Haswell Preview Benchmark - Page 3

Batalov · 2013-06-02, 17:36

Post real-life benchmarks!

fivemack · 2013-06-02, 18:23

My Haswell is arriving on Tuesday

(4770, GA-B85M-HD3 motherboard, 2x8G RAM, 120G Samsung 840 SDD)

Hoping to be able to retire the dual-Xeon-L5420 and the Q6600.

I have a box of power meters to try to calibrate how much I'd be saving by using the Haswell instead of the five-year-old kit.

sdbardwick · 2013-06-03, 01:01

Attached is i7-4770K bone stock with HT enabled.

sdbardwick · 2013-06-03, 01:15

And without HT:

sdbardwick · 2013-06-03, 01:25

No HT, but using XMP profile (1600 vs 1333)

sdbardwick · 2013-06-03, 01:34

HT with XMP1600

Batalov · 2013-06-03, 04:00

Looks quite nice. Thanks!

henryzz · 2013-06-03, 11:16

I didn't realize that TF was affected by memory bandwidth.

Prime95 · 2013-06-03, 14:10

Quote:

Originally Posted by henryzz

I didn't realize that TF was affected by memory bandwidth.

There is also a 200MHz difference in CPU speed between the slow-mem and faster-mem benchmarks.

Prime95 · 2013-06-03, 14:15

sdbardwick,

Could you run 4 LL tests of the same (or very close) exponents. Then report the timings for just 1 worker running, then 2 workers, 3 workers, and 4 workers?

I'm curious if Haswell hits the same memory bandwidth wall as Sandy/Ivy Bridge.

sdbardwick · 2013-06-03, 15:12

1600K AVX FFT

Code:

Note values are averages of two runs (4,3,2,1  and 1,2,3,4 workers)
1 Worker      8.2ms
2 Workers     8.7ms
3 Workers    10.32ms
4 Workers    12.50ms

For funsies:
1 worker x 4 threads = 3.053 ms
1 worker x 3 threads = 3.34
1 worker x 2 threads = 4.352
1 worker x 1 thread  = 8.2

This box uses PC3-1600 RAM recycled from a SB i5-2500; I don't have any faster memory ATM.

2013-06-02, 18:23	#24
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 7²·131 Posts	My Haswell is arriving on Tuesday (4770, GA-B85M-HD3 motherboard, 2x8G RAM, 120G Samsung 840 SDD) Hoping to be able to retire the dual-Xeon-L5420 and the Q6600. I have a box of power meters to try to calibrate how much I'd be saving by using the Haswell instead of the five-year-old kit. Last fiddled with by fivemack on 2013-06-02 at 18:26

2013-06-03, 15:12	#33
sdbardwick Aug 2002 North San Diego County 1010101101₂ Posts	1600K AVX FFT Code: Note values are averages of two runs (4,3,2,1 and 1,2,3,4 workers) 1 Worker 8.2ms 2 Workers 8.7ms 3 Workers 10.32ms 4 Workers 12.50ms For funsies: 1 worker x 4 threads = 3.053 ms 1 worker x 3 threads = 3.34 1 worker x 2 threads = 4.352 1 worker x 1 thread = 8.2 This box uses PC3-1600 RAM recycled from a SB i5-2500; I don't have any faster memory ATM. Attached Thumbnails Last fiddled with by sdbardwick on 2013-06-03 at 15:46 Reason: more funsies

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Haswell-E Prelim. Benchmark	sdbardwick	Hardware	37	2015-02-10 18:49
Prime95 and Haswell	Pleco	Information & Answers	22	2014-07-13 16:03
Haswell Rig	Mini-Geek	Hardware	64	2014-05-27 13:22
Prime95 version 27.1 early preview, not-even-close-to-beta release	Prime95	Software	126	2012-02-09 16:17
Missing mouse-over preview text	retina	Forum Feedback	1	2011-09-12 15:32

2013-06-02, 17:36	#23
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3⁶·13 Posts	Post real-life benchmarks!

2013-06-03, 04:00	#29
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3⁶×13 Posts	Looks quite nice. Thanks!

2013-06-03, 11:16	#30
henryzz Just call me Henry "David" Sep 2007 Cambridge (GMT/BST) 2³×3×5×7² Posts	I didn't realize that TF was affected by memory bandwidth.

2013-06-03, 14:15	#32
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2×53×71 Posts	sdbardwick, Could you run 4 LL tests of the same (or very close) exponents. Then report the timings for just 1 worker running, then 2 workers, 3 workers, and 4 workers? I'm curious if Haswell hits the same memory bandwidth wall as Sandy/Ivy Bridge.