I'll try to add to this table for reference in due course.
TL.
Duron (Spitfire Core)
550 1.5v--21.1w
600 1.6v--27.4w
650 1.6v--29.4w
700 1.6v--31.4w
750 1.6v--33.4w
800 1.6v--35.4w
850 1.6v--37.4w
900 1.6v--39.5w
950 1.6v--41.5w
Duron (Morgan Core)
900 1.75v--42.7w
950 1.75v--44.4w
1.0G 1.75v--46.1w
1.1G 1.75v--50.3w
AMD T-Bird Family:
Athlon 1200------1.20 GHZ-----65.7 watts
Athlon 1300------1.30 GHZ-----68.3 watts
Athlon 1333------1.33 GHz-----69.8 watts
Athlon 1400------1.40 GHz-----72.2 watts
AMD MP Family:
AthlonMP 1000--1.00 GHz-----46.1 watts
AthlonMP 1200--1.20 GHz-----54.7 watts
AMD XP Family:
AthlonXP 1500--1.33 GHz-----60.0 watts
AthlonXP 1600--1.40 GHz-----62.8 watts
AthlonXP 1700--1.47 GHz-----64.0 watts
AthlonXP 1800--1.53 GHz-----66.0 watts
Intel P4 Socket 478 P4 Socket 423 (1.75v)
1.50 Ghz........57.9W (478) 1.50Ghz.........57.8W(423)
1.60 Ghz........60.8W (478) 1.60Ghz.........61.0W(423)
1.70 Ghz........63.5W (478) 1.70Ghz.........64.0W(423)
1.80 Ghz........66.1W (478) 1.80Ghz.........66.7W(423)
1.90 Ghz........72.8W (478) 190Ghz.........69.2W(423)
2.00 Ghz........75.3W (478) 2.00Ghz.........71.8W(423)
Hmmm, I seem to have forgotten about this table as it is now _way_ out of date :lol:
TL.
QuoteOriginally posted by TeaLeaf@Oct 30 2004, 03:13 PM
Hmmm, I seem to have forgotten about this table as it is now _way_ out of date :lol:
TL.
[post=68272]Quoted post[/post]
Maybe I can be of some assistance.....
AMD & Intel CPUs': Max Voltage!Intel Celeron® Processor up to 1.10 GHz
Section 2.9 - Table 4 states:
Max Voltage: [Typical Operating voltage from Table 5] V + 0.1 V = [MaxVCore] V
Intel Celeron Processor for the PGA370 Socket up to 1.40 GHz on 0.13 Micron Process
Section 2.10 - Table 6 states:
Max Voltage: 1.75 V
Intel® Celeron Processor in the 478-Pin Package at 1.80 GHz
Section 2.9 - Table 5 states:
Max Voltage: 2.10 V
Intel Celeron Processor on 0.13 Micron Process in the 478-Pin Package
Section 2.10 - Table 6 states:
Max Voltage: 1.75 V
Intel Celeron D Processors 335, 330, 325, and 320
Section 2.10 - Table 2-7 states:
Max Voltage: 1.55 V
Coppermine Documentation (500 MHz to 1.13 GHz)
Section 2.9 - Table 6 states:
Max Voltage: 2.10 V
Intel Pentium 4 Processor in the 423-pin Package at 1.30, 1.40, 1.50, 1.60, 1.70, 1.80, 1.90 and 2 GHz (Willamette Socket 423)
Section 2.9 - Table 4 states:
Max Voltage: 2.10 V
Intel Pentium 4 Processor in the 478-pin Package at 1.40 GHz, 1.50 GHz, 1.60 GHz, 1.70 GHz, 1.80 GHz, 1.90 GHz, and 2 GHz (Willamette Socket 478)
Section 2.10 - Table 5 states:
Max Voltage: 2.10 V
Intel Pentium 4 Processor with 512-KB L2 Cache on 0.13 Micron Process and Intel Pentium 4 Processor Extreme Edition Supporting Hyper-Threading Technology Documentation (Northwood)
Section 2.10 - Table 2-5 states:
Max Voltage: 1.75 V
Intel Pentium 4 Processor on 90 nm Process Documentation (Prescott)
Section 2.9 - Table 8 states:
Max Voltage: 1.55 V
Mobile Intel Pentium 4 Documentation
Section 2.10 - Table 6 states:
Max Voltage: 1.75 V
Mobile Intel Pentium M Documentation (Centrino)
Section 3.9 - Table 4 states:
Max Voltage: 1.75 V
Intel Pentium M Processor on 90nm Process with 2-MB L2 Cache Documentation
Section 3.8 - Table 4 states:
Max Voltage: 1.55 V
Intel Pentium 4 Processor 560, 550, 540, 530 and 520
Section 2.10 - Table 2-7 states:
Max Voltage: 1.55 V
Intel Pentium 4 Processor Extreme Edition on 0.13 Micron Process in the 775-Land Package
Section 2.10 - Table 2-4 states:
Max Voltage: 1.55 V
AMD
AMD Duron 600 MHz to 950 MHz
Section 7.9 - Table 8 States:
Max Voltage: 1.7 V
AMD Duron 900 MHz to 1300 MHz
Section 7.8 - Table 7 States:
Max Voltage: 1.75 V + 0.5 V = 1.8 V
AMD Duron 1400 MHz to 1800 Mhz
Section 7.8 - Table 7 States:
Max Voltage: 1.5 V + 0.5 V = 1.55 V
AMD Athlon XP 1500+ to 2100+
Section 6 - Table 1 States:
Max Voltage: 1.75 V + 0.5 V = 2.25 V
AMD Athlon XP With 266 FSB (1700+ to 2200+)
Section 8.8 - Table 13 States
Max Voltage: 1.65 V + 0.5 V = 2.15 V
AMD Athlon XP With 333 FSB (2500+ to 3000+)
Section 8.8 - Table 14 States:
Max Voltage: 1.65 V + 0.5 V = 2.15 V
AMD Athlon XP With 400 FSB (3000+ to 3200+)
Section 8.8 - Table 14 States:
Max Voltage: 1.65 V + 0.5 V = 2.15 V
Hope it helps OCers!
