ElectroSoldier wrote on 2024-02-06, 17:08:

I think the clue is in the title of the thread.

SkullTrail is powerful but its also power hungry.
It nearly doubles my power usuage when I turn it on. Cant believe I used it as long as I did without realising just how much power it was using.

150W TDP, per CPU socket + some GTX 295 (almost 300W each) cards in SLI seem like they would heat things up quite nicely.

As usual, 3DMark 2001 can answer many of life's difficult questions.

-Socket 478 sees a pretty impressive increase, going from 8238 points (Celeron-1700, 7800 GS) to 24372 (P4 3.4, 7800 GS). The best CPU for this socket is 2.96 times faster than the slowest one.
-Slot 1 scales from 2163 points (PII-233, 9800 Pro) to 8170 (PIII-1000EB, 9800 Pro). Difference = 3.77x
-Socket A shows impressive range, scoring 5119 at the bottom end (Duron 600, 6800 GT) and eventually ending up with 21448 (Athlon XP 3200+, 6800 GT). Difference = 4.19x
-The legendary Socket 775? It starts out at a leisurely 14706 (Celeron D 2.53GHz, GTX 680) and finishes at a dizzying 62450 (C2D 3.33GHz, GTX 680). Difference = 4.25x
-Finally, like a few of you guys were predicting, Socket 370 takes the cake, going from a plodding 2428 (Celeron 300A, 9800 Pro) to a quite nice 11496 (PIII-1400S, 9800 Pro). Difference = 4.73x

Socket AM4 is probably saying "hold my beer" right about now, but we can't consider it a vintage socket yet since the 5700X3D launched five freakin' days ago. But going from Excavator to Zen 3? Yeah, I'm sure there's at least a 5x single-threaded boost there, and that's not even taking into account the doubling of AVX2 throughput that Zen 2 added along the way. But hey, I only benchmark with 3DMark2001, so none of that fancy instruction-dense AVX stuff even matters. MMX and chill, baby.

Except you cant run a PIII-s 1400 in a PPGA socket 370 because its FC-PGA.
You can modify it of course, but then its not the same.

Standard Def Steve wrote on 2024-02-07, 01:31:

As usual, 3DMark 2001 can answer many of life's difficult questions. […]
Show full quote

As usual, 3DMark 2001 can answer many of life's difficult questions.

-Socket 478 sees a pretty impressive increase, going from 8238 points (Celeron-1700, 7800 GS) to 24372 (P4 3.4, 7800 GS). The best CPU for this socket is 2.96 times faster than the slowest one.
-Slot 1 scales from 2163 points (PII-233, 9800 Pro) to 8170 (PIII-1000EB, 9800 Pro). Difference = 3.77x
-Socket A shows impressive range, scoring 5119 at the bottom end (Duron 600, 6800 GT) and eventually ending up with 21448 (Athlon XP 3200+, 6800 GT). Difference = 4.19x
-The legendary Socket 775? It starts out at a leisurely 14706 (Celeron D 2.53GHz, GTX 680) and finishes at a dizzying 62450 (C2D 3.33GHz, GTX 680). Difference = 4.25x
-Finally, like a few of you guys were predicting, Socket 370 takes the cake, going from a plodding 2428 (Celeron 300A, 9800 Pro) to a quite nice 11496 (PIII-1400S, 9800 Pro). Difference = 4.73x

Socket AM4 is probably saying "hold my beer" right about now, but we can't consider it a vintage socket yet since the 5700X3D launched five freakin' days ago. But going from Excavator to Zen 3? Yeah, I'm sure there's at least a 5x single-threaded boost there, and that's not even taking into account the doubling of AVX2 throughput that Zen 2 added along the way. But hey, I only benchmark with 3DMark2001, so none of that fancy instruction-dense AVX stuff even matters. MMX and chill, baby.

Why did you stop at the E8600?, there exists the E8800 ES chips at 3.666Ghz and also the mighty X5470 which will happily clock out to 4.3 Ghz and supports dual GTX 295's. The fact the E8800 is an ES chip shouldnt exclude it from this discussion as it does physically exist and has been benched, and do modded 771 Xeons get excluded when 771 and 775 are the same socket just rotated 90 degrees.

