Athlon 64's

[Solaris17]

well we all remember the athlon xp's on socket A and we all remember that the naming schemes were based on the T-bred core EX: 2800+ = preformance of a T-bred at 2.8Ghz but what about the athlon 64's? i mean 3700+...........compared to what? i mean i dont think they would still compair it to the T-bred because its such old tech now..any ideas?

 

[gygabite]

Good question, perhaps they say a 3700+ has the performance of a P4 @ 3.7GHz

 

[_33]

well we all remember the athlon xp's on socket A and we all remember that the naming schemes were based on the T-bred core EX: 2800+ = preformance of a T-bred at 2.8Ghz but what about the athlon 64's? i mean 3700+...........compared to what? i mean i dont think they would still compair it to the T-bred because its such old tech now..any ideas?

The AMD Athlon XP numbers came in around 2001 when they went from SDR memory to DDR memory and added SSE to the new core. The processors then came with what is aclled a PR rating scheme. So, with the added features, they decided to use the PR rating in the name and so an Athlon of 1,54 Ghz was an Athlon XP 1800+.

Then, things continued to evolve, the circuitry got shrinked from 0,18 down to 0,13 (don't remember the unit it uses, microns?) added some more cache memory, and added on board memory controller, then the margins between PR rating and actual gigahertz went a little wider.

Finally, AMD added 64bit, SSE2, increased the memory controller capability (128bit HTT, if I'm not mistaken), and in some cases 1024K of L2 cache instead of 512K, chip circuitry got smaller, passing from 0,13 to 0,09, again PR rating started to get a little crazy, to the point where you would buy an AMD 64 rated 3000+ but actually runs at 1.8Ghz.

This PR rating nonsense striked me, when I decided to compare my 4 year old AMD XP 1800+ (1,54Ghz) to my brand new AMD 64 3000+ (1,8Ghz) in sisoft sandra. Well, the results speak from themselves, as the XP 1800+ hits over 6000 mips (million instructions per second), while the A64 3000+ is barely faster an edge over 8000 mips. So, YES the PR rating doesn't make any sense.

The solution for me was overclocking. I took my A64 3000+ and my DFI Lanparty mobo, and started overclocking the CPU. I managed to bump speed from the stock 1,8 ghz up to a sizeable 2,8 ghz, which clocks in a neat 12000+ mips in sisoft sandra, basically doubling the speed over my 4 year old XP 1800+ rig. From then, I tought; "Hey, 4 years apart, and only twice the speed... Great job AMD!!! Keep those PR rating climbing!!!"

A lesson to learn? Screw moore's law!

 

[trog100]

the 64 amd rating is compared to a P4 at xxx speed .. the sempron is compared to a celeron at xxx speed..

sandra readings dont give true performance comparisons either..

folks used to buy or measure cpus on the clock speed.. being as the amd cpu does more work per clock cycle.. amd had little choice but to come up with the PR system..

its not entirely accurate but thats the basis of it..

course we now have single core amd PRs compared with dual core amd PRs.. a single core with a speed of 2.4 would have PR of 4000 while a dual core at 2.4 would have PR of 4800..

i assume the second PR would be a single core P4 at 4800.. not as any such thing exsists..

interestingly amd seem to think a dual core at 2.4 gig would have roughly the same performance as a single core at about 3 gig.. price wise they also seem to cost about double for the similar speed dual core as the single core..

trog

 

[_33]

sandra readings dont give true performance comparisons either..

trog, the concept of MIPS (million instructions per second) has always existed in computers, it's the primary reference when looking at calculation power on the instruction level. The Motorola 68000 at 8mhz performed 0,8 mips, the commodore 64 did 0,2 mips. Every CPU has their mips performance. I've used sandra for a lot of reference info, and it has never failed or proven wrong in evaluating. I really don't want to convince you, but...

 

[trog100]

according to sandra a dual core cpu has exactly twice the performance of single core.. u gonna tell me that reflects real life performance in any way shape or form..

the other factors that sandra dosnt take into account and amd PR ratings do is the fsb speed or the cache size.. all these things factor in to real life performance which is why i say

"sandra readings dont give true performance comparisons either"..

it quite simply dosnt..

trog

 

[wazzledoozle]

The AMD Athlon XP numbers came in around 2001 when they went from SDR memory to DDR memory and added SSE to the new core. The processors then came with what is aclled a PR rating scheme. So, with the added features, they decided to use the PR rating in the name and so an Athlon of 1,54 Ghz was an Athlon XP 1800+.

Then, things continued to evolve, the circuitry got shrinked from 0,18 down to 0,13 (don't remember the unit it uses, microns?) added some more cache memory, and added on board memory controller, then the margins between PR rating and actual gigahertz went a little wider.

Finally, AMD added 64bit, SSE2, increased the memory controller capability (128bit HTT, if I'm not mistaken), and in some cases 1024K of L2 cache instead of 512K, chip circuitry got smaller, passing from 0,13 to 0,09, again PR rating started to get a little crazy, to the point where you would buy an AMD 64 rated 3000+ but actually runs at 1.8Ghz.

