More AM2 Benchmarks

[jlbenedict]

Originally posted by: Brunnis

Originally posted by: jlbenedict
only when DDR2 speeds are ramped up, will AM2 take advantage...
from what I understand, DDR2-800 is about equal to DDR1-400

No, that is not the problem here. Even Intel's DDR2-533 platform with a 1066MHz FSB CPU beats this AM2 system with DDR2-800. SiSoft Sandra doesn't care much about latency, it measures bandwidth. DDR2-800 in dual channel mode has a theoretical bandwidth of 12.8GB/s and it's certainly not some inherent problem with the DDR2 memory that limits the bandwidth here. The system is given 12.8GB/s to work with and can make use of less than 50% and that indicates a seriously inefficient implementation.

Yeah.. i'd say that is inefficient

 

[leexgx]

:frown:
just to note None of these tests are official realy as its an Test CPU the DDR2 800 contorler was broken on that test chip (was running more like ddr2 400 speeds@ 5-5-5-12 even tho it was set to 800mhz) so there is no point in bashing it or loveing it untill we get an real one, when both come out then we can see if the new cpu is better or not (amd and intel)

do not go off these tests like tomshardware states its an test chip an old one as well (same chip every one has had)

from what I understand, DDR2-800 is about equal to DDR1-400

DD2 is about 2x the speed of ddr1 (not the other way round DDR2 400 will out perform DDR1 400 in bandwith tests)

some one post linkys about ram DDr types

 

[robertk2012]

Originally posted by: leexgx
:frown:
just to note None of these tests are official realy as its an Test CPU the DDR2 800 contorler was broken on that test chip (was running more like ddr2 400 speeds@ 5-5-5-12 even tho it was set to 800mhz) so there is no point in bashing it or loveing it untill we get an real one, when both come out then we can see if the new cpu is better or not (amd and intel)

do not go off these tests like tomshardware states its an test chip an old one as well (same chip every one has had)

from what I understand, DDR2-800 is about equal to DDR1-400

DD2 is about 2x the speed of ddr1 (not the other way round DDR2 400 will out perform DDR1 400 in bandwith tests)

some one post linkys about ram DDr types

I dont think so. Neither of you are right.

 

[KBM]

I don't think chipset will make any difference..
Since AMD uses onboard MC, chipset would have no effect..

 

[Artanis]

Switching to DDR2 may only help the future AMD quad/multi-core CPU's.

 

[Duvie]

Originally posted by: Artanis
Switching to DDR2 may only help the future AMD quad/multi-core CPU's.

This is what I believe as well...

DDR2 now has more to deal with availability of the memory market where iNtel and its huge marketshare has pushed DDR2 to the fore front even if the design is rather lackluster versus a much older DDR standard...

Quad cores will be when AMD chips will start to become starving for bandwidth.....


With AMD and AM2....I give INtel the next 1 year at least as the leader in the desktop market...Quad core 65nm Opterons will be here before INtels and AMD wil continue to make in roads in the more lucrative business sector....

 

[robertk2012]

Originally posted by: Duvie

Originally posted by: Artanis
Switching to DDR2 may only help the future AMD quad/multi-core CPU's.

This is what I believe as well...

DDR2 now has more to deal with availability of the memory market where iNtel and its huge marketshare has pushed DDR2 to the fore front even if the design is rather lackluster versus a much older DDR standard...

Quad cores will be when AMD chips will start to become starving for bandwidth.....


With AMD and AM2....I give INtel the next 1 year at least as the leader in the desktop market...Quad core 65nm Opterons will be here before INtels and AMD wil continue to make in roads in the more lucrative business sector....

It was also thought that the x2s would be bandwith starved. I think ddr2 is just a marketing gimmick otherwise they would release AM2 with the quadcores. I personally dont plan on buying an AM2 until there is a quad core chip. On the other hand I will take a hard look at conroe when it is released. Things can change though ;-)

 

[jiffylube1024]

So far AM2 is completely underwhelming. I think the only benefit we will see is when faster chips ultimately come out on AM2 than were available on S939.

 

[Zebo]

Originally posted by: keysplayr2003

Originally posted by: Zebo

Originally posted by: Amaroque
I'm about AM2. I had much higher expectations. But as I've said before, I will always go with the highest performing platform regardless of manufacturer.