Source: http://www.driverheaven.net/showthread.php?t=16750 (http://www.driverheaven.net/showthread.php?t=16750)
....and some help with Athlon 64Bit setups
You can find the 'core code' (sometimes refered to as the 'stepping code', though technically it's not a stepping in the true sense of the word, but can indicate the stepping) located on the heatspreader or underneith the heatspreader on a black label at the edge of processor. Should look something like:
ADA3400AEP5AP
AAALC0341MPMW
F447141J30044
Here is a brief guide to understanding these 3 lines of code on AMD processors:
ADA3400AE5AP - 1st line
ADA ----- ADA = AMD A64 Desktop, AMA = AMD A64 Mobile DTR, AMN = AMD A64 Mobile
3400 --â€" model number or performance rating (PR)
A â€"------- package type (A =754pin, B=754pin Unlidded, C=940pin, D=939pin)
E â€"------- default voltage (C=1.55v, E=1.50V, I=1.40v, Q=1.2v)
P â€"------- max case (heatspreader?) (I=63oC, O=69oC, P=70oC, X=95oC)
5 â€"------- L2 cache (3=256KB, 4=512KB, 5=1MB)
AP â€"----- core revision (AK=C0, AP=C0, AR=CG, AT=CG, AX=CG)
AAALC0341MPMW - 2nd line
AAALC is the 'core code'. In the past comparing when comparing 'core codes' that have the same starting letter (ie A), then the higher up the alphabet the other letters are, the better it is. Thus AAALC should be better than AAAHC.
0341, the 4 numbers after the core code refer to the date of manufacture. The first 2 numbers refer to the year and the second 2 numbers refer to the week. In the above case 03=2003 and 41=week 41. Generally the newer the processor the better it should be. So you might find a week 41 AAALC that overclocks better than a week 34 one. Though some weeks tend to be better than others, so it's not always true.
MPMW, supposedly one user claims that in the 'WPMW' code, the letter M (ie the 3rd letter) refers to the batch or wafer. Supposedly the lower this letter, the better. Another user agreed and cited that they had seen and XPAW do very well, XPMW do averagely well and TPXW do worse than average. Some cited a AMD T-Bred B datasheet as pointing this out. Even saying that if it's A, B, C they are early batches of the week and good. But if they are an 'M' it could refer to combined lots from many batches that needed to be rated lower. Plus others like R, I, S, and even digits like 1 to 9, with it being suggested the numbers are worse lots than the letters. However, another user also seems to think this 3rd letter is somehow related to the to default vcore or threshold voltage required to run at stock speeds and that the higher the letter, the better.
F447141J30044 - 3rd line
The third line has been suggested as stating the production line (first character, ie F) and some kind of batch and serial number identifier (ie 447141J30044). Certainly the first character of the third line in the past with Athlons, has been used suggest good overclocking cores. Ones with letters (Y, K, etc) would seem to overclock a bit better than ones with numbers (9, 7, etc). Y's used to be the best overall, though this was only noted by studying T-Bird Athlons. No study has been done to see if this holds true with AthlonXPs, so most only go by 'core code' and date of manufacture now.
Noticed a comment about some seem to think the last 4 digits (ie 0985) refer to the cores number in that batch and that a lower number is better (one user claims anything 0150 or below is very overclockable). Maybe could be related to the location of the core on the silicon wafer, lower number means closer to the centre. In theory, cores made from the centre of the silicon wafer are of better quality than cores made on the outer edge. However, no proof to back up such theory as yet. Infact one user claimed with their two same stepping AthlonXPs, the 0112 one overclocked better than the 0014 one.
The core type and revision/stepping can be identified by the following 2 letter code at the end of the OPN (ie ADA3400AEP5AP) and by CPUID. A current overview:
A64 Desktop
AP -----: S754 130nm Clawhammer (Rev SH7-C0, CPUID F48h, 512K/1MB)
AR -----: S754 130nm Clawhammer (Rev SH7-CG, CPUID F4Ah, 512K/1MB)
AX -----: S754 130nm Newcastle (Rev DH7-CG, CPUID FC0h, 512K)
AS -----: S939 130nm Sledgehammer (Rev SH7-CG, CPUID F58h, 1MB)
AW ----: S939 130nm Newcastle (Rev DH7-CG, CPUID FC0h, 512K)
BI ------: S939 90nm Winchester (Rev DH8-D0, CPUID ????, 512K)
A64 Mobile
AP -----: S754 130nm Clawhammer (Rev SH7-C0, CPUID F48h, 1MB)
AR -----: S754 130nm Clawhammer (Rev SH7-CG, CPUID F4Ah, 1MB)
AX -----: S754 130nm Newcastle (Rev DH7-CG, CPUID FC0h, 512K)
?? ------: S754 90nm Oakville (Rev ???-???, CPUID ????, ????)
A64 FX
AK -----: S940 130nm Sledgehammer (Rev SH7-C0, CPUID F58h, 1MB)
AT -----: S940 130nm Sledgehammer (Rev SH7-CG, CPUID F5Ah, 1MB)
AS -----: S939 130nm Sledgehammer (Rev SH7-CG, CPUID F58h, 1MB)
The Clawhammer C0 revision based on 0.13-micron process. Voltage is 1.50v for desktop and 1.40v for mobiles.
