http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2771&p=1
I know what chip my next cpu will be based on
I know what chip my next cpu will be based on
anandtech gets a hold of a 2.93ghz conroe...
- Thread starter Moloch
- Start date
The performance difference is not as large as was shown in the Intel controlled preview. 20%
from a 2.9Ghz conroe vs 2.8Ghz athlon fx as opposed to almost 40% from a 2.66 conroe versus a 2.8 Afx. All AMD needs to do is drop their prices and they will still remain very competitive with the Conroe. It's amazing how a now old a64 design still keeps up with intels latest.
from a 2.9Ghz conroe vs 2.8Ghz athlon fx as opposed to almost 40% from a 2.66 conroe versus a 2.8 Afx. All AMD needs to do is drop their prices and they will still remain very competitive with the Conroe. It's amazing how a now old a64 design still keeps up with intels latest.
Yes it's just using "standard" memory but on the other hand you have the same hardware on both systems. Before this all tests used closed systems and you didn't know that they were using the same hardware on both systems. It is possible that the conroe systems gets a performance boost on other motherboards or later bios revisions so the Intel demos might not be that incorrect. Even though these results are quite a bit lower than those shown before it still rocks.
EDIT: Spelling.
EDIT: Spelling.
Bajzel said:
I'd call 50%+ better in FEAR for minimum frame rates pretty damn impressive. Minimum Framerates are after all the most important to an enthusiast fanatical gamer of any type. Of all the games they tested FEAR is certainly the most CPU bound so far. Oblivion and HL2 will be great tests as they mentioned since both games have serious CPU limitations as well. In otherwords AMD may be in just as much trouble as people thought they were with those prelimilary benchmarks. We need more high end CPU limited games to prove it. Minimum framerates if possible. I dont know how anyone could look through that though, thats seriously impressive. 50%+...wow.
Xentropy said:
Latency timings dont do much at all to impact AMD 64 chips performance....it really doesnt matter because of the onboard mem controller. Its only a real problem for Intel chips to try to keep good timings.
Last edited by a moderator:
I'm not to sure about that.. did you see the memory latency they have?
Only a tad short of amd.
Truely amazing imo.
They don't have all the draw backs of an onboard MC and they still have awesome latecy.
No changing sockets for memory type changes and no "half assed" (tech report) ddr2 800 implementation for all but the highest clocked chips.
I was truely shocked when i saw amd didn't run ddr2 800 at full speed for all speed grades and transistors used up by it.
Also digit-life shows amd's ddr1 platform is showing better memory performance in some of their tests, and not be a small margin is a few, also intel's P4 is showing better peformance as well.
http://www.digit-life.com/articles2/mainboard/ddr2-800-am2.html
Only a tad short of amd.
Truely amazing imo.
They don't have all the draw backs of an onboard MC and they still have awesome latecy.
No changing sockets for memory type changes and no "half assed" (tech report) ddr2 800 implementation for all but the highest clocked chips.
I was truely shocked when i saw amd didn't run ddr2 800 at full speed for all speed grades and transistors used up by it.
Also digit-life shows amd's ddr1 platform is showing better memory performance in some of their tests, and not be a small margin is a few, also intel's P4 is showing better peformance as well.
http://www.digit-life.com/articles2/mainboard/ddr2-800-am2.html
radeonic2 said:
It's the same with DDR1 on K8. You always run a few MHz below spec because of what looks like "math precision" on the dividers I'd guess. For example, a K8 running PC2700 runs the RAM at like 160 instead of 167. It doesn't matter in the least if you're an overclocker, and the performance difference is probably so negligible otherwise that it doesn't matter.
I'm personally not at all surprised with what Conroe can do. It's a rethinking of Pentium M without the need to run super low power. They couldn't really lose, IMO.
AMD should be ashamed for not having reworked their core for years, essentially. The DDR2 K8 is basically the same as what arrived in 2002. And, K8 is basically an Athlon from 1999 with some tweaks and AMD64. All that really makes it faster is the memory controller. It's more like a Pentium M, which is a Pentium Pro with tweaks. And Pentium M / Core 1 can keep up with it in most real-world apps (and use a ton less power).