The thing that bothers me most is that Intel can jack prices sky high again if there isn't any fierce competition. Also they won't have any pressure from AMD to innovate. Who benefited from the GHz race?

Remember, the reason we have Conroe today, is because of AMD. Without AMD, we'd all probably still be using P4 space heaters for the next couple few years.

Unless AMD can pull a rabbit out of a hat, I'm pretty sure what my next upgrade will be though.

Dude it's only 20% lead - little in CPU terms - and only 20% in cherry picked benches intel chose to run. You don't think they'd play anything X2 was strong in do you? Also Lets wait and make sure they did'nt "accidently" have a 3.6 Ghz conroe in the box reviewers wernt allowed to open before saying the sky is falling.

I think you guys will be surprised once reviewers get them and a full battery of tests is done. I think Conroe will lead by 10% clock for clock - you heard it here first. Nothing exactly earth shattering.

Ummm, didn't you just buy some Intel stock? Forgive me if I'm mistaken, but if you did, this last comment from you does not have a consistency with your confidence in Intel as indicated by your stock purchase. Again, if I am mixing you up with someone else, apologies.
Keys

Not saying I hav'nt thought about it but Yes you are mistaken. If anything I'd bet against AMD but Charles convinced me this is not a good bet either.

You really think intel will lead by 20% or more in every test? Come on Intel ran the damn test and would do nothing to show AMD is positive light..

There are many possibilites intel could have used to juice.

1.The most unlikly - Intel outright cheated - many ways to do from as subbtle as putting a raptor in Intel and a old 5400 RPM drive in AMD to outright overclocking intel chip and re-programming CPU identifiers.

2. More likly - Inel just chose tests which thier CPU was most dominate in - yeilding 20% dominance over X2 - negecting anytthing AMD was strong in

3. Somewhat unlikly - Custom drivers to help intel and hurt AMD


I have said for a year or more now Conroe will whoop AMD X2 and my thoughts hav'nt change but you may be surprised to find lead was'nt as large a lead as Intel has you thinking. Because I know Intel - would stab mother in back for a nickle. And Marketing types would stab her twice just to make sure she could'nt fight back. AMD is mother.

 

[Furen]

Originally posted by: robertk2012

Originally posted by: leexgx

from what I understand, DDR2-800 is about equal to DDR1-400

DD2 is about 2x the speed of ddr1 (not the other way round DDR2 400 will out perform DDR1 400 in bandwith tests)

some one post linkys about ram DDr types

I dont think so. Neither of you are right.

Yes, they're both wrong. DDR2 800 latencies can be compared directly to DDR1 400 latencies, so at 5-5-5 DDR2 has about 250% the latency as DDR1 at 2-2-2. DDR2 has twice the prefetch as DDR1 (4 vs 2) but this is taken into account in the "effective" clock rate, so DDR2 400 has the same theorethical bandwidth as DDR1 400 but with higher latencies (which leads to a lower effective bandwidth).

Of course, if you want to do a proper comparison you have to look at both prefetch and latency, so DDR2 800 has twice the prefetch (theorethical bandwidth) at 250% the latency. Most people tend to say that A64s prefer low latencies over bandwidth, but this is in single-core mode. Having two cores requires you to be able to feed data into both of them, so the benefit from increased bandwidth may be higher with the X2s.

By the way, the efficiency of the memory controller has nothing to do with the chipset, since the mem controller is on-die chipset manufacturers have almost no control over this except to tell the CPU what settings to run at. I suppose the problem could be that the motherboard is forcing single-channel mode or something of the sort but I'd be more inclined to say that AMD's mem controller (the chip was a week 47, 2005 part, I believe) just plain sucks or that there is a bottleneck elsewhere within the CPU (the SRI perhaps? I highly doubt this since it would be too glaring of an error to make...).

 

[Zebo]

Most people tend to say that A64s prefer low latencies over bandwidth, but this is in single-core mode. Having two cores requires you to be able to feed data into both of them, so the benefit from increased bandwidth may be higher with the X2s.
-------------------

Theory is nice but experimentation tells us TWICE the bandwitdh in socket 940 dual CPU got beat by cheap 939 dual core solutions with half the bandwith. Why is primarly the 3-3-3 timings and extra wait state of EEC ram compared to LL 2-2-2 1T 939. This is why for a year now I've not been exactly going out on a limb predicting DDR2 will actually hurt AMD unless timings get way down there (well plus Intel did worse initially too vs LL DDR) We don't just say it - we see it though testing by others and ourselves A64 loves LL much more than bandwidth.

