AMD's and this is being a bit general
Skt A 256
Skt 754 512
Skt 939 512 or 1024
Skt AM2 1024 or 2048
It was a bit general indeed... for 939 and AM2 (non-mobile)
For 939 single core (Winchester / Venice, San Diego), only the San Diego core models (3700+ and 4000+) are 1MB. All others are 512K.
For 939 X2, nearly all are 512K with the exception of the 4400 and 4800.
Single core AM2 is 512K.
For AM2 flavors of the X2 line, both cache level units out there. AMD has mostly gone to dedicated 512K per core on the lower end. I'm seeing 1MB x 2 starting at the 5200+ (AM2). Apparently the higher clock rate models are impacted a little bit more by having more cache than the lower clocked units, likely because even though K8 isn't memory starved, as the clock speed ramps up the memory consumption will ramp up too.
All FX models and all Opterons for both 939 and AM2 are 1MB per core.
So, seeing as how my San Diego core 3700+ has 1MB, either more than that is needed to not see a penalty...or it's not all that much cache-influenced... :shrug:
AMD's and this is being a bit general
Skt A 256
Skt 754 512
Skt 939 512 or 1024
Skt AM2 1024 or 2048
It was a bit general indeed... for 939 and AM2 (non-mobile)
For 939 single core (Winchester / Venice, San Diego), only the San Diego core models (3700+ and 4000+) are 1MB. All others are 512K.
For 939 X2, nearly all are 512K with the exception of the 4400 and 4800.
Single core AM2 is 512K.
For AM2 flavors of the X2 line, both cache level units out there. AMD has mostly gone to dedicated 512K per core on the lower end. I'm seeing 1MB x 2 starting at the 5200+ (AM2). Apparently the higher clock rate models are impacted a little bit more by having more cache than the lower clocked units, likely because even though K8 isn't memory starved, as the clock speed ramps up the memory consumption will ramp up too.
All FX models and all Opterons for both 939 and AM2 are 1MB per core.
So, seeing as how my San Diego core 3700+ has 1MB, either more than that is needed to not see a penalty...or it's not all that much cache-influenced... :shrug:
So, I'm an Intel man, cause thats what I get from work, at cost -25%.
And WD HDD's.
Here is an update with some extra boxes added to the above list and more performance data (CCH) based on completed results rather than estimates. Once again (C) is Coppermine, (T) is Tualatin, (W) is Williamette and (B) is Barton. The speed given is actual CPU speed (ie the Barton cored Athlon is overclocked - same for both OS's).
*A. PIII 866 Dual Xeon (C) - 35% increase in CCH (validated results)
*B. PIII 866 Dual Xeon (C) - 36% increase in CCH (completed results)
*C. PIII 866 Dual (C) - 41% increase in CCH (estimated - results @ 80% done)
*D. PIII 1266 Dual (T) - 33% increase in CCH (completed result)
*E. PIII 1200 Single (T) - 29% increase in CCH (validated - result)
*F. P4 1700 Single (W) - 2% decrease in CCH (estimated - result @ 30% done)
*G. AMD Athlon 2200 (B) - 47% increase in CCH (completed result)
EDIT:
More details about each of the boxes so far converted to Linux.
ID Machine Details (speed/mem/HD/L2 cache) Win CCH Linux CCH Change
A. Compaq SP750 Workstation 866/512/18/256 4.23 5.73 +35%
B. Compaq SP750 Workstation 866/256/18/256 4.23 5.76 +36%
C. HP Netserver E800 Server 866/384/40/256 4.30 6.07 +41%
D. IBM e-Server xSeries 1266/1024/40/512 6.54 8.68 +33%
E. Dell Optiplex GX-150 1200/512/40/256 6.05 7.81 +29%
F. Dell Optiplex GX-240 1700/256/20/256 7.29 7.16 -2%
G. Clone Soltek nForce2 2200/512/20/512 11.96 17.61 +47%
PS:
If you compare E and F you will see why I love Tualatins and hate early P4s :).