This PR rating nonsense striked me, when I decided to compare my 4 year old AMD XP 1800+ (1,54Ghz) to my brand new AMD 64 3000+ (1,8Ghz) in sisoft sandra. Well, the results speak from themselves, as the XP 1800+ hits over 6000 mips (million instructions per second), while the A64 3000+ is barely faster an edge over 8000 mips. So, YES the PR rating doesn't make any sense.

The solution for me was overclocking. I took my A64 3000+ and my DFI Lanparty mobo, and started overclocking the CPU. I managed to bump speed from the stock 1,8 ghz up to a sizeable 2,8 ghz, which clocks in a neat 12000+ mips in sisoft sandra, basically doubling the speed over my 4 year old XP 1800+ rig. From then, I tought; "Hey, 4 years apart, and only twice the speed... Great job AMD!!! Keep those PR rating climbing!!!"

A lesson to learn? Screw moore's law!

Superpi is a better indicator of cpu speed, run it on your old 1.5 ghz athlon and compare it to your 3000 at stock speed. The 3000 will blow it away.

Sandra does not indicate any real world performance. According to it, my old Athlon XP at 2.6 GHz was equivalent to a 3.6 ghz Pentium 4, which anyone know is completely untrue in real world performance.

 

[noneed4me2]

Superpi is a better indicator of cpu speed, run it on your old 1.5 ghz athlon and compare it to your 3000 at stock speed. The 3000 will blow it away.

Sandra does not indicate any real world performance. According to it, my old Athlon XP at 2.6 GHz was equivalent to a 3.6 ghz Pentium 4, which anyone know is completely untrue in real world performance.

You know i wanted to upgrade so bad from my old P3 1.2ghz 100mhz bus intel based board, after reading all the raves of AMDs 64 cpu's (and gaming just sucked on my old system) but with the exception of gaming and encoding, it seems at times my old P3 is as zippy in other everyday use stuff like internet, AVI files, mp3s and the like, and my dads newer HP 2.6 celeron just seems so slow even after a fresh OS install and driver update compared to my old workhorse. It seems to me for people who don't use PCs for anything other than internet and multimedia the technology has just peaked and anything in todays market will do if your not an enthusiast.

 

[Quake2owns]

a pentium 3 1.ghz should overclock to atleast 1.3 ghz with good sdr 168 pin ram you can buy a socket 370 motherboard with agpx4 slot and use it with a x850xtpe or a 7800gs or 7800 GT.

 

[yogurt_21]

hmm wonder how much of a bottleneck that'd be? lol

 

[Polaris573]

a pentium 3 1.ghz should overclock to atleast 1.3 ghz with good sdr 168 pin ram you can buy a socket 370 motherboard with agpx4 slot and use it with a x850xtpe or a 7800gs or 7800 GT.

I've put a 6800 in a PII system before (400Mhz). I should have run some benchmarks on it while it was in there. Just for curiosity sake (Yes I know it would suck).

 

[noneed4me2]

a pentium 3 1.ghz should overclock to atleast 1.3 ghz with good sdr 168 pin ram you can buy a socket 370 motherboard with agpx4 slot and use it with a x850xtpe or a 7800gs or 7800 GT.

already got a new system, that one (the p3) went to my kids, it has 512mb ram 80gb hd runs XP pro fine and can play all there Disney and PBS kids games no problem, it has a 9550se with 128 ram and it plays Halo great which is my 5 year olds current favorite. Anything better as far as graphics would be a waste in my opinion, the 9550 was the last upgrade i was willing to put into that machine and the card in it before was an ancient AIW Radeon, one of the first, and it cost over $300 when i bought new; before that one it was a 16mb voodoo 3 (i still miss those guys, use to have an original SLI setup).

 

[Quake2owns]

hmm wonder how much of a bottleneck that'd be? lol

well my old socket 370 466 mhz celeron with a 9800 pro can play ut2k4/ doom3/quake 4 lol

 

[noneed4me2]

You just prove my point i think some of that old tech would last longer if people write better code instead of pushing for constant upgrades, i mean i think its great but It seems i was just getting use to AGP and then there is PCI/e and forget about all the different cpu variants, almost everywhere i read saidd Halo would not run on my old AIW radeon but it ran fine if all the settings were low, and would run on the 9550 with settings high and people said that it shouldn't. I am glad i keep most of my old stuff. I got windows 2000 installed on a Pentium 133 with i think had 32mb ram, way under system requirements, just to see if I could. It took several hours to install but once it did it ran ok, better than winows 3.1.

 

[Quake2owns]

im thinking about building a cheap pentium 3 rig ,

 

[noneed4me2]

you could definately do it on the cheap, maybe set it up as your own server, i've been itching to do that with one of my old systems.

 

[Quake2owns]

you could definately do it on the cheap, maybe set it up as your own server, i've been itching to do that with one of my old systems.

maybe sff?

 

[_33]

Superpi is a better indicator of cpu speed, run it on your old 1.5 ghz athlon and compare it to your 3000 at stock speed. The 3000 will blow it away.