 

[Furen]

Hah, a 2 socket vs dual-core comparison is hardly fair. Accessing another CPU's memory through HT more than doubles the actual access latency (not the module's latency, which is usually much lower than the access latency) which pretty much kills any performance improvement you might have from the doubled bandwidth. Also, cache-coherency traffic latency would be massive compared to dual-core.

 

[Zebo]

You don't like those two reasons? Ok third I found 200Mhz 2-2-2 = 300Mhz 3-3-3 in my own real world testing even when sandra says 2000Pts higher for the DDR600 setup. Or we could just wait for DDR2-667 to show us the dog it is.

Edit: assuming they don't use 3-2-2 DDR2-667 , then it might tie 2-2-2 DDR400.

 

[JackPack]

Originally posted by: jlbenedict
only when DDR2 speeds are ramped up, will AM2 take advantage...
from what I understand, DDR2-800 is about equal to DDR1-400

LOL.

DDR2-800 is already the end of the road for DDR2. There might be one more speed bump to 1066 but not likely, given that DDR3 is well on track

 

[Zebo]

is 800 or 667 default for am2?

If it's 800 good luck overclocking at all since 400Mhz pretty much is limitation of chipsets and mem controller..

 

[JackPack]

Socket AM2 will support DDR2-800 from the start.

 

[Zebo]

Originally posted by: JackPack
Socket AM2 will support DDR2-800 from the start.

Fine but that did'nt answer my question. Will multiplier be setup to support 667 or 800?

More succinctly will a 2400Mhz chip have a 6 or 7 default multiplier.

 

[Furen]

The HTT will be set at 200MHz regardless of the ram you use. Kind of how current E revision chips can run a DDR533/566 divider eventhough they use a 200MHz default FSB.

 

[svsnow]

im happy..now i wont be temped to upgrade..

 

[Viditor]

One thing I noticed that appears to have missed everyones attention...

Go to page 4 at the bottom, and look at the difference in the size of the cache between F and E. I think it's quite possible that AMD has already pushed through the Z-Ram if that picture is accurate. The surface area of the Rev F looks to be almost half that of E.

 

[Gatt]

Are AMD's bandwidth starved?

I thought AMD's had plenty of headroom on bandwidth because they were more sensitive to latency.

The benches I've seen seem to bear that out, If you drop from CAS 2 to 2.5 you've gotta move up quite a few mhz with DDR to make that up. Aren't DDR2's at like CAS 3 or 4? I'd expect a huge chunk of the increased bandwidth to go to making up for the latency loss.

If that is the case, I wouldn't expect much out of DDR2.

As far as Conroe goes, I don't see any reason why they couldn't get a 20% bump out of the chip, assuming they reengineered the things and improved things like branch prediction. Mhz isn't the only way to increase performance, the other options just have finite limits and faster diminishing returns.

I mean, it is Intel we're talking about, they do have top rate engineers. I'd expect them to be able to pull something off.

Either way though, doesn't matter to me, as long as it all translates to faster cheaper chips I'm a happy camper.

Btw, before anyone calls me a fanboy, I'm using an AMD 4200+. I like AMD. I like AMD competing with Intel even better.

 

[Absolute0]

these benchies make me want conroe

 

[dexvx]

Originally posted by: Viditor
One thing I noticed that appears to have missed everyones attention...

Go to page 4 at the bottom, and look at the difference in the size of the cache between F and E. I think it's quite possible that AMD has already pushed through the Z-Ram if that picture is accurate. The surface area of the Rev F looks to be almost half that of E.

Z-Ram is not low latency enough to be used as L2 cache. The boys at Ars thinks it can only be used on L3.

 

[ribbon13]

If Conroe can really pull off 11IPC I'll be damned to not put vapor-phase on one.

 

[Furen]

Hardly Z-ram. Like dexvx said, it's most likely not fast enough to be L2. Also, Z has a much greater density that what we're seeing in rev F K8s. Remember that Intel's cache is extremely dense, so there's other ways to get lower cache density besides going Z.