A64 Desktop S754 ---: AP (130nm Clawhammer, SH7-C0 revision, CPUID F48h)
A64 Mobile S754 ------: AP (130nm Clawhammer, SH7-C0 revision, CPUID F48h)
Thus:
A64 2800+ -----------: ADA2800AEP4AP (S754, 1.8GHz, 512k, 1.50v, 130nm Clawhammer, SH7-C0, F48h) ?
A64 3000+ -----------: ADA3000AEP4AP (S754, 2.0GHz, 512k, 1.50v, 130nm Clawhammer, SH7-C0, F48h)
A64 3200+ -----------: ADA3200AEP5AP (S754, 2.0GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-C0, F48h)
A64 3400+ -----------: ADA3400AEP5AP (S754, 2.2GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-C0, F48h)
A64-M 3000+ DTR ---: AMA3000BEX5AP (S754, 1.8GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-C0, F48h)
A64-M 3200+ DTR ---: AMA3200BEX5AP (S754, 2.0GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-C0, F48h)
A64-M 3400+ DTR ---: AMA3400BEX5AP (S754, 2.2GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-C0, F48h)
A64-M 2800+ ---------: AMN2800BIX5AP (S754, 1.6GHz, 1MB, 1.40v, 130nm Clawhammer, SH7-C0, F48h)
A64-M 3000+ ---------: AMN3000BIX5AP (S754, 1.8GHz, 1MB, 1.40v, 130nm Clawhammer, SH7-C0, F48h)
A64-M 3200+ ---------: AMN3200BIX5AP (S754, 2.0GHz, 1MB, 1.40v, 130nm Clawhammer, SH7-C0, F48h)
Some overclocking results spotted around the net (some claim the Brown ones overclock better than Green ones, but unsure):
AAALC 0337MPM @ 2621MHz w/ 1.7v (Mach I)
CAAMC 0339WPMW 9446548130306 @ 2500MHz (aircooled)
CAAMC 0339XPMW Brown @ 2540MHz w/1 95v (Chilled watercooled)
CAAMC 0342 @ 2655mhz (up to 2.0v?) (chilled watercooled, -20oC)
CAAAC 0345RPMW @ 2440MHz w/ 1.7v (watercooled)
CAAAC 0345 @ 2.6GHz w/ 1.8v (Mach I)
CAAMC 0349 RPMW @ 2400MHz w/ 1.6-2.0v (unstable)
CAAOC 0350 XPMW @ 2600MHz w/ 1.85v (watercooled)
A64 3000+ 'AP' CAATC 0350XPMW @ 2508MHz w/ 1.65v
A64 3000+ 'AP'? CAAOC 0352 @ 2704MHz w/ 1.92v
A64 3000+ 'AP'? CAAOC 0352 XPMW Brown 2.5GHz w/ 1.85v
A64 3200+ 'AP' @ 2.8GHz w/ 1.9v (dry ice)
A64 3200+ 'AP'? CAAOC @ 2.5GHz w/ 1.66v (stock aircooled)
A64 3200+ 'AP'? @ 2830MHz w/ 1.9v (triple cascade, -46oC)
A64 3200+ 'AP'? @ 2608MHz w/ 1.68 (Mach I)
A64 3200+ 'AP'? @ 2756MHz w/ 1.8v (Mach II)
A64 3200+ 'AP' CAACK 0330UPWM 9438245200435 @ 2380MHz w/ 1.7v (stock aircooled), 2500mhz (Mach I)
A64 3200+ 'AP' CAAKC 0334TPMW 9441535240131 @ 2400MHz w/ 1.64v, 2450MHz w/ 1.71v (stock aircooled)
A64 3200+ 'AP' CAAMC 0339WPMW 9446635J30492 @ 2.5GHz w/ 1.7v (watercooled)
A64 3200+ 'AP'? CAAMC 0339WPMW 9 0072 Green @ 2.36GHz w/ 1.7v (stock aircooled)
A64 3200+ 'AP'? CAAMC 0339WPMW 9 Green @ 2609MHz w/ 1.77v
A64 3200+ 'AP'? 0341 Brown @ 2767MHz w/ 1.88v (Mach I)
A64 3200+ 'AP' CAAAC 0345SPMW @ 2750MHz w/ 1.904v (Mach I)
A64 3200+ 'AP' CAAMC 0347VPMW 9359625k30176 @ 2349MHz w/ 1.65v (watercooled)
A64 3200+ 'AP' CAAOC 0351 9 0160 @ 2.7GHz w/ 1.9v (Vapochill PE)
A64 3400+ 'AP'? @ 3.0GHz (Mach I w/ R404a) - in 3DMark 400-500 points behind FX-51 (@ 3.0GHz?)
A64 3400+ 'AP'? @ 3.1GHz w/ 2.04v (Cascade)
A64 3400+ 'AP'? @ 2.75GHz w/ 1.7v
A64 3400+ 'AP' CAAOC 0350XPMW 9832334L30188 @ ~2.5GHz w/ 1.7v (stock aircooled)
A64 3400+ 'AP' CAAOC 0351XMPW @ 2.8GHz (Mach II)
A64 3400+ 'AP' CAATC 0351XPMW @ 2784MHz w/ 1.7v
A64 3400+ 'AP' CAAOC 0352TPMW @ 2821MHz w/ 1.7v, 2929MHz w/ 1.8v (Mach II w/ r404a)
A64 3400+ 'AP' CAAOC 0401SPMW 9839644A40514 @ 2508MHz w/ 1.7 v (stock aircooled)
A64-M 3400+ DTR 'AP'? @ 2920MHz w/ 1.85v (Mach II)
Some commented they have seen quite a few CAAOC 0352XPMW all do at least 2.4GHz on aircooling. Plus between week 0351-0352 supposedly are good cores, along with XPMW ones. They cite their A64 3400+ XPMW doing 2.6GHz on aircooling.