If we exclude Xeons then the QX9770 will also clock out to 4Ghz and with dual GTX 580's should happily cruise past any score the E8600 could get.

Ive run a Titan Maxwell on a QX9770 ...so if we include that then Socket 775 is easily the socket with the largest power gains with no hardware modding, along with being one of the longest if not the longest lived Intel socket.

ElectroSoldier wrote on 2024-02-07, 01:49:

Except you cant run a PIII-s 1400 in a PPGA socket 370 because its FC-PGA.
You can modify it of course, but then its not the same.

But then you have to get that pedantic with all of them, Slot 1 launched on a 440FX board, that chipset and volt reg ain't running a 1Ghz PIII, the launch socket 478 board ain't running anything over 400fsb 2ghz and so on.

BitWrangler wrote on 2024-02-07, 01:57:

ElectroSoldier wrote on 2024-02-07, 01:49:

Except you cant run a PIII-s 1400 in a PPGA socket 370 because its FC-PGA.
You can modify it of course, but then its not the same.

But then you have to get that pedantic with all of them, Slot 1 launched on a 440FX board, that chipset and volt reg ain't running a 1Ghz PIII, the launch socket 478 board ain't running anything over 400fsb 2ghz and so on.

I own one of them Slot1 440FX boards, quite an interesting board visually if not a little odd since it has no AGP support. Almost looks like skunk works product where some one thought to just throw a slot 1 socket onto a socket 8 board.

BitWrangler wrote on 2024-02-07, 01:57:

ElectroSoldier wrote on 2024-02-07, 01:49:

Except you cant run a PIII-s 1400 in a PPGA socket 370 because its FC-PGA.
You can modify it of course, but then its not the same.

But then you have to get that pedantic with all of them, Slot 1 launched on a 440FX board, that chipset and volt reg ain't running a 1Ghz PIII, the launch socket 478 board ain't running anything over 400fsb 2ghz and so on.

By that standard, aren't the AMD sockets sure to win? I think their forward compatibility is much better than most of Intel's where you need new chipsets/motherboards even if the socket technically didn't change...

VivienM wrote on 2024-02-07, 02:07:

BitWrangler wrote on 2024-02-07, 01:57:

ElectroSoldier wrote on 2024-02-07, 01:49:

Except you cant run a PIII-s 1400 in a PPGA socket 370 because its FC-PGA.
You can modify it of course, but then its not the same.

But then you have to get that pedantic with all of them, Slot 1 launched on a 440FX board, that chipset and volt reg ain't running a 1Ghz PIII, the launch socket 478 board ain't running anything over 400fsb 2ghz and so on.

By that standard, aren't the AMD sockets sure to win? I think their forward compatibility is much better than most of Intel's where you need new chipsets/motherboards even if the socket technically didn't change...

Yeah I mean it sure seems old enough...
Its got to be close to 20 years since it was in use... Putting a Athlon 3200+ in the same age range as a Pentium 100...

Its not a question of basic maths... slowest CPU to fastest CPU on a socket and getting the percentage differences between them.
Its more a question of what is vintage.

ElectroSoldier wrote on 2024-02-07, 02:20:

Yeah I mean it sure seems old enough... Its got to be close to 20 years since it was in use... Putting a Athlon 3200+ in the sam […]
Show full quote

VivienM wrote on 2024-02-07, 02:07:

BitWrangler wrote on 2024-02-07, 01:57:

But then you have to get that pedantic with all of them, Slot 1 launched on a 440FX board, that chipset and volt reg ain't running a 1Ghz PIII, the launch socket 478 board ain't running anything over 400fsb 2ghz and so on.

By that standard, aren't the AMD sockets sure to win? I think their forward compatibility is much better than most of Intel's where you need new chipsets/motherboards even if the socket technically didn't change...

Yeah I mean it sure seems old enough...
Its got to be close to 20 years since it was in use... Putting a Athlon 3200+ in the same age range as a Pentium 100...

Its not a question of basic maths... slowest CPU to fastest CPU on a socket and getting the percentage differences between them.
Its more a question of what is vintage.

