AMD Shanghai/Deneb Review Thread

[iocedmyself]

Oliverda yes, 1.35 is rated stock though as has been shown, it's a bit erring on the side of caution on AMD's part since these chips undervolt very nicely as well as overclock with little or no voltage bump to some very surprising frequencies.

Yeah GTA IV seems to be buggy i'll admit, but it happened to be what i've been killing time playing latey as well as the fact that it has an in game bench. Though as i was watching task manager during benches on second display i did notice a couple things.

One, during the bench all four cours have the load split pretty evenly across them.

Two, the benchmark cpu utilization report seems to show the MAXIMUM cpu load, not the average load. From what i saw the cpu load was staying between 40 and 50%, largely only jumping up to the reported load a few times during the benchmark, so something to take into consideration. But yes i need to bench some other games, and have been working towards reinstalling everything to that end

Zucker, i am very very pleased to be able to provide these surprisingly delightful results, thanks for the interest and to everyone else as well for the enthusiasm and interest in the performance of the latest from AMD. Ill admit i've been less than unbiased towards AMD in the past but i'm trying to stay objective here and let the results speak for themselves. I also think it important to consider that the phenom II's and dragon platform was not aimed at the performance crown, but more mainstream market with room for us overclocking OCD types or "enthusiasts" if you will. AMD seems to look more at the entire picture in terms of cost, upgrade paths, power consumption, power efficiency as well as performance, and they have shown in my mind anyway remarkable improvements over the Phenom 1 chips all around. The power consumption clockability and performance was even a suprise to AMD's own people from what i've read so their is only room for refinement, these are just the early chips that are popping up with 4-4.5ghz air clocks and 6.2 LN cooled. So 5 with water is looking promising.

Which i plan to start working on very soon. =)

Also pushing past 1.5 on air doesn't seem to show results that scale well, however as others have said these chips adore the cold.


But for now i need to hit my own sleep button as i've been stress tested for 24h now =P Tomorrow i will be finishing up and posting game benchs and start pushing the clock up a bit, tomorrow seems like a good day to bench 3.8.

[noki]

iocedmyself don't forget 3dmark 06 at default resolution pleasee!!!..have a sweet dreams..jeje! Tomorrow seems a long testing day...

[Oliverda]

IOECEDMYSELF

[B]I have only 4 fast questions :

1 ) What is your air heatsink ?B]

He has already answered this question:

My heatsink through out all testing has been TT DuoOrb http://www.thermaltakeusa.com/Produc...1&ID=1748#Tab4

[amdcian]

He has already answered this question:

My heatsink through out all testing has been TT DuoOrb http://www.thermaltakeusa.com/Produc...1&ID=1748#Tab4

Uh oh, File or directory not found Please fix link

[AbelJemka]

http://www.thermaltakeusa.com/Produc...8&ID=1748#Tab0

[ryboto]

(btw 3.0ghz at 1.1v has a TDP of around 55-60w)

How are you estimating that TDP? If that's true, it's going to be neck-and-neck with penryn consumption(assuming you undervolt) at the CPU level, on the system level it ought to be less considering the IMC in the AMD, and the Intel NB's using a bit more power than an AMD.

[informal]

I guess he used the Vcore^3 and 125W for stock TDP as the most accurate way since it takes into account leakage currents too.This way you come near the 60W mark(68W to be precise ,but with VRM efficiency taken into account too,it is near 55-60W for the chip alone).

[AbelJemka]

PII 920
http://www.hwbot.org/quickSearch.do?hardwareId=CPU_1767
http://valid.canardpc.com/show_oc.php?id=452162


PII 940 same user
http://www.hwbot.org/quickSearch.do?hardwareId=CPU_1765
http://valid.canardpc.com/show_oc.php?id=452168


Motherboard high HTT search for PII 920 4GHz overclocking!

[Caveman787]

Did he really have to destroy the the "rated fsb" for high HTT overclocking?

[informal]

920 @3.8Ghz

Great find AbelJemka

@Caveman787

Rated FSB has almost zero influence on performance with K10.Maybe in some extreme case of triple SLI or CF,but even then i think it wouldn't be a bottleneck.What's important is the northbridge/L3 clock speed since it can be higher than HyperTransoprt(and not vice versa).If Deneb can work stable with @2.5-2.6Ghz NB clocks and 4Ghz for core clocks,it will be one tough nugget to crack .

[Caveman787]

Well thanks for informing me off that informal

Didn't notice that till I typed it.

Also I was under the impression that high hypertranport link combined with high nb speed helped increase performance a big deal.

Why the need for 2000 ht if it's not so important then?

[informal]

Because when the chip is rated at 2000Mhz for HyperTr. it means the NB is 2Ghz too ,since HT can't be higher than NB .But NB clocks can be higher than HyperTransport,so here's hoping it will hit even higher with retail chips.SoF already said his chip ran 4Ghz/2.2Ghz for core/NB stable on air.