The Sledgehammer C0 revision based on 0.13-micron process, same as used by Opteron 1xx. Voltage is 1.50v.
A64 FX S940 ---------: AK (130nm Sledgehammer, SH7-C0 revision, CPUID F58h)
Thus:
A64 FX-51 -------------: ADAFX51CEP5AK (S940, 2.2GHz, 1MB, 1.50v, 130nm Sledgehammer, SH7-C0, F58h)
Overclocking results:
A64 FX-51 'AK' @ 3009MHz w/ 1.728v (Chilled Ethanol, -65oC)
The new Clawhammer CG revision, has a few tweaks including energy saving, memory compatibility and performance. Still 1.50v Vcore for desktops and 1.40v for mobiles. Infact there are actually 3 CG revisions SH7-CG, DH7-CG and CH7-CG.
A64 Desktop S754 ---: AR (130nm Clawhammer, SH7-CG revision, CPUID F4Ah)
A64 Mobile S754 -----: AR (130nm Clawhammer, SH7-CG revision, CPUID F4Ah)
Thus:
A64 3000+ -----------: ADA3000AEP4AR (S754, 2.0GHz, 512K, 1.50v, 130nm Clawhammer, SH7-CG, F4Ah)
A64 3200+ -----------: ADA3200AEP5AR (S754, 2.0GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-CG, F4Ah)
A64 3400+ -----------: ADA3400AEP5AR (S754, 2.2GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-CG, F4Ah)
A64 3700+ -----------: ADA3700AEP5AR (S754, 2.4GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-CG, F4Ah)
A64-M 3000+ DTR ---: AMA3000BEX5AR (S754, 1.8GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-CG, F4Ah)
A64-M 3200+ DTR ---: AMA3200BEX5AR (S754, 2.0GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-CG, F4Ah)
A64-M 3400+ DTR ---: AMA3400BEX5AR (S754, 2.2GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-CG, F4Ah)
A64-M 3700+ DTR ---: AMA3700BEX5AR (S754, 2.4GHz, 1MB, 1.50v, 130nm Clawhammer, SH7-CG, F4Ah)
A64-M 2800+ ---------: AMN2800BIX5AR (S754, 1.6GHz, 1MB, 1.40v, 130nm Clawhammer, SH7-CG, F4Ah)
A64-M 3000+ ---------: AMN3000BIX5AR (S754, 1.8GHz, 1MB, 1.40v, 130nm Clawhammer, SH7-CG, F4Ah)
A64-M 3200+ ---------: AMN3200BIX5AR (S754, 2.0GHz, 1MB, 1.40v, 130nm Clawhammer, SH7-CG, F4Ah)
Cool'n'Quiet feature has changed with the CG revision. So instead of the previous powering saving state of 800MHz and 1.30v (35W TDP), the CG stepping can manage 1.0GHz and 1.10v (22W TDP). The A64 3400+ also gets an extra power saving state too, 1.8GHz and 1.30v. There is also reduced requirements for AMD64 interface logic energy consumption.
The CG revision is also meant to bring better memory performance via a "2T memory timing" that allows for higher density or frequency memory. It should also improve memory compatibility by no longer requiring identically paired modules for the 2nd and 3rd slots with 3 dimm slots.
However, some claim the CG revision is slightly slower than the C0 revision. Though, there are some C0 vs CG benchmarks seeing how different memory config and timings effect things. With 1 module no real difference. 3 mixed modules the CG works, the C0 doesn't. With 3 identical modules C0 doesn't work with 2.5-3-3-7 but the CG did. At CL3-3-3-7 the CG is clearly a bit slower than the C0, but not much. Comparing CG 2.5-3-3-7 vs C0 CL3-3-3-7, you find some benchmarks (inc 3DMark) the CG is slightly faster and some the C0 is slightly faster, though both end up with same SuperPI 1M time of 44s.
Some say the CG stepping tends to overclock to 2.4-2.5GHz on aircooling and that 1MB L2 is around 5% faster than only having 512K L2. As for overclocking results:
A64 3000+ 'AR' SH7-CG CAATC 0350XPMW @ 3.0GHz w/ 1.968v (r22 phase change)
A64 3000+ 'AR' SH7-CG CAATC 0404XPMW 0256 @ 2655 w/ 1.9v (watercooled)
A64 3200+ 'AR' SH7-CG CAAPC 0408TPMW 9561925C40400 @ 2.5Ghz w/ 1.7v (watercooled)
A64 3200+ 'AR' CAAPC 0405TPMW @ 2.53Ghz w/ 1.71v (aircooled)
A64 3200+ 'AR' CAAPC 0348TPAW F441225L30104 @ 2.6Ghz w/ 1.75v (watercooled)
A64 3200+ 'AR' CAABC 0416SMPW 9543755D40417 @ 2350Mhz w/ 1.65v (aircooled)
A64 3200+ 'AR' AAAPC 0346XPMW 5837680L30083 @ 2800Mhz w/ 1.8v (VapoLS)
A64-M 3200+ 'AR' CAABC 0416MPMW @ 2525MHz w/ 1.7v (aircooled)
A64-M 3200+ DTR 'AR' SH7-CG @ 2508MHz w/ 1.74v (1.7v in BIOS) (aircooled)
A64 3400+ 'AR' SH7-CG CAAPC 0408TPWM 9567014C40733 @ 2.7GHz w/ 1.7v (Mach I)
A64 3700+ 'AR' CAACC 0420UPMW 9772012E40064 @ 2944MHz w/ 1.8v (R404 phase change)
The new Newcastle CG revision is meant to be proper 512K core, not just a 1MB Clawhammer core with half the L2 cache disabled like previous 512K Athlon64s. Assume it has all the CG tweaks as per the Clawhammers.
