are real cores.I would think 8 hours for 7 cores it would still run 8 hours
on say just 4 cores. I dont see this kind of thing on the 3800x. Now
iam just going by the progress bar % in B/M, i see very little change if
i run 6 or 7 cores or if i run 3 or 4 cores on the 3800x. I hope you
understand what iam trying to say. it seems the core's take much more
of a hit on the real cores then the ryzen's do.
Now let me go soak my sore fingers.
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
Again: The size of the data which cpuGFN18 is working on is 4 MB, per each program instance.

Ryzen 3800X can keep 8x 4 MB in its processor caches.
i9-9900K can keep 4x 4 MB in its processor caches.

If you make them work on more data than that simultaneously, then
  • only some of the data can be cached,
  • the rest of the data needs to be read from RAM/ written to RAM all the time,
  • yet this I/O from/to RAM has very high latency and very low bandwidth compared with cache accesses,
  • therefore, the vector arithmetic units of the processors go idle for much of the time.

The Ryzen 3800X is computing something most of the time.
The i9 is waiting for data most of the time.

You can make the i9 compute most of the time by running merely 4 simultaneous cpuGFN18 tasks. This will leave 1/2 of the vector arithmetic units idle all the time though. Alternatively, you can run more than 4 tasks at once, and thereby use respectively more of the vector units. But then each of the vector units in use will repeatedly fall idle due to having to wait for data which did not fit into the caches anymore.

In other words,
  • both processors have far more hardware threads than you can use for cpuGFN18,
  • the i9 has twice as many physical cores than you can use for cpuGFN18,
  • because the i9 has only half the cache size per core compared with Ryzen 3000.
 
  • Like
Reactions: lane42

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,220
3,803
75
Day 2 stats:

Rank___Credits____Username
83_____1314753____Lane42
124____756765_____emoga
154____559271_____biodoc
165____506137_____Ken_g6
235____270739_____SlangNRox
288____154773_____zzuupp
333____86206______geecee
388____43096______waffleironhead
542____1725_______Kiska

Rank__Credits____Team
19____4629295____BOINCstats
20____4570174____Ultimate Chaos
21____3828372____Space Quest
22____3693468____TeAm AnandTech
23____3431079____Ukraine
24____3353451____BOINC@Poland
25____3105887____Iowa State University

The Quest for Space Quest continues...
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,220
3,803
75
Day 3 stats:

Rank___Credits____Username
82_____2019101____Lane42
130____1118997____emoga
162____834727_____Kiska
170____760216_____Ken_g6
185____690975_____biodoc
245____417418_____SlangNRox
293____253553_____zzuupp
333____148318_____geecee
379____89752______waffleironhead

Rank__Credits____Team
17____8097421____Dutch Power Cows
18____7727239____The Knights Who Say Ni!
19____7496980____BOINCstats
20____6333060____TeAm AnandTech
21____5942388____Ukraine
22____5870274____Space Quest
23____5630198____BOINC@Poland

I'm slower than an Aussie in the summertime?! I should put my 750Ti in my main computer here.
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,220
3,803
75
Merry Christmas, everybody!

Day 4 stats:

Rank___Credits____Username
85_____2612018____Lane42
125____1613417____emoga
173____1047398____Ken_g6
193____900583_____Kiska
221____690975_____biodoc
244____574513_____SlangNRox
297____365701_____zzuupp
343____200097_____geecee
420____93199______waffleironhead

Rank__Credits____Team
18____10417806___The Knights Who Say Ni!
19____10263497___BOINCstats
20____8450384____Ukraine
21____8097904____TeAm AnandTech
22____7906338____BOINC@Poland
23____7749128____Space Quest
24____7319045____Iowa State University

HAHAHA :D

Fear my power! :p
What happened? :(
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
BTW, there is a brief discussion over at SETI.Germany about the GPU-GFN application's built-in benchmark which probes for the optimum OpenCL transform at the start of each task. In case of GFN-16 (not part of this challenge), this benchmark is currently superfluous because there is only one transform left which works for the current parameter space. And in case of the other GFN subprojects, the benchmark sometimes chooses a suboptimal transform; e.g. if there is a concurrent transient load on the GPU while the benchmark is performed. pschoefer posted an app_config which makes the application skip the benchmark and go for the user-defined transform.
 
  • Like
Reactions: emoga

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,220
3,803
75
Day 5 stats:

Rank___Credits____Username
85_____3209428____Lane42
142____1712342____emoga
167____1321102____Kiska
169____1291349____Ken_g6
249____733336_____SlangNRox
251____725701_____biodoc
297____464636_____zzuupp
341____267422_____geecee
421____132959_____waffleironhead

Rank__Credits____Team
19____12468883___BOINC@AUSTRALIA
20____10950559___Ukraine
21____10291650___BOINC@Poland
22____9858279____TeAm AnandTech
23____9662451____Space Quest
24____9279081____Iowa State University
25____7036235____Ultimate Chaos

We're falling! :eek: :(
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,220
3,803
75
Day 6 stats:

Rank___Credits____Username
84_____3847227____Lane42
136____2174675____emoga
165____1598040____Ken_g6
172____1544115____Kiska
231____997086_____SlangNRox
277____725701_____biodoc
302____569201_____zzuupp
351____329585_____geecee
397____209212_____waffleironhead

Rank__Credits____Team
19____14059375___BOINC@AUSTRALIA
20____13387012___Ukraine
21____12328178___BOINC@Poland
22____11994845___TeAm AnandTech
23____11511488___Space Quest
24____11202578___Iowa State University
25____8205135____Ultimate Chaos

I finally installed my 750 Ti. Take that, @Kiska! Or don't! ;)
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,220
3,803
75
Day 7 stats:

Rank___Credits____Username
83_____4450845____Lane42
140____2405974____emoga
164____1859826____Ken_g6
198____1544115____Kiska
228____1247214____SlangNRox
291____725701_____biodoc
300____676026_____zzuupp
347____395241_____geecee
414____209212_____waffleironhead

Rank__Credits____Team
19____15612495___Ukraine
20____15253844___BOINC@AUSTRALIA
21____14309944___BOINC@Poland
22____13514159___TeAm AnandTech
23____13229970___Space Quest
24____12922100___Iowa State University
25____9354379____L'Alliance Francophone

Looks like now our quest is just staying ahead of the questers behind us.
 
Last edited:

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,220
3,803
75
Day 7.5 stats:

Rank___Credits____Username
85_____4791823____Lane42
147____2505171____emoga
165____2043704____Ken_g6
210____1544115____Kiska
221____1407127____SlangNRox
303____725701_____biodoc
307____709085_____zzuupp
346____435000_____geecee
426____214392_____waffleironhead

Rank__Credits____Team
19____16885460___Ukraine
20____15965167___BOINC@AUSTRALIA
21____15687277___BOINC@Poland
22____14376123___TeAm AnandTech
23____14254991___Space Quest
24____13963338___Iowa State University
25____10068442___L'Alliance Francophone

At this rate, we're going to lose to Space Quest (a team of one) and Iowa State University (a team of one). Can we get a few more GPUs in here?

Edit: I added a few more cloud CPU cores.
 
Last edited:

lane42

Diamond Member
Sep 3, 2000
5,721
624
126
Thanks for the stats g6 :)
All i have left is a GTX260 which, well, won't cut it.
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
During the last several challenges at which I showed up at all, I tended to run only in the second half or last third. So from that perspective, yes, 3 days are somewhat short indeed. :-)

Maybe the organizers have insider info about market supply of new computing gear recovering in time for the 2021 challenge series. ;-)