Six-core 32nm Westmere Full Review @ HKEPC

Printable View

Show 100 post(s) from this thread on one page

08-10-2009, 08:58 PM
informal

Quote:

Originally Posted by hurleybird

It's system power consumption, using low power components (gts 210)

That's correct. The only thing i couldn't quite understand was the author's i7-965 comment under the chart with power consumption numbers. Both CPUs tested should be 2.4Ghz models.

Quote:

Originally Posted by kl0012

Bulldozer isn't mentioned in AMDs 2011 roadmap anymore. It was expectable since AMD had to adopt Intel AVX extension. BTW Q4-10/Q1-11 is a time of Sandy Bridge intro which was taped out 2 months ago, probably because of longer debug/test cycle of new gen + graphics.

If BD is not 2011 thing anymore(as you say),when is it supposed to launch then?!
BD cores are now already in 2nd major revision,first being "canceled" back in the 2007.Since then,it "gained" 256bit AVX support and SSE5 on top of that(XOP etc. in new renamed nomenclature).
So Chad is right,BD is in best case scenario late 2010,or in worst case Q2 2011.
08-10-2009, 11:52 PM
grimREEFER

if theres a cheap model of this, theres no reason to not get it.
this is great for ppl who do bluray transcoding.
x264 is fully multithreaded, so ppl should get huge gains.
especially if you use constant rate factor mode instead of 2 pass mode.
08-11-2009, 12:31 AM
largon

Is there a 32nm QC (native or not) part with 192bit memory coming anytime?
08-11-2009, 12:36 AM
R101

This can encode 6 WMV9 streams and deinterlace in real time. :D
08-11-2009, 12:54 AM
jcool

Naw, 12 --> HT :D
08-11-2009, 12:58 AM
ajaidev

Quote:

Originally Posted by kl0012

Bulldozer isn't mentioned in AMDs 2011 roadmap anymore. It was expectable since AMD had to adopt Intel AVX extension. BTW Q4-10/Q1-11 is a time of Sandy Bridge intro which was taped out 2 months ago, probably because of longer debug/test cycle of new gen + graphics.

Orochi = Bulldozer

Any estimations on how big the die size will be?? I would guess around 280mm2. Also is the speed suppose to be 2.4Ghz or faster because even i7 965 had a ES with the proper speed so maybe this ES also has the correct speed??

Quote:

Originally Posted by largon

Is there a 32nm QC (native or not) part with 192bit memory coming anytime?

I think 2011!! Sandy and Bull both will support 192 bit memory. That is of course only possible if Intel does not release a i7 "Westmere" before 2011. Ahh if you reallly crave it that much get a i9 :>
08-11-2009, 01:01 AM
BrowncoatGR

I think this chip will, initially at least, have the same problem AMDs 3cores and 6cores have. Which is software not properly supporting CPUs with a core count that is not a power of 2.
08-11-2009, 03:03 AM
Hornet331

Quote:

Originally Posted by LOE

3x2=6...

http://en.wikipedia.org/wiki/Power_of_two

theres no six in the "power of two". :p:
08-11-2009, 04:03 AM
Chad Boga

Quote:

Originally Posted by Hornet331

http://en.wikipedia.org/wiki/Power_of_two

theres no six in the "power of two". :p:

Whilst I obviously don't disagree with your post, I am wondering if the notion of having cores to the power of two matter much any more?

It appears that Intel will be bringing out a 10 core successor to the 8 core Nehalem-EX. :shrug:
08-11-2009, 04:09 AM
zalbard

Quote:

Originally Posted by Chad Boga

It appears that Intel will be bringing out a 10 core successor to the 8 core Nehalem-EX. :shrug:

10? I can see them using 12, but not 10... :shrug:
08-11-2009, 04:24 AM
Hornet331

Quote:

Originally Posted by Chad Boga

Whilst I obviously don't disagree with your post, I am wondering if the notion of having cores to the power of two matter much any more?

It appears that Intel will be bringing out a 10 core successor to the 8 core Nehalem-EX. :shrug:

There where/are some apps out there that have a "hardcoded" ammount of threads, 1,2,4 etc (power of two). If they are confronted with a cpu that has an odd number of threds (3,6,12 etc) they fall back to the next lower number.

But most apps shouldn't have a problem, cause the either don't scale past a certain number of threads or allow any number of threads.

The problem is even lower for HPC (beckton) software since there you code your software to fit the hardware its run on.
08-11-2009, 05:18 AM
Particle

Don't get too excited about x264 performance. If you're transcoding, chances are you're using Avisynth. Avisynth kills efficiency when talking about lots of threads going at it in x264.
08-11-2009, 05:58 AM
Chumbucket843

Quote:

Originally Posted by Chad Boga

Whilst I obviously don't disagree with your post, I am wondering if the notion of having cores to the power of two matter much any more?