A64 Desktop S754 ---: AX (130nm Newcastle, DH7-CG revision, CPUID FC0h)
A64 Mobile S754 -----: AX (130nm Newcastle, DH7-CG revision, CPUID FC0h)
A64 Desktop S939 ---: AW (130nm Newcastle, DH7-CG revision, CPUID FF0h)
Thus:
A64 2800+ -----------: ADA2800AEP4AX (S754, 1.8GHz, 512K, 1.50v, 130nm Newcastle, DH7-CG, FC0h)
A64 3000+ -----------: ADA3000AEP4AX (S754, 2.0GHz, 512K, 1.50v, 130nm Newcastle, DH7-CG, FC0h)
A64 3200+ -----------: ADA3200AEP4AX (S754, 2.2GHz, 512K, 1.50v, 130nm Newcastle, DH7-CG, FC0h)
A64 3400+ -----------: ADA3400AEP4AX (S754, 2.4GHz, 512K, 1.50v, 130nm Newcastle, DH7-CG, FC0h)
A64-M 2700+ LV -----: AMD2700BQX4AX (S754, 1.6GHz, 512K, 1.20v, 130nm Newcastle, DH7-CG, FC0h)
A64-M 2800+ LV -----: AMD2800BQX4AX (S754, 1.8GHz, 512K, 1.20v, 130nm Newcastle, DH7-CG, FC0h)
A64 3500+ -----------: ADA3500DEP4AW (S939, 2.2GHz, 512K, 1.50v, 130nm Newcastle, DH7-CG, FF0h)
A64 3800+ -----------: ADA3800DEP4AW (S939, 2.4GHz, 512K, 1.50v, 130nm Newcastle, DH7-CG, FF0h)
Overclocking results:
A64 2800+ 'AX' CBAVC 0420TPMW 9764412E40176 @ 2451MHz w/ 1.55v (aircooled), 2608MHz w/ 1.65v (aircooled)
A64 2800+ 'AX' CBAUC 0420TPMW 9764112E40515 @ 2451MHz w/ 1.52-1.57v (aircooled)
A64 2800+ 'AX' CBASC 0419UPMW 9718925E40028 @ 2565MHz w/ 1.65v (aircooled)
A64 3000+ 'AX' DH7-CG CBAPC 0404VPFW 9847235B40119 @ 2.4GHz w/ 1.68v (1.65v in BIOS) (stock aircooled)
A64 3000+ 'AX' DH7-CG CBASC 0406UPMW @ 2.6GHz w/ 1.67v
A64 3000+ 'AX' CBASC 0416VPMW @ 2565MHz w/ 1.7v (watercooled)
A64 3000+ 'AX' CBASC 0416RPMW @ 2555MHz w/ 1.55v (watercooled)
A64 3000+ 'AX' CBASC 0404XPMW 9833965C40262 @ 2400MHz w/ 1.7v (aircooled)
A64 3200+ 'AX' CBAVC 0420TPMW 9764224E40104 @ 2500MHz w/ 1.7v (aircooled)
A64 3200+ 'AX' CBAEC 0423TPMW 1044734F40126 @ 2400MHz w/ 1.7v (aircooled)
A64 3200+ 'AX' CBASC 0415XPMW @ 2618MHz w/ 1.65v (watercooled)
A64 3200+ 'AX' CBAEC @ 2500MHz w/ 1.55v (aircooled)
A64 3200+ 'AX' CBASC 0407MPMW @ 2413MHz w/ 1.55v (aircooled)
A64 3200+ 'AX' ?????? @ 2.7GHz w/ 1.65v (watercooled), 2.8GHz w/ 1.75v (IHS removed, watercooled)
A64 3200+ 'AX' CBASE 0416 @ 2693MHz w/ 1.7v (aircooled)
A64 3500+ 'AW' @ 2.55GHz w/ 1.65v (aircooled)
A64 3500+ 'AW' CBAEC 0421MPMW 1035634F40441 @ 2530MHz w/ 1.62 (stock aircooled), 2900MHz w/ 1.62v (VapoLS)
The new Sledgehammer CG revision, assume it has the same tweaks as the Clawhammer's CG stepping mentioned above.
A64 FX S940 -----------: AT (130nm Sledgehammer, SH7-CG, CPUID F5Ah)
A64 Desktop S939 ----: AS (130nm Sledgehammer, SH7-CG, CPUID F7Ah)
A64 FX S939 -----------: AS (130nm Sledgehammer, SH7-CG, CPUID F7Ah)
Thus:
A64 FX-51 -------------: ADAFX51CEP5AT (S940, 2.2GHz, 1MB, 1.50v, 130nm Sledgehammer, SH7-CG, F5Ah)
A64 FX-53 -------------: ADAFX53CEP5AT (S940, 2.4GHz, 1MB, 1.50v, 130nm Sledgehammer, SH7-CG, F5Ah)
A64 4000+ ------------: ADA4000DEP5AS (S939, 2.4GHz, 1MB, 1.50v, 130nm Sledgehammer, SH7-CG, F7Ah)
A64 FX-53 -------------: ADAFX53DEP5AS (S939, 2.4GHz, 1MB, 1.50v, 130nm Sledgehammer, SH7-CG, F7Ah)
A64 FX-55 -------------: ADAFX55DEI5AS (S939, 2.6GHz, 1MB, 1.50v, 130nm Sledgehammer, SH7-CG, F7Ah)
The A64 4000+ is effectively a rebranded FX-53, but upper multiplier locked like all non-FX processors.