Machine G shows just how good Athlon XPs were in their day as well.
I don't know that L2 cache size is all that significant here.
So, I'm an Intel man, cause thats what I get from work, at cost -25%.
And WD HDD's.
Wasn't trying to knock you...just adding clarity. :-)
WD you say? Ever get the ole Click-o-Death?
BTW, what's a RaptorX go for across the pond?
We have not had many HDD failures, but we don't do much database type work, which I believe, but maybe wrong as usual, is where the problem lies. Plus compared to number of cpu's and memory chips we use our HDD count is very small.
Raptor X, which we don't use, is listed at £126 less taxes. I could get but must order full box of 10 min. There are discounts for bulk purchases, and I think the company gets about 8% off for ordering in 100's.
And we only use Enterprise class, not the slightly cheaper mainstream ones.
We have not had many HDD failures, but we don't do much database type work, which I believe, but maybe wrong as usual, is where the problem lies.
I built about a thousand (yes, thousand) machines over the course of about a year when I was working for a small OEM back in 1998. This was consumer grade, not enterprise... At any rate, we had a mix of Fujitsu, WD, Maxtor, and Seagate. The largest number of failures happened with the WD and Seagate drives. The Fujitsu drives were just painfully slow. Maxtor was the faster and most reliable. There were 3 other builders (some were faster than me) so somewhere between 4500 and 5000 installed drives, perhaps 5500ish ordered drives. Not a real big run, but enough to get a general feel of the performance and reliability...
Remembering all of that, I had serious reservations about getting the Seagate that's in my system. The first one that arrived was defective. It caused the system to hang at boot, but with a misleading indicator on my LED diagnostic board that goes with my motherboard. It was pointing at FDD failure. The only way I figured it out, since I didn't have access to multiple drives like I did back then, was the fact that SATA is hot pluggable. Boot to a floppy and while the floppy is first loading, plug the cable in. Ran Seagate Tools and it said it was bad. I did an RMA with Newegg and got the second one and it has worked like a champ...
To prove the point that it's cache-related on Intels, I'll try to negotiate with my brother... he's got one of those really nice late Pent Ms with an enormous 2 MB of L2 cache (it's called a Dothan iirc), a likely machine to get away totally unaffected if we are correct... doubt he'll let me run it off a live CD, though... anyone else got a machine like this?
Just found the T2060 on
)
Just found the T2060 on Intels site
Confirmed as 1 MB, but I will not be critical as Samsung (UK) have it listed as 2MB.
Andy
RE: AMD's and this is
)
It was a bit general indeed... for 939 and AM2 (non-mobile)
For 939 single core (Winchester / Venice, San Diego), only the San Diego core models (3700+ and 4000+) are 1MB. All others are 512K.
For 939 X2, nearly all are 512K with the exception of the 4400 and 4800.
Single core AM2 is 512K.
For AM2 flavors of the X2 line, both cache level units out there. AMD has mostly gone to dedicated 512K per core on the lower end. I'm seeing 1MB x 2 starting at the 5200+ (AM2). Apparently the higher clock rate models are impacted a little bit more by having more cache than the lower clocked units, likely because even though K8 isn't memory starved, as the clock speed ramps up the memory consumption will ramp up too.
All FX models and all Opterons for both 939 and AM2 are 1MB per core.
So, seeing as how my San Diego core 3700+ has 1MB, either more than that is needed to not see a penalty...or it's not all that much cache-influenced... :shrug:
RE: RE: AMD's and this
)
So, I'm an Intel man, cause thats what I get from work, at cost -25%.
And WD HDD's.
Andy
Or it is cache-related on
)
Or it is cache-related on Intels, but affects all AMDs. Maybe it isn't even ONE problem with the Windows app but two unrelated ones...
RE: Coppermine 256KB @ 1/2
)
IIRC, the Coppermine's L2 runs on full clk speed.
Anayway, the "All intels with small chache and all AMD are affected" theory sounds plausible to me.
