| View previous topic :: View next topic |
| Author |
Message |
bart_nessux
n00b
Joined: 30 Jan 2004
Posts: 11
|
 Posted: Mon Feb 16, 2004 12:19 am Post subject: P4 3.2 Extreme with 2M cache vs. AMD64 3200+ |
 |
|
Can anyone comment on the performance difference between these two procs?
Currently, I have a Athlon 1.4MHz Thunderbird. I've had it for more than two years. Looking to upgrade. Leaning toward the P4 right now... the 875 chipset is super-fast.
Any opinions?
Thanks,
Bart |
|
| Back to top |
|
 |
sneakerski
Apprentice
Joined: 14 Oct 2003
Posts: 168
|
| Posted: Mon Feb 16, 2004 12:50 am Post subject: |
|
|
http://www.linuxhardware.org/article.pl?sid=04/02/02/1654234&mode=thread
just looked at it again more closely, they compiled it with flags which slightly benefit the p4. just a consideration, nothing really major.
_________________
Athlon 64 X2 @ 2ghz on an Asus nForce 4 w/ 1gig
Radeon X850 using OSS Radeon drivers |
|
| Back to top |
|
|
crazycat
l33t
Joined: 26 Aug 2003
Posts: 838
Location: Hamburg, Germany
|
| Posted: Mon Feb 16, 2004 4:25 am Post subject: |
|
|
Well i think amd is worth buyng this time because of great 64 bit performance on linux. And u'll get nearly-40-minutes-bootstrap-time (measured on my athlon64 3000+)  .
check this:
http://www20.tomshardware.com/cpu/20030422/opteron-17.html
dual opertron *dreaming*
here in germany u can get dual board for 400 Euro and two 1.4 opertrons 200 euro each, witch is about a price of p4ee alone. |
|
| Back to top |
|
|
ndraak
n00b
Joined: 29 Jan 2004
Posts: 7
|
| Posted: Mon Feb 16, 2004 5:53 pm Post subject: |
|
|
My opinion would go hands down to the amd64. Strictly for 3 reasons:
1. AMD64's would let you go 64bit gentoo, which rocks 
2. AMD64's compile almost twice as fast as most chips. There was a test site that talked about compilation times and the 3200+ was around 16 minutes for firebird or something and the p4ee came in around 24 minutes along with the other barton series chips comming in even higher.
3. Somewhat of a repeat, but your a gentoo user so you have 64bit at your fingertips, why waste it |
|
| Back to top |
|
|
thumper
Guru
Joined: 06 Dec 2002
Posts: 554
Location: Venice FL
|
| Posted: Tue Feb 17, 2004 1:51 am Post subject: |
|
|
Here is an article with some benchmarks: http://www.linuxhardware.org/Features/03/12/17/189239.shtml
It references an Opteron, compairing to a Xeon.
The Opterons appear a bit faster than the AMD64 family.
My AMD64 3200+ did this (in 64bit mode):
# genlop -t mozilla-firebird
* net-www/mozilla-firebird
Merged at Tue Feb 17 01:37:22 2004 (mozilla-firebird-0.7-r2)
merge time: 26 minutes and 18 seconds.
George |
|
| Back to top |
|
|
Evangelion
Veteran
Joined: 31 May 2002
Posts: 1087
Location: Helsinki, Finland
|
| Posted: Tue Feb 17, 2004 9:23 am Post subject: |
|
|
| thumper wrote: | | The Opterons appear a bit faster than the AMD64 family. |
Opteron is AMD64, just like Athlon 64 is. the difference between the two is that Opteron has 128bit mem-channel whereas A64 has 64bit
_________________
My tech-blog | My other blog |
|
| Back to top |
|
|
thumper
Guru
Joined: 06 Dec 2002
Posts: 554
Location: Venice FL
|
| Posted: Tue Feb 17, 2004 6:35 pm Post subject: |
|
|
| Evangelion wrote: |
Opteron is AMD64, just like Athlon 64 is. the difference between the two is that Opteron has 128bit mem-channel whereas A64 has 64bit |
Yes you are indeed correct, so that makes the Athlon 64 the slow children of the AMD64 family. 
