» Search forum


Index :: Challenge Chats :: {H]ard|OCP's VGTU test for potential inclusion into the DC-Vault


Pages: 1 [2] 3
Coleslaw
BAM!ID: 60911
Joined: 2008-10-31
Posts: 189
Credits: 4,609,530,251
World-rank: 744

2016-04-04 01:30:53

Welcome SETI.USA. Showing up late and kicking it pretty hard.
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-04 15:00:50

Dirk, give me an example where a workload runs better on AMD vs Intel when using Linux. Some projects prefer Linux and some Windows.
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-04 16:40:50

And some do better on Wintel altogether...
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-04 16:57:35

Some projects are even Windows-only
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-04 16:59:24

And most are X86 (and/or X86-64) only
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-04 17:00:23

I'm not here with a Bias on which is better. I try to share information that allows people to get the most out of what they have. I am in a unique position to resolve issues with Intel CPU's. On Intel cpu's HT will increase the time per work unit but the total work done with the extra threads more than makes up for the longer run time. I still have more testing to do but it looks like the added AES instructions give this client a boost in performance.
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-04 17:37:35

Future AMD Zen will also have HT
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-04 17:47:53

Cool! Competition is a good thing. Speaking of which nice to see Mumps joining in.
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-04 21:21:15

Hmm, looks like AES doesn't add anything. Linux is ruling over Windows.
X5690(3.47Ghz) windows HT on: 2.18 hours/wu
X5690(3.47Ghz) Linux HT on: 1.14 hours/wu
X5690(3.47Ghz) Linux HT off: .663 hours/wu
X5570(2.93Ghz) Linux HT on: 1.375 hours/wu
I7 2700(3.5Ghz) Linux HT on: 1 hour/wu
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-04 23:02:22

Hmm something in the second generation I7 vs first generation i7 is making a significant improvement.
X5690(3.47GHz) Linux HT on: 1.14 hours/wu
E5-2860 (2.7GHz) Linux HT on: 1.13 hours/wu.
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-04 23:24:51

'Something' might be F16C/16-bit Floating-Point conversion, which is absent in the 1st generation i7, but present in the 2nd.
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-04 23:25:48

Athlon 5350 does have that, too...
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-04 23:42:01

29% improvement. Many nice things were missing in the 1st generation that showed up in the second. Third on made small improvements, mostly in power consumption.
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-04 23:49:58

F16C/16-bit Floating-Point conversion is a 3rd generation thingy, the 1st generation i7 did not even have AVX
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-05 00:04:32

1st gen was mainly a complete redesign from the old core arch. Big improvements in speed along with large power demands. 2nd added AVX, AES instructions, and lots of other speed tweaks and reasonable power usage. 3rd lowered power usage and lost some OC headroom. 4th added AVX2 and more power reductions.
Mumps
BAM!ID: 83603
Joined: 2010-04-01
Posts: 1
Credits: 799,102,760,339
World-rank: 6

2016-04-05 00:11:32

Heya. Yeah, I'm here MODist, watching you being selfish with all the WU's. Seems an admin forgot to enter Seti.USA on time. Great run ya got going there MODist!
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-05 00:17:05

Hey Mumps. My goal was to get our team to second place overall before the end of the challenge. Had you not stalled at the start it would have been a close race to the finish.
Coleslaw
BAM!ID: 60911
Joined: 2008-10-31
Posts: 189
Credits: 4,609,530,251
World-rank: 744

2016-04-05 00:59:42

Glad you guys showed up anyways. The more stress we can put on the servers the better.
Mumps
BAM!ID: 83603
Joined: 2010-04-01
Posts: 1
Credits: 799,102,760,339
World-rank: 6

2016-04-05 03:20:59

So sad that there aren't enough WU's to go around.
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-05 06:56:11

4th generation also had FMA3, 5th generation brought TSX and faster memory support
Dirk Broer
 
BAM!ID: 117974
Joined: 2012-03-01
Posts: 343
Credits: 6,575,465,560
World-rank: 574

2016-04-05 06:57:07

oops, faster memory came with 6th generation
Coleslaw
BAM!ID: 60911
Joined: 2008-10-31
Posts: 189
Credits: 4,609,530,251
World-rank: 744

2016-04-05 13:14:17

Mumps, they should be making work units on the fly. Are you seeing them run completely out for more than a few minutes? They typically have 50 available to send at all times but someone with the amount of hardware you have may drain them quicker than they can create new. Please take a look and report in.
Coleslaw
BAM!ID: 60911
Joined: 2008-10-31
Posts: 189
Credits: 4,609,530,251
World-rank: 744

2016-04-05 14:04:33

Since my last post, saw the server status go from 6 available to 201 available. So work is being generated. I just wonder if they need to adjust the server to create larger batches to keep things smooth.
MODist
BAM!ID: 203618
Joined: 2016-03-24
Posts: 0
Credits: 0
World-rank: 0

2016-04-05 14:33:35

We are averaging about 22 wu/min right now.
Coleslaw
BAM!ID: 60911
Joined: 2008-10-31
Posts: 189
Credits: 4,609,530,251
World-rank: 744

2016-04-05 15:58:04

The admin has confirmed raising the amount of available work units due to the issue brought up by Mumps. Please post if you still have any issues.
Pages: 1 [2] 3

Index :: Challenge Chats :: {H]ard|OCP's VGTU test for potential inclusion into the DC-Vault
Reason: