AMD VR INTEL

Message boards : Number crunching : AMD VR INTEL

To post messages, you must log in.

AuthorMessage
Profile Rangers
Avatar

Send message
Joined: 27 Oct 08
Posts: 39
Credit: 6,502,073
RAC: 159
Message 71307 - Posted: 22 Sep 2011, 0:18:57 UTC

is there a large structural difference as far as number crunching goes with the phenom vs i7. Ive heard about floating point math or something like that that handles all the math on amd procesors, is this still the case? I googled and couldnt find a difference so I will ask. I would be able to do alot more with amd.
ID: 71307 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Rangers
Avatar

Send message
Joined: 27 Oct 08
Posts: 39
Credit: 6,502,073
RAC: 159
Message 71308 - Posted: 22 Sep 2011, 0:21:13 UTC - in response to Message 71307.  

and i am willing to give up ht just for the cheaper price really, input ?
ID: 71308 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 114,398,661
RAC: 54,591
Message 71310 - Posted: 22 Sep 2011, 8:50:15 UTC

Hi benjamin

As far as Rosetta is concerned, the fastest i5 (which is no doubt overclocked) has a similar average credit to the fastest AMD 1090T. I don't know what speeds they're running at but the i5 has 4 cores and the 1090T has 6, so the i5 is faster per core, but then you can get more AMD cores for the same money. With HT, the top end i7's listed are getting up to 20% more throughput again, but that might be at the expense of more than 20% more electricity. I'd like to give a straight answer but I don't have the info I'm afraid!

To be honest though, if you can wait a few weeks then it will be worth seeing how the 8-thread Bulldozer performs - latest rumour is 12th Oct release and that seems more robust as prices are appearing (quite low prices too) and the server versions are already shipping to the top tier builders.
ID: 71310 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
.clair.

Send message
Joined: 2 Jan 07
Posts: 274
Credit: 26,399,595
RAC: 3
Message 71319 - Posted: 22 Sep 2011, 20:34:18 UTC

I have had my new build on hold for quite some time now.
Sooner or later AMD acctualy will release Bulldozer to us that would like play with it.
Though it is being a long wait.
ID: 71319 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1847
Credit: 7,991,670
RAC: 8,700
Message 71404 - Posted: 12 Oct 2011, 9:57:00 UTC - in response to Message 71319.  

Sooner or later AMD acctualy will release Bulldozer to us that would like play with it.


See the test of BZ on the net.
It's a fake, bad processor...
The next Ivy Bridge of Intel will destroy it
ID: 71404 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 114,398,661
RAC: 54,591
Message 71406 - Posted: 12 Oct 2011, 17:34:58 UTC - in response to Message 71404.  

Sooner or later AMD acctualy will release Bulldozer to us that would like play with it.


See the test of BZ on the net.
It's a fake, bad processor...
The next Ivy Bridge of Intel will destroy it


Not sure what you mean by fake? Bulldozer's strengths are multi-threaded apps and its price point for those apps. It doesn't compete very well on low threaded workloads but Rosetta is massively multithreaded (at quite a coarse level) so it might do very well against Sandy Bridge on Rosetta, pound for pound.
ID: 71406 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile granno21
Avatar

Send message
Joined: 19 Jan 06
Posts: 6
Credit: 314,667
RAC: 0
Message 71417 - Posted: 17 Oct 2011, 2:40:28 UTC

AMD's Bulldozer has been released to some underwhelming reviews. I was wondering if there was any information on how it performs for Rosetta.

I'm looking at a new crunching computer and I'm not sure which route to go.


ID: 71417 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
.clair.

Send message
Joined: 2 Jan 07
Posts: 274
Credit: 26,399,595
RAC: 3
Message 71421 - Posted: 17 Oct 2011, 18:20:35 UTC
Last modified: 17 Oct 2011, 19:12:23 UTC

The main problem i see with Bulldozer is that it has only fore floating point units,
Each pair of integer core`s share one FPU, even though the FPU`s have been `optimized` to take account of this and do have supprizing performance as they are,
With BOINC projects that are FPU intensive surely this will be a bottle neck for us,
And in the few tests so far published the 1100T can keep up in many of the ways we are likely to use them when the CPU is subjected to 24/7 100% load and the fact of eight work units trying to use the limited number of FPU`s at the same time, only time will tell and when someone is brave enough to buy one and put it on Rosetta to see what it can do in real world tests.
I would still like to buy one but will wait a bit longer to see how things work out.

There is also a thread on seti@home about Bulldozy
http://setiathome.berkeley.edu/forum_thread.php?id=65764
ID: 71421 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 114,398,661
RAC: 54,591
Message 71424 - Posted: 17 Oct 2011, 22:19:03 UTC - in response to Message 71421.  

The main problem i see with Bulldozer is that it has only fore floating point units,
Each pair of integer core`s share one FPU, even though the FPU`s have been `optimized` to take account of this and do have supprizing performance as they are,
With BOINC projects that are FPU intensive surely this will be a bottle neck for us,
And in the few tests so far published the 1100T can keep up in many of the ways we are likely to use them when the CPU is subjected to 24/7 100% load and the fact of eight work units trying to use the limited number of FPU`s at the same time, only time will tell and when someone is brave enough to buy one and put it on Rosetta to see what it can do in real world tests.
I would still like to buy one but will wait a bit longer to see how things work out.

There is also a thread on seti@home about Bulldozy
http://setiathome.berkeley.edu/forum_thread.php?id=65764


I'm fairly sure I recall someone stating that Rosetta has more integer work than float. Not sure if that's true, and I can't find the thread but have foudn this: https://boinc.bakerlab.org/rosetta/forum_thread.php?id=4907...
ID: 71424 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
.clair.

Send message
Joined: 2 Jan 07
Posts: 274
Credit: 26,399,595
RAC: 3
Message 71433 - Posted: 19 Oct 2011, 1:11:36 UTC

Thanks for the link,

ID: 71433 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Prom

Send message
Joined: 21 Jun 06
Posts: 23
Credit: 931,604
RAC: 0
Message 71452 - Posted: 21 Oct 2011, 0:43:23 UTC - in response to Message 71424.  

The main problem i see with Bulldozer is that it has only fore floating point units,
Each pair of integer core`s share one FPU, even though the FPU`s have been `optimized` to take account of this and do have supprizing performance as they are,
With BOINC projects that are FPU intensive surely this will be a bottle neck for us,
And in the few tests so far published the 1100T can keep up in many of the ways we are likely to use them when the CPU is subjected to 24/7 100% load and the fact of eight work units trying to use the limited number of FPU`s at the same time, only time will tell and when someone is brave enough to buy one and put it on Rosetta to see what it can do in real world tests.
I would still like to buy one but will wait a bit longer to see how things work out.

There is also a thread on seti@home about Bulldozy
http://setiathome.berkeley.edu/forum_thread.php?id=65764


I'm fairly sure I recall someone stating that Rosetta has more integer work than float. Not sure if that's true, and I can't find the thread but have foudn this: https://boinc.bakerlab.org/rosetta/forum_thread.php?id=4907...

Either way it doesn't really matter. Bulldozer has one shared FPU per module for the new 256-bit instructions but for the traditional 128-bit instructions it's still 2 FPUs so no difference for Rosetta or any other app currently available. It would make sense therefor to restrict all the BOINC apps to 128-bit only.

I would like to see some Bulldozer Rosetta benchmarks as opposed to the synthetic ones we are seeing now.
BBLounge - Broadband and Technology forum
ID: 71452 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
.clair.