RE: *PIII 866 Dual Xeon
)
Here is an update with some extra boxes added to the above list and more performance data (CCH) based on completed results rather than estimates. Once again (C) is Coppermine, (T) is Tualatin, (W) is Williamette and (B) is Barton. The speed given is actual CPU speed (ie the Barton cored Athlon is overclocked - same for both OS's).
*B. PIII 866 Dual Xeon (C) - 36% increase in CCH (completed results)
*C. PIII 866 Dual (C) - 41% increase in CCH (estimated - results @ 80% done)
*D. PIII 1266 Dual (T) - 33% increase in CCH (completed result)
*E. PIII 1200 Single (T) - 29% increase in CCH (validated - result)
*F. P4 1700 Single (W) - 2% decrease in CCH (estimated - result @ 30% done)
*G. AMD Athlon 2200 (B) - 47% increase in CCH (completed result)
EDIT:
More details about each of the boxes so far converted to Linux.
ID Machine Details (speed/mem/HD/L2 cache) Win CCH Linux CCH ChangeA. Compaq SP750 Workstation 866/512/18/256 4.23 5.73 +35%
B. Compaq SP750 Workstation 866/256/18/256 4.23 5.76 +36%
C. HP Netserver E800 Server 866/384/40/256 4.30 6.07 +41%
D. IBM e-Server xSeries 1266/1024/40/512 6.54 8.68 +33%
E. Dell Optiplex GX-150 1200/512/40/256 6.05 7.81 +29%
F. Dell Optiplex GX-240 1700/256/20/256 7.29 7.16 -2%
G. Clone Soltek nForce2 2200/512/20/512 11.96 17.61 +47%
PS:
If you compare E and F you will see why I love Tualatins and hate early P4s :).
Machine G shows just how good Athlon XPs were in their day as well.
I don't know that L2 cache size is all that significant here.
Cheers,
Gary.
RE: So, I'm an Intel man,
)
Wasn't trying to knock you...just adding clarity. :-)
WD you say? Ever get the ole Click-o-Death?
BTW, what's a RaptorX go for across the pond?
RE: RE: So, I'm an Intel
)
We have not had many HDD failures, but we don't do much database type work, which I believe, but maybe wrong as usual, is where the problem lies. Plus compared to number of cpu's and memory chips we use our HDD count is very small.
Raptor X, which we don't use, is listed at £126 less taxes. I could get but must order full box of 10 min. There are discounts for bulk purchases, and I think the company gets about 8% off for ordering in 100's.
And we only use Enterprise class, not the slightly cheaper mainstream ones.
Andy
RE: We have not had many
)
I built about a thousand (yes, thousand) machines over the course of about a year when I was working for a small OEM back in 1998. This was consumer grade, not enterprise... At any rate, we had a mix of Fujitsu, WD, Maxtor, and Seagate. The largest number of failures happened with the WD and Seagate drives. The Fujitsu drives were just painfully slow. Maxtor was the faster and most reliable. There were 3 other builders (some were faster than me) so somewhere between 4500 and 5000 installed drives, perhaps 5500ish ordered drives. Not a real big run, but enough to get a general feel of the performance and reliability...
Remembering all of that, I had serious reservations about getting the Seagate that's in my system. The first one that arrived was defective. It caused the system to hang at boot, but with a misleading indicator on my LED diagnostic board that goes with my motherboard. It was pointing at FDD failure. The only way I figured it out, since I didn't have access to multiple drives like I did back then, was the fact that SATA is hot pluggable. Boot to a floppy and while the floppy is first loading, plug the cable in. Ran Seagate Tools and it said it was bad. I did an RMA with Newegg and got the second one and it has worked like a champ...
Brian
To prove the point that it's
)
To prove the point that it's cache-related on Intels, I'll try to negotiate with my brother... he's got one of those really nice late Pent Ms with an enormous 2 MB of L2 cache (it's called a Dothan iirc), a likely machine to get away totally unaffected if we are correct... doubt he'll let me run it off a live CD, though... anyone else got a machine like this?