Strange AMD DualCore Performance FX62 Versus FX57

[East17]

How come the DualCore version of the same architecture with the same clockspeed gets lower results than the Single Core version ?





YES I know the benchmarks are in the SingleThreaded version but hey ... it's the same task on the same CORE @ the same clockspeed .

Link

[dnd]

Logic latency from the memory controllor deciding which core on the FX62 should process the data?

[TheGlasMan]

Looking at the system specs the AM2 was using 667 memory at 4-4-4, don't know what the FX 57 was using. But fairt to say someting wrong with the benchmark. they may have explained it in German, but that doesn't do us any good (translations cannot be trusted)

[_dangtx_]

think 800mhz ddr2 benches to ddr400 so that could be it indeed

[Pehu]

The article credit the bad performance of the AM2 chips on the higher latency memory used. However, they can not explain the bad AM2 performance of the 'multi-threaded 7-zip'-test and simply say 'There is something wrong with this test.'

That article is simply worthless: They actually only had an AM2 4000+ processor available and used different overclocks on that chip to simulate a 'virtual FX62, etc.', so none of the results actually show the true performance of any other chip, other than said AM2 4000+ (which uses DDR2-667 memory btw).

[PF Prophet]

Pehu, u can use pretty much any ddr2 with the am2 chips, most peoples problems with ddr2 come from not setting proper voltage b4 they set the timings.

and as long as the clocks the same there really isnt any diffrance in perf between an fx chip and an x2 or normal a64, unlike intel chips the extra cache on amd chips is almost worthless, thats why from what i have read amd is removing 1mb cache chips and moving to 512 as the upper end and 256 as normal, so 256+256 x2's should be on the way.
There are slect apps that benefit from more cache, folding can be one of them, as can things like superPI, but in real world apps and games the diffrances are not worthy of note, so there review is valid imho, they show the aprox perf of a chip FAR higher clocked and FAR more expincive.

who has that kinda $ to spend on a cpu?????
nobody i know, sure i know 2 ppl with quad opti setups but thats more then a cpu and they both have 3dlabs 640mb video cards as well, but they do high end gfx work so they need the power, or so they say :P

but back to pehu, Saying that this isnt valid because they used overclocks is silly, its like saying that my 3000+@(400mhz oc)2400mhz was slower then my 3700+@2400mhz(stock) when in reality the perf diffrance dosnt excist other then on paper, oh and in super pi, tho i think part of my diffrance there is that i now have a64 tweaker

[Pehu]

but back to pehu, Saying that this isnt valid because they used overclocks is silly, its like saying that my 3000+@(400mhz oc)2400mhz was slower then my 3700+@2400mhz(stock) when in reality the perf diffrance dosnt excist other then on paper, oh and in super pi, tho i think part of my diffrance there is that i now have a64 tweaker

The multipliers needed to simulate the processors wanted on the 4000+ simply do not exist on that proc. So, instead they opted to fiddle with the bus clocks and have had to reduce e.g. memory clocks to get those 'simulated' results. They aren't valid and that's a fact.

[PF Prophet]

changing memory devider has been showen many many many times to have no negitive effect on a64 systems, unlike older chipset based memory controlers the a64's dosnt suffer the latancy issues assotaited with older chipsets.

as i have had to explain over and over to a few people i know its like compairing onboard sound NOW or compairing soundstorm audio to ac97 audio in 97/98 when its sound quility was poor and it used alot of cpu time/power, those assumptions are nolonger valid, the diffrance in latancy is not an issue anymore as long as you keep the devider setup so that your ram is running at aprox he same clock as it would be stock on another cpu.

this is easy to do in my exp, my 3700+ is at 305x8 insted of 12x200, so my ram is slitly clocked and the cpu is slitly over stock. but memory perf isnt any worse or better then if i lowerd the fsb and ran it 1:1 with cpu with higher multi, again due to the fact that a64 dont suffer a perf hit using deviders vs 1:1.
intel on the other hand suffer more if you run async due to chipset based memory controler use.

as i said its a valid way to test to see how diffrent clocked cpus will perform, i personaly laught at ppl who buy extreamly pricy cpus thinking its the only way to get that level of performance.