The FX-55 is the first A64 processor to use streched 'strained' SOI (Silicon on Insulator) process, while previous A64 processors have used the normal SOI. Although some suggest by Q4/2004 all new batches of 130nm and 90nm will probably also use streched 'strained' SOI. Hopefully should help reach higher clock speeds. In future AMD could also introduce the better compressed 'strained' silicon that Intel has had success with.
Also the FX-55 has a TCASE Max (heatspreader?) temp of only 63oC, compared to the previous 70oC.
Overclocking results:
A64 FX-53 'AT' SH7-CG @ 3423MHz w/ 2.096v (Triple Cascade, -112oC)
A64 FX-53 'AT' SH7-CG CAABC 0410MPMW @ 3404MHz w/ 1.987v (LN2)
A64 FX-53 'AT' SH7-CG @ 3201MHz w/ 1.45v, 3602MHz w/ 1.95v (Cascade -95oC)
When it comes to overclocking the A64/FX processors, it's important to realise things operate a little differently.
(1) Firstly, only the A64 FX processors have fully unlocked multipliers. The normal A64 processors (and A64-Ms?) are 'multiplier upwards locked', in that you can select the default multiplier and any value below, but no higher than the default. As yet, no way to unlock the upper multipliers.
(2) Secondly, the A64/FX processors no longer use a single FSB system bus. Normally the FSB system bus is used to connect the CPU to the memory and the rest of the system via northbridge chipset. However, the memory controller has been removed from the northbridge and intergrated into the A64/FX processors. So the A64/FX has replaced the single FSB with two seperate buses:
Memory bus
HyperTransport (HT) system link
The 'memory bus' connects the on-board memory controller directly to the system memory, which is of course currently DDR memory. While the 'HyperTransport (HT) system link' connects the CPU to the northbridge and rest of the system.
HyperTransport (formerly LDT, Lightning Data Transport) is a high-speed, packet based 'point-to-point' interconnection technology developed by AMD. It has been designed with features such as high speed, low latency and simple design (few wires). It is DDR based, so effectively operates at twice it's actual clock speed. Plus it's also bi-directional (full-duplex), which allows it to send data in both directions (uplink and downlink) at the same time, effectively doubling the bandwidth.
(3) Because of all this, the clock speeds of the system (processor, HT bus, memory, pci/agp, etc) are derived slightly differently. The A64/FXs still uses a PLL clock generator to drive the northbridge and it provides the reference frequency (clock speed) for the HyperTransport (HT) system link and CPU clock speed. This reference clock speed is commonly refered to as the 'HTT' and replaces the traditional FSB:
CPU Clock Speed (MHz) ---------------- = HTT (MHz) x CPU Multiplier
Memory bus ------------------------------ = CPU Clock Speed (MHz) / Memory Controller Divider
HyperTransport (HT) Link (MHz) ------ = HTT (MHz) x LDT Multiplier
Current A64/FX processors have a HTT of 200MHz and are designed with an 800MHz HT (4x LDT) in mind. AMD plans to move to 250MHz HTT and 1000MHz HT with the upcoming S939 versions.
As you can see the memory bus is now derived from the CPU speed, not the HTT (aka FSB). This is because the memory controller is intergrated into the processor, thus runs at the CPU speed (memory requests only) instead of the northbridge speed (ie HTT). Therefore to get the correct 'memory bus' clock speed for current DDR memory, the memory controller uses a set of dividers.
Supposedly in the BIOS you'll get memory speed or mem/cpu ratio options that do relate to the HTT clock (ie 200MHz), like 200 (1:1), 166 (5:6), 133 (2:3), 100 (1:2), etc. These are not memory locks though. When you select one, the memory controller will see this as the 'target memory speed'. It then derives the actual memory bus speed by taking the CPU Clock Speed and selecting a 'Memory Controller Divider' value that results in a speed as close as possible to the required 'target memory speed'. Depending on the values involved, sometimes the resulting Memory Bus is spot on, other times it's slightly out. Supposedly a table stating what the Memory Controller Divider will be for 200MHz HTT processors is as follows:
CPU Multiplier ----------------------- DRAM Speed setting -------------------------
----------------------- 200MHz --- 166MHz --- 150MHz --- 133MHz --- 100MHz-
16.0x ----------------- 16.0 -------- 20.0 ------- 22.0 ------- 24.0 ------- 32.0 --
15.0x ----------------- 15.0 -------- 18.0 ------- 20.0 ------- 23.0 ------- 30.0 --
14.0x ----------------- 14.0 -------- 17.0 ------- 19.0 ------- 21.0 ------- 28.0 --
13.0x ----------------- 13.0 -------- 16.0 ------- 18.0 ------- 20.0 ------- 26.0 --
12.0x ----------------- 12.0 -------- 15.0 ------- 16.0 ------- 18.0 ------- 24.0 --
11.0x ----------------- 11.0 -------- 14.0 ------- 15.0 ------- 17.0 ------- 22.0 --
10.0x ----------------- 10.0 -------- 12.0 ------- 14.0 ------- 15.0 ------- 20.0 --
9.0x ------------------- 9.0 --------- 11.0 ------- 12.0 ------- 14.0 ------- 18.0 --
8.0x ------------------- 8.0 --------- 10.0 ------- 11.0 ------- 12.0 ------- 16.0 --
7.0x ------------------- 7.0 --------- 9.0 --------- 10.0 ------- 11.0 ------- 14.0 --
6.0x ------------------- 6.0 --------- 8.0 --------- 8.0 --------- 9.0 -------- 12.0 --
For example, if you overclock to 278MHz HTT x 9.0 (2502MHz) and use 166 (5:6) in the BIOS, the 'memory controller divider' will be 11.0. Thus the memory bus will actually be 2502MHz / 11.0 = 227.45MHz and not 166MHz. In contrast on a P4 with an FSB and northbridge memory controller, you'd of normally worked it out like (278MHz FSB / 6) x 5 = 231.66MHz.