Vintage, its a rather nebulous term that doesn't have a definitive age range to work with, but I'm sure most here would consider anything after normal socket 7 to be retro rather than vintage. Some may even consider anything newer than a 486 to not be Vintage, I doubt we will ever have a unified consensus on the subject.

For me I see anything prior to the 386 as being a Vintage setup, essentially using the introduction of 32bit as the cut off, anything after and including the 386 is retro.

ElectroSoldier wrote on 2024-02-07, 01:49:

Except you cant run a PIII-s 1400 in a PPGA socket 370 because its FC-PGA.
You can modify it of course, but then its not the same.

Fair enough. Here's a revised, FC-PGA only Socket 370 result. According to my screencap, I actually used a Slot 1 PIII-500E. However the socketed version should be identical in performance.

From 4905 with the PIII-500E + 9800 Pro to 11496 with the PIII-1400S + 9800Pro. Difference = 2.34x. Boo, that's not nearly as dramatic!

Trashbytes wrote on 2024-02-07, 01:55:

Why did you stop at the E8600?, there exists the E8800 ES chips at 3.666Ghz and also the mighty X5470 which will happily clock out to 4.3 Ghz and supports dual GTX 295's. The fact the E8800 is an ES chip shouldnt exclude it from this discussion as it does physically exist and has been benched, and do modded 771 Xeons get excluded when 771 and 775 are the same socket just rotated 90 degrees.

If we exclude Xeons then the QX9770 will also clock out to 4Ghz and with dual GTX 580's should happily cruise past any score the E8600 could get.

Ive run a Titan Maxwell on a QX9770 ...so if we include that then Socket 775 is easily the socket with the largest power gains with no hardware modding, along with being one of the longest if not the longest lived Intel socket.

I was aiming strictly for single-threaded performance at stock clocks. E8600 is the fastest official CPU for benching 3DMark2001 on LGA 775.
I can post some overclocked results for all of these sockets, but I'd have to rummage through my poorly organized collection of benchmark notes and screencaps again...yikes! 😜

IMO unmodified socket - fastest CPU at launch compared to fastest CPU available is the only fair way to consider it. Taking the slowest CPU a socket supports and comparing it to the fast CPU that can be hacked to work is disingenuous IMO.

The problem with slot 1 and socket 370 is that the CPUs at release were too good. A 333mhz pentium II was available at launch for slot 1 and the fastest slot 1 drop in replacement is katmai@600mhz. You need slotket adapters to make copper mine work which is not really the original socket anymore.

The same is true for socket 478 ad 775. There were multiple chipset revisions. The earliest 775 board could not support a core2duo.

I think socket 5 is probably the winner. You could get a brand new socket 5 board with a Pentium at 66mhz in 1994 and that was top of the line. Later you could buy a 3.3v Pentium MMX @ 200mhz as a drop in replacement - more than tripling the clock speed and giving you MMX.

Socket 486 is another strong contender. At launch in 1989 the fastest chip available was a DX 25mhz. You could drop in replace a dx4@75mhz (100mhz would require changing the crystal to a 33mhz and a AMD 586 would require a socket adapter).

Socket 1 might be just as as good as the original 486 socket. If you had a board with a 33mhz crystal and paired it with a 486DX33 you could drop in replace with a dx4 100.

If you are looking at no rules, whatever it takes, and anything goes then yeah it's socket 486/socket 1. Technically it's possible to run a 486sx@16mhz and a AMD 586dx@133+ mhz on the same board. But you would need to modify the board a lot in order to do it.

Standard Def Steve wrote on 2024-02-07, 03:00:

Fair enough. Here's a revised, FC-PGA only Socket 370 result. According to my screencap, I actually used a Slot 1 PIII-500E. How […]
Show full quote

ElectroSoldier wrote on 2024-02-07, 01:49:

Except you cant run a PIII-s 1400 in a PPGA socket 370 because its FC-PGA.
You can modify it of course, but then its not the same.

Fair enough. Here's a revised, FC-PGA only Socket 370 result. According to my screencap, I actually used a Slot 1 PIII-500E. However the socketed version should be identical in performance.

From 4905 with the PIII-500E + 9800 Pro to 11496 with the PIII-1400S + 9800Pro. Difference = 2.34x. Boo, that's not nearly as dramatic!