It appears that Intel will be bringing out a 10 core successor to the 8 core Nehalem-EX. :shrug:

depends on the program.
08-11-2009, 09:06 AM
Face

some info dugg from the article:

Quote:

32nm system in the next blessing, Intel plans in the second quarter of 2010 to launch the first of six core DT processor core, code-named "Gulftown".

Quote:

"According to Intel to disclose to the motherboard industry, Intel Gulftown six core eventually named after a very good chance they would not use the Core i9 family, but Core i7-1000 family, because there is no micro-architecture Gulftown changes, only the core of the increase in the number not constitute a new brand."

So apparently Q2 2010 release with the Core i7-1000 moniker.
08-11-2009, 09:10 AM
poke349

Quote:

Originally Posted by Hornet331

There where/are some apps out there that have a "hardcoded" ammount of threads, 1,2,4 etc (power of two). If they are confronted with a cpu that has an odd number of threds (3,6,12 etc) they fall back to the next lower number.

Or just round up to the next power of two and let the scheduler take care of it. :rofl::rofl::rofl:
08-11-2009, 09:26 AM
Schlotkins

Is this going to run on current X58 boards? Thanks.
08-11-2009, 09:27 AM
Tenknics

Quote:

Originally Posted by Face

some info dugg from the article:

So apparently Q2 2010 release with the Core i7-1000 moniker.

I think that article could be right about i7 branding and not using i9 but it has the wrong reasons. Westmere, like the last tick Penryn, will have micro-architecture changes. I remember reading in many places Westmere has changes to it's core, they might be minor but they are still new changes. Penryn didn't get a brand refresh or specific new brand. It adopted its tock brand Core2 Duo/Quad just with different/higher model #s. So its not that shocking that Westmere will continue the same brands as Nehalem but with higher model #s. And so far there have been 0 32nm 4core codenames or news or anouncements. 32nm with be 6/12 and 2/4 for the most part. But then again this whole iX things makes no sense anymore and its Intel so who knows!:shrug:

edit: Found it.

http://img.hexus.net/v2/articles/Int...Advantages.png

Quote:

because there is no micro-architecture Gulftown changes, only the core of the increase in the number not constitute a new brand."

Riiiiiiight:down:
08-11-2009, 10:23 AM
Katanai

Quote:

Originally Posted by Schlotkins

Is this going to run on current X58 boards? Thanks.

Yes. Though I don't know what the point of this chip, in the desktop segment, is. Now, in the server arena it does make sense if it's priced right.
08-11-2009, 11:06 AM
gOJDO

Quote:

Originally Posted by Face

So apparently Q2 2010 release with the Core i7-1000 moniker.

There is no logic in such naming or there would be no Core i5 and Core i3, but Core i7-500 and Core i7-300.
08-11-2009, 11:16 AM
Jacky

Quote:

Originally Posted by Katanai

Yes. Though I don't know what the point of this chip, in the desktop segment, is. Now, in the server arena it does make sense if it's priced right.

I suppose turbo mode could be the saving grace for single threaded performance. Does anyone know how far it will go? (or will it get not much of a turbo mode á la i7)
08-11-2009, 11:24 AM
Serra

I'm a little sad to see that there are really no improvements except for AES handling, but I'm assuming it's more of a software issue. That's fine with me though, I mean even if it can do the same old things at the same old speed, it can handle 50% more of the same old things simultaneously.

Need DP version released now! 24 threads would be much nicer than 16.

I want to see this AES handling in action though. I like the example use given as full hard drive encryption, and I would love to know how much it consumes in terms of CPU resources and how much of a performance impact it is to the disk subsystem. For example, would encrypting your whole disk on a 6-core give you about the same performance as a 4-core non-encrypted system, with a slight hit to disk performance... or is it better/worse? To me this is a huge question.
08-11-2009, 12:20 PM
jcool

Quote:

Originally Posted by Serra

Need DP version released now! 24 threads would be much nicer than 16.

Gulftown automatically comes with a 2nd QPI link just like Gainestown, so you can run them in 1P and 2P rigs. That's the whole point really, and the one thing AMD is missing with the Istanbuls.

Quote:

I want to see this AES handling in action though. I like the example use given as full hard drive encryption, and I would love to know how much it consumes in terms of CPU resources and how much of a performance impact it is to the disk subsystem. For example, would encrypting your whole disk on a 6-core give you about the same performance as a 4-core non-encrypted system, with a slight hit to disk performance... or is it better/worse? To me this is a huge question.

I fear this AES business is pretty useless.. I ran a Truecrypt bench on my old Harper more than a year ago, as you can see it's more than fast enough arleady to handle plain 128bit AES ;)

http://database.he-computer.de/Bilder/truecryptz7s.jpg

On the Dual X5450 it's like 1,2GB/s... pretty hard getting a storage system to max out the CPus with simple AES ;)
08-11-2009, 12:25 PM
Helloworld_98

0.072v on a quad?

Show 100 post(s) from this thread on one page