(4) There have been concerns over A64 processors dying from high Vdimm voltages, since now the memory controller is on-board the processor core. Supposedly AMD suggest at most 2.9v Vdimm and 1.65v Vcore. So it's possible this issue is mainly related to boards with Vdimm Vmods. Some noted when using Vmods, leaving the BIOS Vdimm to AUTO and changing it via the Vmod is best. Otherwise changing with Vdimm in the BIOS with a Vmod in place can cause it to spike as high as 3.78v, which maybe what does the damage.
1MB vs 512K L2 cache
Supposedly 1MB L2 cache can be 2-10% (some say on average 5%) faster than 512K depending on the apps and assuming all things equal (same clock speed, same socket, etc). In games some suggest 1MB L2 can be 3-9% faster than 512K, again assuming all things equal. But it can really depend on the software.
A Doom3 CPU review showed only a 3.4% benefit of 1MB L2, when comparing same speed S939 dual-channel processors (A64 3400+ vs FX-51). Citing the on-board memory controller as probably helping 512K not be such a limit. Comparing another set of same speed S939 dual-channel processors (A64 3800+ vs FX-53), showed 3.48% (3.6fps) benefit of 1MB vs 512K.
Anandtech in another review noted a 3.37% (3.4fps) benefit of 1MB vs 512K in Doom3 (1024x768, High Quality) when comparing same speed S939 dual-channel processors (A64 3800+ vs A64 4000+). In Counterstrike: Source (1024x768, High Quality, they noted a 7.5% (14fps) benefit of 1MB vs 512K. So can depend on the game.
S939 (128-bit dual-channel) vs S754 (64-bit single-channel)
Supposedly S939 has 80% more effective memory bandwidth than S754. This provides better performance, especially in memory intensive apps, scientific calcs, etc and somewhere it's been suggested that can be as high as 20-80%. Supposedly in games, it's claimed S939 can be around 5% faster than S754 assuming all things equal (same clock speeds, same L2 cache size, etc). But again it can depend on the software.
A Doom3 CPU review using same speed 1MB L2 cache processors (A64 3400+ vs FX-51), showed there was only a 2.94% (2.9fps) benefit of S939 (dual-channel) at 600x800. While comparing same speed processors (A64 3400+ vs A64 3500+) showed a 0.83% (0.8fps) benefit of S754/1MB over S939/512K.
Anandtech compared same speed 512K L2 processors, the A64 3400+ (S754, 2.4GHz/512K) vs A64 3800+ (S939, 2.4GHz/512K) and noted the following average benefits of S939 (dual-channel):
Business/General Use -------------: 5.4% (only tied in 1 benchmark)
Multitasking Content Creation ---: 3.2%
Video Creation/Photo Editing ----: 4.2%
A/V encoding ------------------------: 4.4% (4 out of 5 benchmarks)
Gaming -------------------------------: 6.3%
3D rendering ------------------------: 5.4%
Workstation -------------------------: 17%
They noted a 12.5% (21.5fps) benefit of S939 in Counterstrike: Source (1024x768, High Quality). While they noted a ~7% (6.8fps) benefit of the S939 in Doom3 with 1024x768 High Quality.
Interestingly in their previous Doom3 CPU review they only noticed 2.94% (2.9fps) benefit in Doom3 at a lower resolution, yet expected it to be even less at high resolutions. Not sure why the Doom3 discrepancy, but in the Doom3 CPU review they were comparing 1MB L2 processors not 512K. Plus the test systems were different. The above review used an nForce4 board w/ low latency memory and X800 XT, while the Doom3 CPU review used an nForce3 board w/ normal memory and 6800 Ultra.
The new Winchester D0 revision (aka Rev D) is a 90nm (0.09-micron) core. Default voltage is now 1.4v and due to the smaller fab process hopefully a bit cheaper than previous 130nm S939 processors. Supposedly no changes to the core otherwise, so performance should be on a par with same spec 130nm ones.
A64 Desktop S939 ---: BI (90nm Winchester, Rev DH8-D0, CPUID ????)
Thus:
A64 3000+ -----------: ADA3000DIK4BI (S939, 1.8GHz, 512K, 1.40v, 90nm Winchester, DH8-D0, ????)
A64 3200+ -----------: ADA3200DIK4BI (S939, 2.0GHz, 512K, 1.40v, 90nm Winchester, DH8-D0, ????)
A64 3500+ -----------: ADA3500DIK4BI (S939, 2.2GHz, 512K, 1.40v, 90nm Winchester, DH8-D0, ????)