George |
|
| Back to top |
|
|
Isaiah
Guru
Joined: 25 Feb 2003
Posts: 359
|
| Posted: Wed Feb 18, 2004 6:30 am Post subject: |
|
|
| $955.00 vs. $282.00 - any more questions |
|
| Back to top |
|
|
secondshadow
Guru
Joined: 23 Jun 2003
Posts: 362
|
| Posted: Wed Feb 18, 2004 6:39 am Post subject: |
|
|
And that my friend is why I opted for the athlon64. If you are REALLY willing to put down just about a grand for a processor, then maybe an AMD Athlon64 FX-51 would be in order. They're supposed to be all kinds of fast. Or maybe a mid- to high-end dual Opteron. |
|
| Back to top |
|
|
lightvhawk0
Guru
Joined: 07 Nov 2003
Posts: 388
|
| Posted: Thu Feb 19, 2004 7:03 am Post subject: |
|
|
Lol, I'm still waiting for to win the lottery. Then I'll buy my Opteron system.
_________________
If God has made us in his image, we have returned him the favor. - Voltaire |
|
| Back to top |
|
|
NewBlackDak
Guru
Joined: 02 Nov 2003
Posts: 512
Location: Utah County, UT
|
| Posted: Thu Feb 19, 2004 5:23 pm Post subject: |
|
|
I've been looking lately. I know I will go with A64. My Athlon-XP @ 2GHz is faster compiling than my PIV@3GHz, and tmy Athlon-XP at 2450 makes them both look like they're slow. I'm just waiting for the interface change now, and hoping the mobos will be PCI-X, and I can doo all of that at one time. I know not everything will work at first, but maybe I can get in on the ground floor on it.
_________________
Gentoo systems.
X2 4200+@2.6 - Athy
X2 3600+ - Myth
UltraSparc5 440 - sparcy |
|
| Back to top |
|
|
molander
Tux's lil' helper
Joined: 20 Jun 2003
Posts: 110
Location: St. Louis
|
| Posted: Fri Feb 20, 2004 4:41 am Post subject: |
|
|
I believe all the opteron boards require ECC RAM too dont they? |
|
| Back to top |
|
|
absinthe
Retired Dev
Joined: 06 Oct 2002
Posts: 111
Location: San Francisco, CA, USA
|
| Posted: Fri Feb 20, 2004 5:43 am Post subject: |
|
|
Yes, AFAIK all Opteron boards are ECC since the Opterons are targetted at the high-end workstation and server market.
However, it's my opinion that everyone should be using ECC RAM, especially on a new architecture such as AMD64. There's no performance difference, and systems are a lot more stable as a result. You get what you pay for.
People with non-ECC RAM often experience crashes that they think are related to application bugs but are actually memory failures. Who knows how many non-reproduceable bugs get filed that are just memory problems.
With ECC RAM, with the marginal difference in price, this problem virtually goes away. I highly recommend putting high-quality ECC memory in your system before blowing your dough on an insane video card that doesn't even have stable 64-bit drivers yet... <g>
Right now we need users with stable hardware (incl. memory) to be able to diagnose amd64-related bugs effectively... |
|
| Back to top |
|
|
ctford0
l33t
Joined: 25 Oct 2002
Posts: 774
Location: Lexington, KY,USA
|
| Posted: Fri Feb 20, 2004 5:22 pm Post subject: |
|
|
| absinthe wrote: |
However, it's my opinion that everyone should be using ECC RAM, especially on a new architecture such as AMD64. There's no performance difference, and systems are a lot more stable as a result. You get what you pay for.
|
Although I agree with you that ECC ram is great, there is about a 10% performance loss. I have a dual MP system that at one time had a gig of registered non-ecc ram and I then replaced it with a gig of ecc ram. Measuring the performance difference on some computational code that I wrote resulted in almost exactly a 10% increase in the time that it took to run the code.