Trashbytes wrote on 2024-02-07, 01:55:

Why did you stop at the E8600?, there exists the E8800 ES chips at 3.666Ghz and also the mighty X5470 which will happily clock out to 4.3 Ghz and supports dual GTX 295's. The fact the E8800 is an ES chip shouldnt exclude it from this discussion as it does physically exist and has been benched, and do modded 771 Xeons get excluded when 771 and 775 are the same socket just rotated 90 degrees.

If we exclude Xeons then the QX9770 will also clock out to 4Ghz and with dual GTX 580's should happily cruise past any score the E8600 could get.

Ive run a Titan Maxwell on a QX9770 ...so if we include that then Socket 775 is easily the socket with the largest power gains with no hardware modding, along with being one of the longest if not the longest lived Intel socket.

I was aiming strictly for single-threaded performance at stock clocks. E8600 is the fastest official CPU for benching 3DMark2001 on LGA 775.
I can post some overclocked results for all of these sockets, but I'd have to rummage through my poorly organized collection of benchmark notes and screencaps again...yikes! 😜

No no .. 🤣 I know what you mean by yikes 🤣, I avoid even opening my backup images folder as even my modern PC groans at having to read that many files.

Like you I really need to organise it if only to make opening the folder faster.

mothergoose729 wrote on 2024-02-07, 03:07:

IMO unmodified socket - fastest CPU at launch compared to fastest CPU available is the only fair way to consider it. Taking the […]
Show full quote

IMO unmodified socket - fastest CPU at launch compared to fastest CPU available is the only fair way to consider it. Taking the slowest CPU a socket supports and comparing it to the fast CPU that can be hacked to work is disingenuous IMO.

The problem with slot 1 and socket 370 is that the CPUs at release were too good. A 333mhz pentium II was available at launch for slot 1 and the fastest slot 1 drop in replacement is katmai@600mhz. You need slotket adapters to make copper mine work which is not really the original socket anymore.

The same is true for socket 478 ad 775. There were multiple chipset revisions. The earliest 775 board could not support a core2duo.

I think socket 5 is probably the winner. You could get a brand new socket 5 board with a Pentium at 66mhz in 1994 and that was top of the line. Later you could buy a 3.3v Pentium MMX @ 200mhz as a drop in replacement - more than tripling the clock speed and giving you MMX.

Socket 486 is another strong contender. At launch in 1989 the fastest chip available was a DX 25mhz. You could drop in replace a dx4@75mhz (100mhz would require changing the crystal to a 33mhz and a AMD 586 would require a socket adapter).

Socket 1 might be just as as good as the original 486 socket. If you had a board with a 33mhz crystal and paired it with a 486DX33 you could drop in replace with a dx4 100.

If you are looking at no rules, whatever it takes, and anything goes then yeah it's socket 486/socket 1. Technically it's possible to run a 486sx@16mhz and a AMD 586dx@133+ mhz on the same board. But you would need to modify the board a lot in order to do it.

If we include modding then we can just take a P45/X48 775 board and make it support both the first and last 775 CPUs with a BIOS mod at which point the performance jump is enormous, I think people forget just how big the performance delta between Netburst and CD2 was. (Especially if its between Presshot and Yorkfield XE)

Trashbytes wrote on 2024-02-07, 03:17:

mothergoose729 wrote on 2024-02-07, 03:07:

IMO unmodified socket - fastest CPU at launch compared to fastest CPU available is the only fair way to consider it. Taking the […]
Show full quote

IMO unmodified socket - fastest CPU at launch compared to fastest CPU available is the only fair way to consider it. Taking the slowest CPU a socket supports and comparing it to the fast CPU that can be hacked to work is disingenuous IMO.

The problem with slot 1 and socket 370 is that the CPUs at release were too good. A 333mhz pentium II was available at launch for slot 1 and the fastest slot 1 drop in replacement is katmai@600mhz. You need slotket adapters to make copper mine work which is not really the original socket anymore.

The same is true for socket 478 ad 775. There were multiple chipset revisions. The earliest 775 board could not support a core2duo.