Sandra does not indicate any real world performance. According to it, my old Athlon XP at 2.6 GHz was equivalent to a 3.6 ghz Pentium 4, which anyone know is completely untrue in real world performance.

Of course it's not real world performance, it's targeted benchmarks, meaning MEMORY, CPU, HARD DISKS? etc etc... It's not meant for general blabla benchmarks...

The AMD XP 1800+ config is sold now, but a Pentium 3 Tualatin with memory of SDR133 speed on an ASUS board gives a run of aproximately 2 minutes and 15 seconds. If you look in sisoft Sandra, the 1.2 Ghz Tualatin is rated 4000 mips, which is roughly 3 times slower than my current OVERCLOCKED processor that clocks 12000+ mips. Now if I run superPI, I get around 35 seconds. The problem with superpi is that it's not representative of ANY application, and it doesn't use cpu instructions in a balanced way. It's just calculating PIs. Sisoft Sandra focuses on CPU power by launching it in a trial mode, which is totally different. Superpi is not MY alltime reference tool for computer speed, definately. Then, a raytracer, would probably be a better way, but even then...

Calculating mips is a reference, believe it or not.

Seriously, it takes a 1/3 speed boost on the P4 processor in order to surpass an AMD, everyone knows that. That is why at 3.6Ghz, the P4 can get nailed by a 2.4Ghz AMD. The reason for this is the netburst architecture with long pipeline and hyperthreading. If you leave hyperthreading off, you get 2 execution pipelines on the P4, as opposed to 3 execution pipelines on the AMD.

 

[_33]

according to sandra a dual core cpu has exactly twice the performance of single core.. u gonna tell me that reflects real life performance in any way shape or form..

the other factors that sandra dosnt take into account and amd PR ratings do is the fsb speed or the cache size.. all these things factor in to real life performance which is why i say

"sandra readings dont give true performance comparisons either"..

it quite simply dosnt..

trog

Duh, there's 2 cores with their own cache memory... what do you expect??? Two applications running both 100% cpu utilisation will perform basically twice the calculation power as opposed to 1 core, at same ghz speed, regardless. No bottlenecks considered. What you can't do with a dual core is launch 1 application and expect it to do double the speed. The benefit you get on dual core is that 1 core takes care of all windoze requests and the other core runs your application.... So for example, your antivirus and windows runs on 1 core, the other runs your game. So your game will basically run at the full potential of 1 core, and the antivirus will be using the maximum of the other core. Or let's say, you encode a DIVX on 1 core, and you run REASON 3.0 on the other core........ 2 CORES!

You want to do a dual core test? Launch SUPERPI twice. I challenge you that both runs will be as fast as a single core of same speed.

 

[trog100]

a p3 was a better chip than a p4.. your p3 at 1.2 is probably as fast as a p4 type celeron at 2.4.. he he he.. praps not quite but not far off..

the new conroe will be more like a p3 than a p4..

trog

 

[yogurt_21]

yup the p3's were great cpu's, then came the 423 p4's with rambus memory *shudder* I think the whole push for the GHZ with intel totally shafted their engineers. I mean all the way up until the p4's intel totally dominated amd, then came the whole rambus fiasco (not to mention the socket winthin a socket) and amd finally had a shot. now we're seeing intel finally having hit the GHZ wall (which is odd cause they said no 4GHZ cpu would be released and anyone with a 3.6 can easily hit 4.5 or better) we're seeing that old intel who slowed down and waited to release new cpu's till they were truly better.

 

[_33]

yup the p3's were great cpu's, then came the 423 p4's with rambus memory *shudder* I think the whole push for the GHZ with intel totally shafted their engineers. I mean all the way up until the p4's intel totally dominated amd, then came the whole rambus fiasco (not to mention the socket winthin a socket) and amd finally had a shot. now we're seeing intel finally having hit the GHZ wall (which is odd cause they said no 4GHZ cpu would be released and anyone with a 3.6 can easily hit 4.5 or better) we're seeing that old intel who slowed down and waited to release new cpu's till they were truly better.

It's a healthy prospect to know that the new Intel Core and Core Duo (that seem to be a combination of P3 Xeon technology and some P4 cache and other tech. + 64 bit(!)) would get the better of AMD in the near future. AMD has screwed the consumer in the long term as they tampered with falsified PR ratings and low tech cache, etc... IMHO, AMD evolved, but in a slower pace than it should, only because the P4 sucked so much. If INTEL would of kept their focus on enhancing the P3 line into upgrading and adding to that, they would of kept the crown. Now it's catch up time, but I feel Intel will have one huge advantage. If I'm not mistaken, the yet to be released Core architecture will run on 4 execution pipelines, as opposed to 3 execution pipelines for P3 & AMD rig. This means that whatever AMD comes up with, the Core will always be about 25% faster at same Ghz Who's laughing now! The only way I see for AMD to get back in the race would be to do the same and equip their Athlons/Opterons with 4 pipelines and that would break it even. Still, that would be hard to do I heard, as changing execution pipeline architecture means redesigning the whole CPU layout