chris |
|
| Back to top |
|
|
absinthe
Retired Dev
Joined: 06 Oct 2002
Posts: 111
Location: San Francisco, CA, USA
|
| Posted: Fri Feb 20, 2004 6:24 pm Post subject: |
|
|
| ctford0 wrote: | Although I agree with you that ECC ram is great, there is about a 10% performance loss. I have a dual MP system that at one time had a gig of registered non-ecc ram and I then replaced it with a gig of ecc ram. Measuring the performance difference on some computational code that I wrote resulted in almost exactly a 10% increase in the time that it took to run the code.
chris |
I've heard other people say this, but I have done a lot of benchmarking and I have not been able to reproduce any significant difference in speed on systems that support both types of RAM. (which I consider a "real" test instead of just disabling parity checking on ECC RAM in the BIOS).
Crucial.com says ECC is 2% slower, but they've had that on their website for years. I believe a slowdown between 1-5% was evident on old chipsets/motherboards, particularly cheap home consumer oriented ones that attempted to support both. Nowadays on newer chipsets, I'd wager if you ran memtest86+ either way it would be <1% difference.
Given the choice, I'd rather have ECC do error correction than hope that the software handles it right. Especially in higher density DIMMs (1G+).
Always test your ECC RAM thoroughly (with parity checking disabled in the BIOS) when you first receive it. Why?
It's important to note what the difference is between two sticks of RAM: one ECC and one non-parity. If someone has a slightly defective stick of ECC, it'll still work usually, just slower. Which might account for why a handful of people see a significant performance drop on ECC RAM -- because they actually have a defective part -- they just don't realize it. ECC RAM, even when it's slightly defective, will compensate for those defects transparently to the end user... resulting in a performance hit -- but things will keep working in most cases.
On the other hand, a person with defective non-parity RAM won't see slowdowns -- they'll have stability issues. So it will be obvious that there's a RAM problem in the system. |
|
| Back to top |
|
|
Kugelfang
Retired Dev
Joined: 28 Nov 2003
Posts: 25
Location: Witten, Germany
|
| Posted: Fri Feb 20, 2004 6:42 pm Post subject: |
|
|
hm, ECC RAM is no considerably slower than non-ECC RAM, but i think molander meant registered RAM (that's not the same as ECC), which is needed for Opterons as well as the 940 Pin Athlon64 FX.
Both Opteron/Athlon64FX and Athlon64 can use ECC and non-ECC RAM. The fact is that the registered RAM is a bit slower than unregistered RAM. |
|
| Back to top |
|
|
ctford0
l33t
Joined: 25 Oct 2002
Posts: 774
Location: Lexington, KY,USA
|
| Posted: Sat Feb 21, 2004 6:16 am Post subject: |
|
|
| absinthe wrote: |
I've heard other people say this, but I have done a lot of benchmarking and I have not been able to reproduce any significant difference in speed on systems that support both types of RAM. (which I consider a "real" test instead of just disabling parity checking on ECC RAM in the BIOS).
|
The test I ran was even disabling the error checking completely in the bios. I also ran memtest86 on the ecc ram and it returned no errors. I'll give you this, the averaged user would probably never notice the slow down. Unless you were actually trying then you probably could not reproduce it. The code that I wrote to benchmark the speed was a basic fortran program that multiplied two enormous arrays just to fill up as much ram as I could. I actually had a series of programs that would run at the same time so that I could fill up the entire gig of ram. While using that much memory I noticed a 10% degradation in speed just switching from a 1 gig ddr 266 registered memory module to a 1 gig ddr 266 registered ecc memory module. Again, this was even with ecc disabled in the bios.