Send message
Joined: 2 Jan 07
Posts: 274
Credit: 26,399,595
RAC: 3
Message 71486 - Posted: 25 Oct 2011, 1:06:20 UTC

A good, and rather long write up on `Dozer and the chips that will follow in it`s tracks can be found here,
I will leave you to decide weather `Dozer is a chip that in some ways arrived to early for what it is intended to do.

http://arstechnica.com/gadgets/news/2011/10/can-amd-survive-bulldozers-disappointing-debut.ars

And on the first page of the review that photo is a loading shovel.
Ok, im`e being fussy :¬)
ID: 71486 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 114,398,661
RAC: 54,591
Message 71509 - Posted: 27 Oct 2011, 14:31:31 UTC

I can now add that my Phenom II X3 (was an X2 with one of the disabled cores unlocked) (3GHz x 3 = 9GHz) slightly outperforms my Q6600 (2.4GHz x 4 = 9.6GHz). My E8400 (Q6600 = 65nm, Phenom II and E8400 = 45nm) gets a similar RAC to the Phenom II per core (both @ 3GHz).

My i3 gets a much lower score per core per GHz - I presume because of the limited cache. I run 2 tasks at a time rather than 4 as 1.5MB L2 cache is already stretched between two tasks. The credit per hour per core increases significantly when running one task on the i3 but running 2 tasks still gets through more work (just far from 2x as much).

I haven't measured the power consumption of any of them for a long time so can't comment on that front...
ID: 71509 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 114,398,661
RAC: 54,591
Message 71898 - Posted: 27 Dec 2011, 14:56:17 UTC

I'm looking to upgrade my main PC and have been comparing some CPUs. Think I'll be going with a Sandy Bridge chip for Autodesk Inventor and Excel performance but in case it's of use, I've just done the maths on a Bulldozer/FX CPU. Looks like an 8-thread 8120 does quite well on Rosetta - this one:

https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1489011 is doing an average of 537 credits/thread/day = RAC of 4300. Don't know if it's overclocked but those figures are similar to an i7-2600. Not sure what the power consumption will be like with Rosetta as it tends to take it toward peak consumption but Prime95 uses more power on my PCs than Rosetta does.

Thought I'd post it seeing as I've checked...
ID: 71898 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Paul

Send message
Joined: 29 Oct 05
Posts: 193
Credit: 65,736,681
RAC: 2,734
Message 72354 - Posted: 18 Feb 2012, 13:37:58 UTC - in response to Message 71898.  

I have only used Intel to date on the project and have been very pleased with them. The good news is that either of these highend processors will do a great job and we look forward to another strong computer donating time to the project.

Buy one, get it up and running, connected to the project and crunching. The sooner you start the better. It would be difficult for either processor to make up for the delay of a perfect decision.

Thanks for your contribution to R@H!
Thx!

Paul

ID: 72354 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 114,398,661
RAC: 54,591
Message 72357 - Posted: 18 Feb 2012, 15:54:26 UTC

I decided to wait for Ivy Bridge for the lower power and so hopefully the IGP will be sufficient for CAD. 8th April is the rumour...
ID: 72357 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
AlphaLaser

Send message
Joined: 19 Aug 06
Posts: 52
Credit: 3,327,939
RAC: 0
Message 72398 - Posted: 27 Feb 2012, 3:37:50 UTC - in response to Message 72357.  

I decided to wait for Ivy Bridge for the lower power and so hopefully the IGP will be sufficient for CAD. 8th April is the rumour...


Unfortunately according to this Ivy Bridge has been delayed...
ID: 72398 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 114,398,661
RAC: 54,591
Message 72400 - Posted: 27 Feb 2012, 14:55:13 UTC - in response to Message 72398.  

I decided to wait for Ivy Bridge for the lower power and so hopefully the IGP will be sufficient for CAD. 8th April is the rumour...


Unfortunately according to this Ivy Bridge has been delayed...


VR-Zone says it's just the dual-core chips that are delayed:
http://vr-zone.com/articles/intel-is-only-delaying-dual-core-mobile-ivy-bridge-cpus/14933.html

Hope that's true!
ID: 72400 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Number crunching : AMD VR INTEL



©2024 University of Washington
https://www.bakerlab.org