A japanese site claimed Winchester 3500+ was just under 5oC hotter at load than Newcastle, yet a Czech site claimed Winchester 3000+ was much cooler than Newcastle. Sudhian.com claimed a 2.0GHz Winchester was 2oC hotter at idle, but 1oC cooler at load. Techreport.com ran an A64 3800+ (2.4GHz) Winchester @ 3500+ (2.2GHz) and claimed it used 28W less than an equal 130nm 3500+. They also measured it's temp at load as 55oC, while the 130nm chip was 61oC.
AMD have been quoted as saying power dissipation is less for their 90nm process than 130nm. Supposedly AMD Thermal Data Sheet specs suggest current 90nm processors are rated up to 67W, compared to up to 89W for 130nm processors (up to 2.4GHz). Plus state 'TCASE Max' (max heatspreader temp?) of 65oC for 90nm (70oC for 130nm).
Anandtech is reporting the 90nm processors seem to perform 1-7% faster (3% avg in games, 1.5-2% in Doom3, 7% in Quake3) than the comparable 130nm ones. Which would seem to suggest the Winchester core contains some tweaks afterall. They also managed to overclock a 90nm A64 3000+ and 3500+ both to 2.6GHz on aircooling, though the 3000+ required slightly more voltage (1.52v?).
Techreport are also reporting the 90nm core seems to have slightly better performing L2 cache compared to 130nm. However, they only got a 0.4% benefit in Counterstrike: Source and UT2004. AMD would only tell them that some small optimizations have been made to the 90nm memory controller and how instructions execute.
They also tested system power consumption and the 90nm 3500+ at load was 15.7% less than the 130nm version. Infact the 90nm 3500+ at load was 37% less than a P4 550 (3.4GHz) at load. It was even 2.7% less at load than a P4 550 at idle! Maybe even more impressive when you consider they claimed a BIOS issue caused the A64s Vcore to be set too high for the tests, 1.6v instead of 1.4v.
Some suggest while the 130nm FX-55 is first to get streched 'strained' SOI, by Q4/2004 new batches of both 90nm and 130nm processors will probably also use it. Plus newer 90nm next year could also benefit from the better compressed 'strained' SOI.
Overclocking results:
A64 3200+ 'BI' ????? 0435 @ 2300MHz w/ default, 2550MHz w/ 1.60v (watercooled)
A64 3200+ 'BI' CBBFD 0435TPAW 1064627I40282 @ 2.5GHz w/ 1.60v (stock aircooled)
A64 3500+ 'BI' CBBFD 043?TPMW 10582355400??? @ 3003MHz w/ 1.48v (VapoLS
AFAIK the A64-Ms are upper multiplier locked, same as the desktop ones. So you'll be able to select the default and below values, but nothing higher. Only the A64-FX that is truely multiplier unlocked it seems. This does mean the slower A64-Ms will have a lower max multiplier, thus require a higher HTT/FSB speed to overclock fully.
Should also note, the A64-Ms lack the intergrated heatspreader (IHS) of the desktop processors, so have an exposed core. Maybe allows better cooling, but may cause problems with some heatsinks not making proper contact? Plus motherboards need a BIOS with supporting microcode, else supposedly they might default to only 800MHz (x4) or not even boot at all. Not sure which boards do have A64-M supp
Compiled by Nitestorm (Overclock.co.uk)
Scarey stuff!! :rolleyes:
.....and the new A64 Mobiles
More info:
--Mobile Athlon64--
*Socket 754 - Newcastle Core Low Voltage(1.2V)
AMD2700BQX4AX 1.6GHz
AMD2800BQX4AX 1.8GHz
Mobile Athlon64 Low Voltage(Newcastle) Example:
AM D 2800 B Q X 4 AX
AM = Brand: AMD Mobile Athlonâ,,¢64
D = Power Limit: 35W
2800 = Model Number: 1.8GHz
B = Package 754 Pin Lidless OµPGA(Organic Micro Pin Grid Array)
Q = Operating Voltage: 1.2V
X = Maximum Die Temperature: 95° Celsius
4 = L2 Cache 512KByte
AX = Newcastle Core Revision CG
*Socket 754 - Clawhammer Core DTR 81.5Watt(Desktop Replacement 1.5V)
AMA3000BEX5AP Athlon64 DTR 3000+ Review with MiTAC 8355 barebone
AMA3200BEX5AP Athlon64 DTR 3200+
AMA3400BEX5AP Athlon64 DTR 3400+
AMA3700BEX5AP Athlon64 DTR 3700+
Mobile Athlon64 DTR(Clawhammer) Example:
AM A 3200 B E X 5 AR
AM = Brand: AMD Mobile Athlonâ,,¢64
A = Power Limit: Desktop Replacement
3200 = Model Number: 2.0GHz
B = Package 754 Pin Lidless OµPGA(Organic Micro Pin Grid Array)
E = Operating Voltage: 1.5V
X = Maximum Die Temperature: 95° Celsius
5 = L2 Cache 1MByte (4=512KByte)
AR = Clawhammer Core Revision CG
*Socket 754 - Clawhammer Core 62Watt (1.4V)
AMN2800BIX5AR 1.6GHz
AMN3000BIX5AR 1.8GHz
AMN3200BIX5AR 2.0GHz
AMN3400BIX5AR 2.2GHz
Athlon64 Mobile Example
AMN 3200 B I X 5 AR
AM = Brand: AMD Athlonâ,,¢64 Mobile
N = Power Limit: 62W
3200 = Model Number: 2.0GHz
B = Package Type: 754 Pin Lidless OµPGA(Organic Mirco Pin Grid Array)
I = Operatring Voltage 1.4V
X = Maximum Die Temperature: 95° Celsius
5 = L2 Cache 1MByte
AR = Clawhammer Core Revision CG