I think socket 5 is probably the winner. You could get a brand new socket 5 board with a Pentium at 66mhz in 1994 and that was top of the line. Later you could buy a 3.3v Pentium MMX @ 200mhz as a drop in replacement - more than tripling the clock speed and giving you MMX.

Socket 486 is another strong contender. At launch in 1989 the fastest chip available was a DX 25mhz. You could drop in replace a dx4@75mhz (100mhz would require changing the crystal to a 33mhz and a AMD 586 would require a socket adapter).

Socket 1 might be just as as good as the original 486 socket. If you had a board with a 33mhz crystal and paired it with a 486DX33 you could drop in replace with a dx4 100.

If you are looking at no rules, whatever it takes, and anything goes then yeah it's socket 486/socket 1. Technically it's possible to run a 486sx@16mhz and a AMD 586dx@133+ mhz on the same board. But you would need to modify the board a lot in order to do it.

If we include modding then we can just take a P45/X48 775 board and make it support both the first and last 775 CPUs with a BIOS mod at which point the performance jump is enormous, I think people forget just how big the performance delta between Netburst and CD2 was. (Especially if its between Presshot and Yorkfield XE)

No if you do that one of the AM sockets is the winner by far. I don't know exactly where the cut offs are but AM3+ supports a lot of very fast and comparatively very slow CPUs.

AM2+ could do a 6 core phenom II and an ultra slow single core sempron - as another example.

mothergoose729 wrote on 2024-02-07, 03:07:

I think socket 5 is probably the winner. You could get a brand new socket 5 board with a Pentium at 66mhz in 1994 and that was top of the line. Later you could buy a 3.3v Pentium MMX @ 200mhz as a drop in replacement - more than tripling the clock speed and giving you MMX.

Socket 5 and 66 MHz Pentium? Haven't they been built for Socket 4 with its 5 Volt?

mothergoose729 wrote on 2024-02-07, 03:07:

IMO unmodified socket - fastest CPU at launch compared to fastest CPU available is the only fair way to consider it. Taking the slowest CPU a socket supports and comparing it to the fast CPU that can be hacked to work is disingenuous IMO.

The problem with slot 1 and socket 370 is that the CPUs at release were too good. A 333mhz pentium II was available at launch for slot 1 and the fastest slot 1 drop in replacement is katmai@600mhz. You need slotket adapters to make copper mine work which is not really the original socket anymore.

Erm, not exactly, the wait to 333mhz was a long one and the 300mhz variant which supposedly launched in May alongside the 233/266 was vaporware in the computer shopper 6 months (nobody would accept orders or put you on the waiting list if you tried to force order one). Maybe different for name brands but I didn’t know anyone with a 300mhz p2 until almost 1998 as none were on the shelf, even if you had deep pockets.

So I’m not sure how much that matters but yes compatiblity issues did abound in that era, theoretically stuck at 66mhz FSB (possibly 75 or 83 if you had a board that could overclock) on launch edition boards which realistically limited you to a rare 733 celeron.

Now to expand, I owned a socket adapter for socket 8 ppro, that allowed you to install socket 370 chips, the 733 celeron I had seemed to work in the ppro motherboard .

In terms of platform if we include ppro we have 133mhz-1400mhz using the architecture.

mothergoose729 wrote on 2024-02-07, 03:07:

I think socket 5 is probably the winner. You could get a brand new socket 5 board with a Pentium at 66mhz in 1994 and that was top of the line. Later you could buy a 3.3v Pentium MMX @ 200mhz as a drop in replacement - more than tripling the clock speed and giving you MMX.

Nope, your thinking socket 4 which was 5 volt and 60/66mhz and using a socket adapter could be upgraded to 400mhz

mothergoose729 wrote on 2024-02-07, 03:07:

Socket 486 is another strong contender. At launch in 1989 the fastest chip available was a DX 25mhz.

Technically the dx20 launched in April as release candidates and was available earlier than the June 486dx25 launch , oddly Intel had to do a recall and late redesign of the 486 motherboard for reasons I forget. Ah well.

mothergoose729 wrote on 2024-02-07, 03:07:

Technically it's possible to run a 486sx@16mhz and a AMD 586dx@133+ mhz on the same board. But you would need to modify the board a lot in order to do it.

