Haswell Preview Benchmark - Page 2

TheJudger · 2013-05-31, 19:44

Quote:

Originally Posted by TheMawn

Performance-wise I can't really see how much of an improvement one will see from DDR4 over DDR3. There aren't many real applications limited by memory speed.

Well... which applications are you talking about? In HPC business I know alot of applications which are totally limited by available memory bandwidth. Especially codes which were initially written for vector CPUs...
e.g. http://en.wikipedia.org/wiki/NEC_SX-8, 16GFLOPS per CPU, 64GB/s per CPU. 4 Bytes per FLOP, compare this with current (Desktop-)CPUs..

Oliver

TheMawn · 2013-06-02, 01:28

Wow. It makes a good amount of sense to me that memory speed should actually matter, considering how much fuss manufacturers make about it. I'm glad my 2400MHz 10-12-11-31 is helping the cause.

I was always under the impression that the CPU was rarely able to handle all the data that memory could supply. I guess it does depend on the type of data.

I might run a few benchmarks when I get bored. I'll clock my RAM down to maybe even 1200MHz if I can manage it and see how much of an impact it makes. I was thinking that since the iterations do take a good amount of time, the CPU has to do a lot more "crunching" than "remembering". I guess a person DOES learn something new every day.

kladner · 2013-06-02, 01:41

Quote:

Originally Posted by TheJudger

(snip)
e.g. http://en.wikipedia.org/wiki/NEC_SX-8, 16GFLOPS per CPU, 64GB/s per CPU. 4 Bytes per FLOP, compare this with current (Desktop-)CPUs..

Oliver

.....from Wikipedia- (I sure understand the sad face better now. Actually, wouldn't the example below end up at 12.5 GB/s? MB/s / 1024?)

Quote:

For example, a computer with dual-channel memory and two DDR2-800 modules per channel running at 400 MHz would have a theoretical maximum memory bandwidth of:
400,000,000 clocks per second × 2 lines per clock × 64 bits per line × 2 interfaces = 102,400,000,000 (102.4 trillion) bits per second (in bytes, 12,800 MB/s or 12.8 GB/s)

The table here http://en.wikipedia.org/wiki/Dynamic..._access_memory gives 12.8 GB/s for the DDR3-1600 RAM that I'm running right now, but I am running dual channel, so maybe this machine is effectively twice that.

sdbardwick · 2013-06-02, 01:50

Quote:

Originally Posted by kladner

.....from Wikipedia- (I sure understand the sad face better now. Actually, wouldn't the example below end up at 12.5 GB/s? MB/s / 1024?)

GiB vs GB

kracker · 2013-06-02, 01:58

http://anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested
http://www.tomshardware.com/reviews/...view,3521.html

Yawn... I was hoping for more.

kladner · 2013-06-02, 02:36

Quote:

Originally Posted by sdbardwick

GiB vs GB

So the binary is GiB, decimal is GB? The piece above still confuses me a bit, as it also states that

Quote:

When referring to computer memory, gigabyte is always a “power of two” - 1,073,741,824 bytes, but when measuring hard drive capacity it is often defined as 1,000,000,000 bytes.

But the numbers being discussed were memory bandwidth. Is this just a case of sloppy usage? (It also seems that I was probably using the wrong denominator in any case.)

Sorry for going O/T on such elementary items.

nucleon · 2013-06-02, 14:35

Didn't we roughly come to the conclusion with P-1, we hit saturation with:

hex core SB-E, was limited with P-1 with 4x cores concurrent.
quad core SB/IB was limited with P-1 with 3x cores concurrent.

Obviously such wide statements have a bit of approximation to them. (i.e. 3.5+GHz cpu clocks, 1600MHz mem clocks)

What makes me curious about Haswell - the execution units have more ports. Makes me wonder if HT has a greater impact to performance. I wonder if more performance can be gained with running 8x processes on the quad core haswell vs *bridge CPUs.

-- Craig

kladner · 2013-06-02, 14:36

It's Heeere!
http://www.microcenter.com/site/bran...r-bundles.aspx

kracker · 2013-06-02, 14:46

It's heeeeree!!!

http://promotions.newegg.com/Intel/13-2142/index.html

chappy · 2013-06-02, 15:13

I woke up to an email from Newegg and an email from Microcenter, taunting me with these chips. Microcenter is just a 7 minute drive, but how much would that end up costing me. :)

kracker · 2013-06-02, 15:16

Quote:

Originally Posted by chappy

I woke up to an email from Newegg and an email from Microcenter, taunting me with these chips. Microcenter is just a 7 minute drive, but how much would that end up costing me. :)

I am so tempted.. but I just got Ivy Bridge a month ago! Damn!

2013-06-02, 14:36	#19
kladner "Kieren" Jul 2011 In My Own Galaxy! 2·3·1,693 Posts	Haswell has Arrived! It's Heeere! http://www.microcenter.com/site/bran...r-bundles.aspx

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, 01:28	#13
TheMawn May 2013 East. Always East. 11·157 Posts	Wow. It makes a good amount of sense to me that memory speed should actually matter, considering how much fuss manufacturers make about it. I'm glad my 2400MHz 10-12-11-31 is helping the cause. I was always under the impression that the CPU was rarely able to handle all the data that memory could supply. I guess it does depend on the type of data. I might run a few benchmarks when I get bored. I'll clock my RAM down to maybe even 1200MHz if I can manage it and see how much of an impact it makes. I was thinking that since the iterations do take a good amount of time, the CPU has to do a lot more "crunching" than "remembering". I guess a person DOES learn something new every day.

2013-06-02, 01:58	#16
kracker "Mr. Meeseeks" Jan 2012 California, USA 2³·271 Posts	http://anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested http://www.tomshardware.com/reviews/...view,3521.html Yawn... I was hoping for more.

2013-06-02, 14:35	#18
nucleon Mar 2003 Melbourne 5×103 Posts	Didn't we roughly come to the conclusion with P-1, we hit saturation with: hex core SB-E, was limited with P-1 with 4x cores concurrent. quad core SB/IB was limited with P-1 with 3x cores concurrent. Obviously such wide statements have a bit of approximation to them. (i.e. 3.5+GHz cpu clocks, 1600MHz mem clocks) What makes me curious about Haswell - the execution units have more ports. Makes me wonder if HT has a greater impact to performance. I wonder if more performance can be gained with running 8x processes on the quad core haswell vs *bridge CPUs. -- Craig

2013-06-02, 14:46	#20
kracker "Mr. Meeseeks" Jan 2012 California, USA 2³×271 Posts	It's heeeeree!!! http://promotions.newegg.com/Intel/13-2142/index.html

2013-06-02, 15:13	#21
chappy "Jeff" Feb 2012 St. Louis, Missouri, USA 13×89 Posts	I woke up to an email from Newegg and an email from Microcenter, taunting me with these chips. Microcenter is just a 7 minute drive, but how much would that end up costing me. :)