K10 demo machine in Akiba

[tictac]

hei.. if the nb speed are not connected to cpu speed.. the memory speed wont change if we overclock the htt speed right.. so imagine 1.9ghz processor with only 1.4ghz nb with memory ratio set at 200mhz that will only give 155mhz memory speed.. so the memory speed is underpower right?

[doompc]

MEM Clock is still connected to CPU Clock:
http://222.151.152.227/c-board/file/...32-Opt2347.png

What isn't clear is where the NB Clock come from.
There might be another multiplier setting.
(lol 3 multipliers: CPU, HTT and now NB)

[Bigchrome]

so almost 10 seconds slower than my 4 year old mid range A64....That's good improvements anyway.


ing NOT!

[SOLDNER-MOFO64]

so almost 10 seconds slower than my 4 year old mid range A64....That's good improvements anyway.


ing NOT!

If even for a second you, or anyone else for that matter imagine this scenario to be correct then you need serious help.

[SnipingWaste]

I think ppl are forgetting that AMD ES CPUs are not as good as the retail CPU's. Look at the A64 benches with the 3100+ A64 ES CPU's that ran at 1.8gig with 1 meg L2. It could only run with T2 and not T1 and the memory controller would error above DDR333. The retail 3200+(2.0gig) running at 1.8gig out performed the 3100+ and I think we will see the same thing with the K10. It would be nice to see some benches off a retail K10 BA or up stepping and see the performance different s.

[KTE]

If even for a second you, or anyone else for that matter imagine this scenario to be correct then you need serious help.

SOLDNER... SPI doesn't bother me but what do you think K10 1.9GHz or 2GHz time will be for a 1M?

[kyosen]

If even for a second you, or anyone else for that matter imagine this scenario to be correct then you need serious help.

SOLDNER... SPI does bother me then what do you think K10 1.9GHz or 2GHz time will be for a 1M?

[leoftw]

Why test it on Vmware ? Thats a horrible idea IMHO .


Here are my quad opteron tests

On 32bit Vista and 2.6 Quad Optys , 4gb DDR2 667 ECC

[knightwolf654]

those are duel cores....

[macci]

NB (inside the CPU, the memcontroller) has its own multiplier
CPU has its own multiplier (for each core that is..)
HT bus has its own multiplier (which can not exceed the NB multiplier)

Memory is connected to HT ref. clock (but not to overall CPU clock anymore), you have 200,266,333,400 and 533MHz... memclk modes.

Memory clock goes up when you adjust HT ref. clock (e.g. if you boot at 200 ref. clock / 400MHz memclk and then increase the ref clock to 205MHz your memclk will be 410MHz (DDR2-820)

since the memclk is driven by the ht ref.clock we dont get wierd memclks with certain multipliers (e.g. 375MHz with x15 in case of K8)

memory speed readings w/ CPU-Z (at least ver 1.41) were probably incorrect (as it assumed memclk is still done the K8-way)

[KTE]

SOLDNER... SPI does bother me then what do you think K10 1.9GHz or 2GHz time will be for a 1M?

....

[doompc]

@ macci, with MEM Clock driven by HT Ref Clock we have 1:2, 3:4, 2:3, etc dividers again?

[leoftw]

those are duel cores....

its two dual cores , not one

[Motiv]

SOLDNER... SPI doesn't bother me but what do you think K10 1.9GHz or 2GHz time will be for a 1M?

From the benches I've been priveledged enough to see the results of ()a 2ghz K10 will score around a 19sec 1M Pi time......at 3ghz approx 12secs.

Think he may of been of with his privileged information. Making claims like that always comes back to bite you on ya ass...

http://www.xtremesystems.org/forums/...&postcount=525

[KTE]

Hmm.. well only he can answer what he saw and didn't.

The only idea I have so far is by what the guys around online have done: ~40-49 sec at 1900MHz. Seems high but the CPUs should be out soon and we'll know for sure the rough range.

[tictac]

NB (inside the CPU, the memcontroller) has its own multiplier
CPU has its own multiplier (for each core that is..)
HT bus has its own multiplier (which can not exceed the NB multiplier)

Memory is connected to HT ref. clock (but not to overall CPU clock anymore), you have 200,266,333,400 and 533MHz... memclk modes.

Memory clock goes up when you adjust HT ref. clock (e.g. if you boot at 200 ref. clock / 400MHz memclk and then increase the ref clock to 205MHz your memclk will be 410MHz (DDR2-820)

since the memclk is driven by the ht ref.clock we dont get wierd memclks with certain multipliers (e.g. 375MHz with x15 in case of K8)

memory speed readings w/ CPU-Z (at least ver 1.41) were probably incorrect (as it assumed memclk is still done the K8-way)

Ok now we wont have weird memory speed with the new architecture...

Hei... are you hidding something from us?

[Bigchrome]

If even for a second you, or anyone else for that matter imagine this scenario to be correct then you need serious help.

So the results we've Just ACTUALLY SEEN are false? Wheras the ones you claim but have no way of verifying are more believeable. Sorry, I forgot my place for a second.

[uwackme]

Macci, is that true.... the cores have seperate multipliers?

Could be cool, keep temps lower on apps that only use one core, etc.

[LowRun]

Macci, is that true.... the cores have seperate multipliers?

Could be cool, keep temps lower on apps that only use one core, etc.

Hehe, dude this is Macci speaking, makes it as reliable as can be