Modify? Not really, I owned a very early Micronics ISA only 486 motherboard with a socketed crystal, the motherboard was designed under the assumption that you had to match the cpu and crystal, besides the voltage needing adapting there was really nothing stopping you from installing any 486 cpu, technically even the 5x86-150 would have been fine on the board as it didn’t seem to be much affected by high or low bus speed crystals.

mothergoose729 wrote on 2024-02-07, 03:31:

Trashbytes wrote on 2024-02-07, 03:17:

mothergoose729 wrote on 2024-02-07, 03:07:

IMO unmodified socket - fastest CPU at launch compared to fastest CPU available is the only fair way to consider it. Taking the […]
Show full quote

IMO unmodified socket - fastest CPU at launch compared to fastest CPU available is the only fair way to consider it. Taking the slowest CPU a socket supports and comparing it to the fast CPU that can be hacked to work is disingenuous IMO.

The problem with slot 1 and socket 370 is that the CPUs at release were too good. A 333mhz pentium II was available at launch for slot 1 and the fastest slot 1 drop in replacement is katmai@600mhz. You need slotket adapters to make copper mine work which is not really the original socket anymore.

The same is true for socket 478 ad 775. There were multiple chipset revisions. The earliest 775 board could not support a core2duo.

I think socket 5 is probably the winner. You could get a brand new socket 5 board with a Pentium at 66mhz in 1994 and that was top of the line. Later you could buy a 3.3v Pentium MMX @ 200mhz as a drop in replacement - more than tripling the clock speed and giving you MMX.

Socket 486 is another strong contender. At launch in 1989 the fastest chip available was a DX 25mhz. You could drop in replace a dx4@75mhz (100mhz would require changing the crystal to a 33mhz and a AMD 586 would require a socket adapter).

Socket 1 might be just as as good as the original 486 socket. If you had a board with a 33mhz crystal and paired it with a 486DX33 you could drop in replace with a dx4 100.

If you are looking at no rules, whatever it takes, and anything goes then yeah it's socket 486/socket 1. Technically it's possible to run a 486sx@16mhz and a AMD 586dx@133+ mhz on the same board. But you would need to modify the board a lot in order to do it.

If we include modding then we can just take a P45/X48 775 board and make it support both the first and last 775 CPUs with a BIOS mod at which point the performance jump is enormous, I think people forget just how big the performance delta between Netburst and CD2 was. (Especially if its between Presshot and Yorkfield XE)

No if you do that one of the AM sockets is the winner by far. I don't know exactly where the cut offs are but AM3+ supports a lot of very fast and comparatively very slow CPUs.

AM2+ could do a 6 core phenom II and an ultra slow single core sempron - as another example.

If we are talking performance delta here there isn't a huge jump between the lowest AM2 and the highest one, you are after all still working with CPUs based around the same Uarch. The jump CD2 made from Netburst surprised even AMD, it really was that big a jump, I like to think of CD2 as the Pentium III on steroids which in a way it is.

One could even make the case that the earlier Athlon AM2 cpus were better IPC wise than the later ones, AMD made some really stupid decisions as they moved towards AM3 and the Phenom II/Bulldozer line and their performance uplift was pretty terrible.

They even made the incredibly boneheaded decisions that Intel made in regards to Clock speed and Power limits and pushed out the FX9590 that while it could hit 5Ghz the lower clocked 2600k shit all over it in both IPC and power usage. It was AMDs Presshot arc and it took Ryzen to pull them out of the hole, I guess you could say Ryzen was their CD2 moment.

Slot 2 is interesting as a contender.
Its raw numbers would be a 100% performance increase from the PIII 500 to the PIII 1000.

Fastest launch CPU for socket 5 was a Pentium 90.
Slot 1 it was in actuality the 266 which was available to go from launch date.

BitWrangler wrote on 2024-02-07, 16:38:

Fastest launch CPU for socket 5 was a Pentium 90.
Slot 1 it was in actuality the 266 which was available to go from launch date.

Socket 5 was the original socket for Pentium was it not? The first Pentium had a clock speed of 66mhz, but it was very quickly followed up by a 90mhz sku.

EDIT: Nope, there was a short lived socket 4. You're right. 90mhz -> 200mhz on socket 5.