Conroe is truly a new design with tons of bottleneck breaking going on.
Blazkowicz
Legend
AMD's integrated MC doesn't prevent from being hurt by DDR2's added latency. but I think it's well overplayed, I can get 2GB of gskill DDR2 667 at 4-4-4-12 for 175 euros, while PC3200 2.5-3-3-6, near equivalent latency-wise, is at 185 euros.
also, with intel you don't need a new CPU when RAM type changes, but you need a new mobo for each half-generation of CPU.
still, AMD will be in trouble but that's Intel beneficing from its smaller process rather than continue wasting it on Netburst. which pisses me off as I'm used to Intel being more expansive and slower. at least AMD will still have biggest gaming bang for the buck (sempron 1.6ghz with big easy o/c)
also, with intel you don't need a new CPU when RAM type changes, but you need a new mobo for each half-generation of CPU.
still, AMD will be in trouble but that's Intel beneficing from its smaller process rather than continue wasting it on Netburst. which pisses me off as I'm used to Intel being more expansive and slower
. at least AMD will still have biggest gaming bang for the buck (sempron 1.6ghz with big easy o/c)Actually the Bad Axe boards (a 975X revision that just started) seems to be the first decent chipset intel has put out for awhile. Xbit is reporting that because it will support Conroe it can theoretically support the first run of Quad cores in 2007 due to having the same power requirments. Although Intel may screw people which they have been known to do like when they introduced 1066FSB and said the 925 was incompatable which was a lie. If they try to push DDR3 adoption by early 2007 that may complicate things as well.
Well, it seems rather likely that the whole reason that Intel has changed their platform so many times in recent years is that they want to maximise their sales, and that includes chipset sales. This makes me really hope that AMD continues to not supply Athlon chipsets.SugarCoat said:
EasyRaider
Regular
No, it's because the divider must be an integer. Example: With 2600 MHz CPU frequency, the lowest useable memory divider is 7, which gives 371 MHz memory frequency, quite a bit lower than 400.swaaye said:
So you can get exact memory freq., but only when CPU freq. is an integer multiple.
Last edited by a moderator:
The_Wolf_Who_Cried_Boy
Newcomer
That review seems very dubious to me, at least in the measurement of the difference between 2GB and 4GB. There's absolutely no reason whatsoever why there should be a difference between 2GB and 4GB in Doom3. This is particularly true if they were running in 32-bit Windows, which seems very likely.
Shogun said:
I never said the difference between 1T and 2T command rates was nothing. I said Latency timings, which that proves. Is your head going to spin off if you lose 2fps on loose timings when you're already doing doing 100+?
Speaking from the position especially as someone who likes to overclock, it is well known in the OC community that loosening the timings (within reason that is) to overclock the chips doesnt hurt performance much at all. Likewise at stock it doesnt do much either like when for example if you wanted to cram more memory into your system.
As far as 1T and 2T command rates go, speaking from experiance on older Athlon 64 systems, i'll take 2gig and the 2T command rate over 1gig and the 1T command rate any day of the week. There may be a small performance hit but the overall system performance is improved and much quicker. Especially for loading.
The_Wolf_Who_Cried_Boy
Newcomer
I suppose my previous effort could have been a little more verbose.
(Hopefully Legit don't mind hotlinking)
This chart is what stuck in my mind, I assumed you were included command rate in your statement as part of the generalised timings parameters.
Nevermind, I get confused easily in my old age....
(Hopefully Legit don't mind hotlinking)
This chart is what stuck in my mind, I assumed you were included command rate in your statement as part of the generalised timings parameters.
Nevermind, I get confused easily in my old age....
Blazkowicz
Legend
I didn't notice the 5-5-5-15 numbers were run at 2T, would have made more sense to run at 1T and the picture would have been less dramatic (while 4-4-4-12 probably still being the sweet spot)
EasyRaider
Regular
It's true for all current K8.darkblu said:
And don't confuse "CPU multiplier" with "memory divider".
memory clock = reference clock * CPU mult. / RAM div.
Last edited by a moderator:
