I'd say it's a toss-up between a Ryzen 9 and an Intel Core i9

Why? Because DDR4 is DDR4-SDRAM, so SDRAM too.

If you mean SDR-SDRAM, you should say so. In that case, as with any "fastest" thing: you need to specify fastest in *what* specifically. Athlon XP had higher IPC and better FPU than P4, plus the architecture was less memory-bandwidth dependent, so overall the fastest 133MHz FSB AthlonXP probably would win, but if an application used SSE2 or was otherwise heavily optimised for P4, an i845B chipset with the fastest 133MHz FSB P4 it could run would be fastest.

dionb wrote on 2020-07-28, 10:13:

If you mean SDR-SDRAM, you should say so.

Why?

Since when did it matter ?

AS long as I can remember their was NEVER any confusion when someone said sd-ram it was interpreted as single data rate.

DDR = Double data rate
DDR2
DDR3
DDR4

And so on.

Eh, no, SD in SDRAM doesn't mean Single Datarate, it means Synchronus Dynamic. So all DDR memory types are in effect SDRAM. I understand why you assume it automatically should mean single data Rate SDRAM, but it really doesn't.

And there's also things like DDR SGRAM...

Yeah but when was it ever confused when DDR ram came along?

Thats the point im getting at, ram was always SD OR DDR in the early days in the consumer market anyway, their was no confusion about the abbreviation "SD" "RAM"

So if you want to be really technical in response to dionb, you should always say "DDR-SDRAM, DDR2-SDRAM, DDR3-SDRAM, DDR4-SDRAM"

SO if its not alright to call SDR-SDRAM SD-RAM, why is it alright to abbreviate DDR, DDR2, etc without the SD-RAM tag on the end?

SOrry I basically think some people like to start arguments.

I had a socket 7 board that came with PC166 168 pin sdram
Slight overclock on the bus speed

I would guess a system with such ram running at a higher bus speed would be the best

dirkmirk wrote on 2020-07-28, 11:15:

Yeah but when was it ever confused when DDR ram came along?

Thats the point im getting at

I don’t think anyone has ever confused the two, at least not in the last 20 years

I didn't know there were SDR-133 P4's. Wouldn't those be slower than a Tualatin P3?

dirkmirk wrote on 2020-07-28, 11:15:

Yeah but when was it ever confused when DDR ram came along?

Thats the point im getting at, ram was always SD OR DDR in the early days in the consumer market anyway, their was no confusion about the abbreviation "SD" "RAM"

If you're putting the dash after "SD" there obviously is a major element of confusion. It's SDRAM or if you must insert a dash, it's S-DRAM (synchronous DRAM). SD is for Secure Digital memory cards.

So if you want to be really technical in response to dionb, you should always say "DDR-SDRAM, DDR2-SDRAM, DDR3-SDRAM, DDR4-SDRAM"

Not at all, if you're comparing DDR3 to DDR4 there's no need or added value to say more. But if you say "SDRAM" without qualification it can apply to the whole lot.

Oetker wrote on 2020-07-28, 11:18:

I didn't know there were SDR-133 P4's. Wouldn't those be slower than a Tualatin P3?

Depends once again on what you were running. Giving a P4 only 44 33% of its desired ram bandwidth (100 quad pumped FSB vs 133MHz single pumped RAM) certainly castrated it, but if code wasn't too branching, sheer clock speed would win, partiularly if you could install a Northwood up to 3GHz. Can't find any benchmarks, because these things predate Tualatin and by the time that came along, nobody was benching i845 with SDR as the DDR versions were out and nobody looked back.

Edit: can't count 😦

dionb wrote on 2020-07-28, 11:44:

Giving a P4 only 44% of its desired ram bandwidth (100 quad pumped FSB vs 133MHz single pumped RAM) certainly castrated it...

You mean 33% right?

I agree that whatever fastest Pentium4 you can squeeze in one of those SDR based motherboards is probably going to be the fastest overall. Another possibility would be a Via KT266A based motherboard with the fastest Athlon XP it can handle, since apparently that chipset also supports SDR although I don't remember ever seeing a motherboard based on this chipset that supported that kind of memory.

devius wrote on 2020-07-28, 12:44:

Another possibility would be a Via KT266A based motherboard with the fastest Athlon XP it can handle, since apparently that chipset also supports SDR although I don't remember ever seeing a motherboard based on this chipset that supported that kind of memory.

I have ecs k7s5a , it handle 0.13 athlon XP and has 2 SDR slots. It's SIS735 based. Not tested it yet.

It works fine with SDR and isn't even that much slower than with DDR. I had it with both kinds back in the day, pity that I don't have proper benchmarks from back then, nor do I have the board anymore.

On a side note, it's not on nForce2 level, of course, not even with DDR memory. But it has a decent SDR controller.

devius wrote on 2020-07-28, 12:44:

[...]

You mean 33% right?

Er yes, maths failure there...

I agree that whatever fastest Pentium4 you can squeeze in one of those SDR based motherboards is probably going to be the fastest overall. Another possibility would be a Via KT266A based motherboard with the fastest Athlon XP it can handle, since apparently that chipset also supports SDR although I don't remember ever seeing a motherboard based on this chipset that supported that kind of memory.

Can't remember one either, there were lots with the original KT266, but that was quite a bit slower, definitely slower than the SiS 735.

alvaro84 wrote on 2020-07-28, 13:41:

It works fine with SDR and isn't even that much slower than with DDR. I had it with both kinds back in the day, pity that I don't have proper benchmarks from back then, nor do I have the board anymore.

On a side note, it's not on nForce2 level, of course, not even with DDR memory. But it has a decent SDR controller.

The Athlon's bus is far less bandwidth-sensitive than the P4's - it's narrower ("only" 2100Mbps vs 3200Mbps) and because the CPU is less pipelined, the impact of a branch miss is much smaller, so the difference between SDR and DDR is much smaller. If you compare i845 board with both it's day and night, by comparison.

That's another reason why optimizations were so important for the P4: if you could generate more 'predictable' code, it would have less pipeline flushes and could perform better. The Athlon was a much better general-purpose beast that just ate whatever you threw at it. Q3A is a prime example: as always John Carmack prided himself on the quality of his code and its optimization. Q3A always performed better on P4, as a showcase of what near-perfect code could do. Other games didn't do anywhere near so well. UT tended to perform much better on Athlon.

That's also why nForce2 was an improvement, but nowhere near as big an improvement as i865 was: the chipset was fine, but the hunger for bandwidth wasn't as huge.

overdrive333 wrote on 2020-07-28, 10:07:

What is it ? May be via kt133 + top athlon xp? Or some s478?

That honour would definitely go to the 266FSB version of the Athlon XP 2600+ on a SiS 735 motherboard.

Netburst with SDRAM is ugly man. I kid you not: combine a 2.6GHz Celeron (128K L2) with SDRAM and it drops to the performance level of a 1GHz Coppermine when SSE2 isn't being used. With a proper 512K Northwood P4, the performance penalty is far less severe, but it's still not going to be as "fast" as an XP 2600+ on SDRAM.

Of course it won't be. Athlon K7 family had very fat L1 cache, so they don't rely on L2 size and memory bandwidth as much.