chris |
|
| Back to top |
|
|
absinthe
Retired Dev
Joined: 06 Oct 2002
Posts: 111
Location: San Francisco, CA, USA
|
| Posted: Sat Feb 21, 2004 8:11 am Post subject: |
|
|
| ctford0 wrote: | | While using that much memory I noticed a 10% degradation in speed just switching from a 1 gig ddr 266 registered memory module to a 1 gig ddr 266 registered ecc memory module. Again, this was even with ecc disabled in the bios.chris |
That may be the case with your particular testbed, but something still smells wrong to me about your particular hardware. I've run exhaustive memory tests between ECC and non-parity using standard benchmarking tools (memtest, cachebench, lmbench, bytemark, etc) and on all modern chipsets I've not noticed any (> 1-2%) difference in performance.
I'm going to have to scare up some hardware now and come back with some numbers to show what I'm talking about. Give me about a week and bump the thread if I forget. |
|
| Back to top |
|
|
Corona688
Veteran
Joined: 10 Jan 2004
Posts: 1204
|
| Posted: Sat Feb 21, 2004 3:50 pm Post subject: |
|
|
| molander wrote: | | I believe all the opteron boards require ECC RAM too dont they? |
I'm running a dual Opteron 242's on a Tyan Thunder K8W, and while it supports ECC, it doesn't need ECC. It does require registered RAM, though, and that was a real bitch to get. Ended up getting registered ECC PC3700 simply because it wouldn't be obselete in 2 weeks. |
|
| Back to top |
|
|
Corona688
Veteran
Joined: 10 Jan 2004
Posts: 1204
|
| Posted: Sat Feb 21, 2004 3:59 pm Post subject: |
|
|
| ctford0 wrote: | | The test I ran was even disabling the error checking completely in the bios. I also ran memtest86 on the ecc ram and it returned no errors. I'll give you this, the averaged user would probably never notice the slow down. Unless you were actually trying then you probably could not reproduce it. The code that I wrote to benchmark the speed was a basic fortran program that multiplied two enormous arrays just to fill up as much ram as I could. |
Whoa there. During some serious compiles, I've watched my Opteron 242's eat up memory REALLY REALLY FAST. It wouldn't take long at all to use a gig of memory. Are you sure you're not measuring the speed of your swapfile? | Quote: | | I actually had a series of programs that would run at the same time so that I could fill up the entire gig of ram. While using that much memory I noticed a 10% degradation in speed just switching from a 1 gig ddr 266 registered memory module to a 1 gig ddr 266 registered ecc memory module. |
That shouldn't happen if ECC's disabled. Were the sticks identical in all other respects, such as timings, etc? Not all RAM is made equal. |
|
| Back to top |
|
|
Corona688
Veteran
Joined: 10 Jan 2004
Posts: 1204
|
| Posted: Sat Feb 21, 2004 4:03 pm Post subject: |
|
|
| Evangelion wrote: | | thumper wrote: | | The Opterons appear a bit faster than the AMD64 family. |
Opteron is AMD64, just like Athlon 64 is. the difference between the two is that Opteron has 128bit mem-channel whereas A64 has 64bit |
Opterons also have that onboard-memory-controller goodness, so running two procs gives you two memory controllers and thus twice the memory bandwidth(In theory. Tyan Tiger boards don't use both memory controllers, but Tyan Thunder boards do.) Or do Athlon64's have that too? |
|
| Back to top |
|
|
ctford0
l33t
Joined: 25 Oct 2002
Posts: 774
Location: Lexington, KY,USA
|
| Posted: Sat Feb 21, 2004 4:11 pm Post subject: |
|
|
| Corona688 wrote: | | That shouldn't happen if ECC's disabled. Were the sticks identical in all other respects, such as timings, etc? Not all RAM is made equal. |
AFAIK the ram was exactly the same except for the one being ecc. This could be the cause since I don't know for sure the timings,etc and since I dont have the non-ecc chip anymore I really can't run the test again.
I had also read many things around the web on the same phenomena (slow down) so I wasnt surprised by the results of my test.
Chris |
|
| Back to top |
|
|
|
Gentoo on AMD64