As for perf. comment,user should try to get the cores and NB(L3) as high as possible.If it means lowering HyperTransport a bit,it's not a biggie,you can still use high HTT for NB clock calculation(ie. 250HTT x10 for NB/L3),while dropping the HyperTransport multi down(HTT x multie=HyperTransport) to much lower value that will allow for higher core/NB clocks . It sounds complicated but it's not .

[SoF]

friend of mine (musicismylife - I oc'ed on G|C with him the phenom 1m record) just posted some information about 300 HTT (benchstable) on Foxconn Destroyer!!! http://www.planet3dnow.de/cgi-bin/ne...?id=1229367623 I say this will be fun

[informal]

Yeah i just wanted to post that link .Thanks ,great info!

Most interesting quote is the Northbridge clocks he achieved (2.7Ghz!!!):

300 MHz reference clock for a Phenom are a very good result, which so far (if at all) only very rarely been achieved. Interessant ist in diesem Zusammenhang auch der erzielte Northbridgetakt von 2,7 GHz. It is interesting in this context also reached the North Bridge clock of 2.7 GHz. Den Screenshot haben wir dabei um die Informationen "erleichtert", die noch bis zum 08.01.2009 unter NDA stehen. The screenshot we have to ensure that information is "relieved" that until 08.01.2009 are under NDA. Das erzielte Ergebnis wollten wir euch jedoch nicht vorenthalten. The result we wanted but you are not deprived.

[G0ldBr1ck]

Hell ya

[demonkevy666]

friend of mine (musicismylife - I oc'ed on G|C with him the phenom 1m record) just posted some information about 300 HTT (benchstable) on Foxconn Destroyer!!! http://www.planet3dnow.de/cgi-bin/ne...?id=1229367623 I say this will be fun

O_O I was going to get one of those but got sb750 instead... TRI SLI and one 280 GTX as phyics card sounds like a benching queen lol can't wait to see that ones vantage score

[LC_Nab]

Stupid google thinks im full of spyware and it wont show me the translated page >.>

[Jazzman]

Rated FSB has almost zero influence on performance with K10.Maybe in some extreme case of triple SLI or CF,but even then i think it wouldn't be a bottleneck.What's important is the northbridge/L3 clock speed since it can be higher than HyperTransoprt(and not vice versa).If Deneb can work stable with @2.5-2.6Ghz NB clocks and 4Ghz for core clocks,it will be one tough nugget to crack .

I am embarrased to say I did not know this! Thanks informal !!

[Oldguy932]

If we can get around 2.6ghz nb on air I'm going to go out on a limb and say we should be able to get 3ghz with cold, but the cpu clocks will be very modest if we hit that fast. Ht link does need to be dropped at higher htt frequencies, but I've never noticed any gain beyond dropping to 6x with current phenoms. The 9850 I put under cold did 275 htt with a 10x multi on everything, but the cpu clocked better but wouldn't do nearly that much htt on air.

[Ghostbuster]

Look what THG brought in We Overclock Phenom II to 4.9GHz plus result (SuperPi)

[Daveburt714]

Cod... I hate SuperPi as a benchmark in the Intel vs Amd wars!

I remember all the P4 guys bragging over SPi results against my 3800 X2 back in the day, even though that was one of the few benches they could actually win....

That being said, it's nice to see the PhII break the 1M run at under 18 seconds (and to see some real numbers)!

If you'll notice there are a few problems with that SS:

1) Actual clock is only 4.02 Ghz
2) Voltage is Rediculously high for what we've seen from 4Ghz PhII's
3) Memory is set at 800 NOT 1066
4) IMC/NB is running @ 1800Mhz

I'm sure all the Intel fanboys will see this SPi score as a weakness of the PhII, while I see it as very positive. In the right hands, on AIR, I could see this score being sub 16s... We'll see where real world stuff goes from there!

[Jerry Rob]

Cod... I hate SuperPi as a benchmark in the Intel vs Amd wars!

Intel has advantage in SuperPi regarding architectural differencies.
Bigger problem is that Super Pi is single threaded app, it shows almoust nothing about perormance of multicore CPU.

sapienti sat

[Blaber]

THG
the best S H I T site all over the world, his dipendents can't OC , and it's PRO intel
IMBESIL !!!!

+1 and also add ANANDTECH TO THE LIST PLEASE, they all suck INTEL'S BALLS

[AbelJemka]

Look what THG brought in We Overclock Phenom II to 4.9GHz plus result (SuperPi)

|quote]To verify that the processor was stable at 4.9 GHz, we conducted some tests and it completed a SuperPi 1M run in 17,769 sec.[/quote]
Screenshot is 4GHz...
Moreover 17.769s 4G Phenom II
Best Phenom I score 19.92s at 3.9GHz with one core overclocked and memory at 558MHz cas 4...
http://www.hwbot.org/quickSearch.do?hardwareId=CPU_1665

[G0ldBr1ck]

you gotta love the crap these sites pull. We can clearly see that the 17 sec. run was at 4.0 but trust me, there will be others that will choose not to see.