AMD is back? Phenom II@3D06

[AndreYang]

Nice andre!

Which bios did you use?

I got new bios which is multi to 39.5. If you want to try, you should pm me your email address.

[chew*]

Andre whats the max suicide you can do on the DFI, also whats the max 3dstable untweaked? If you can say. How's the digital PWM react to the cold?

[FatAlbert]

can you send to me too ?

did try spi on this speed ?

[MAS]

does anybody have a proof Phenoms B2/B3 are coldbugged? if there is no proof it will be usefull to try to overclock 9950 BE under nitrogen

[massman]

Most of them are bugged; some at -10°C, some at -50°C. It's always nice to see someone wrestle with the coldbug :-).

[SeLecT]

I got new bios which is multi to 39.5. If you want to try, you should pm me your email address.

Thats very nice! Thanks!

You've got PM

[knightwolf654]

holy smokes!! that is awesome!!!

[imsochobo]

I maxed out the Dryice on my 9850BE at 52- and maybe even more!.

STill didnt allow higher clocks, 3600 mhz is my limit, air, water Dice, doesnt matter, clock just the same =/ worth noticing, -52 @ 2000 mhz HT i think it was, booted first on ALL stock.

acc settings doesnt allow it higher, does 3 ghz 1.25 V =) so its a good 24/7 ocing chip.

Anyways, check this HT link scaling http://www.madshrimps.be/?action=get...54&articID=895

Sure you cant get that HT-link up a notch ? same freq and run 3dmark and check the scaling then and score, maybe 30K =) my bet though.

to 4x lower temp, 988 HT link might do it, youre just running 3x at the moment. 3x247.

wellwell, youre chip might be awfully alot coldbugged(on the ht link) i dunno.

The amd overclocking is going to be, finding an chip which first of all clocks high, but also can have an high HT link at those temps needed for those freqs. for the 3dmark benchies though, could be interesting.


***Btw, why do you run the memory in ganged ? havnt you learned about ganged/unganged ? read it up if you havnt, my facial expression was an ? at first here so :p great understanding if you havnt understood it yet.
It doesnt change the 3dmark score more than 10-20 max, but gives an 2 gb/sec more bandwidth. (At max)

Phenom 9850 BE 2000 mhz HT 3300 mhz 4870CF 3dmark06
Phenom 9850 BE 2200 mhz HT 3310 mhz 4870CF 3dmark06

10 % change of HT.
0.33% change of clock speed.
2% change in total score.

Youre score should be :
1% of 29581 is : 295,81 * 2 = 591,62 + 29581 =
30172 if the scaling is the same, if ran @ 988 mhz HT link. my guess: just over 30k.

but... math and theory is 2 diffrent things, but madshrimps link up shows 1:1 scaling much across the board =)

***EDIT: 1:1 scaling on cpu clock, not on ht-link, gives roughly ~10-20 % depending on the application/task on ht-link.

[MAS]

graphics performance is no doubt restricted by low HT speed. try to achieve at least 1200-1500 MHz

[drizzt5]

how do the new phenom II's perform with superpi?

[informal]

how do the new phenom II's perform with superpi?

Better than Agena by some 12% per clock,but still lower than C2Qs.SuperPi is one of those little programs that prefers intel's low latency legacy fp instructions+huge cache mix.Oh and superPi is really slow at calculating Pi .

[nachthymnen]

Very nice!!
Finally AMD seems to be back
And finally no CB problems

[imsochobo]

dont say superpi stuff, its just a total waste for comparison amd-intel.

Superpi likes, much, fast low latency cache, which phenoms doesnt exactly excell at either.
Their L3 cache isnt exactly fast to say it mildly, it got boosted quite abit in the phenom 2.

But for the HT link, and all that, what i think is that its gonna be an battle for highest clock with high HT link for 3dmarks.
Ive seen 2100 mhz Ht link at -110 so it is possible with good results! it was "only" 5100 mhz too.



If you can go way down on the temp, and way up on the HT link andre =).

and check up unganged, ganged memory, and NB speed always have to be higher than HT link or equal.

and thanks for the results, makes me want the phenom 2 even more

and also... seems like 4 dimms is best for superpi aswell.

for 6G run 1.95 Volts or maybe even higher! but test ht links first, might even boost score more than some mhz

That is what other say though, my experience is solely based on temps @ 70+ and down to 50- and not LN2. and not even PH2, but they should behave the same when it comes down to memory, ht links and NB clocks (not exactly, PH2 tend to reach higher HT link speeds.)

[demonkevy666]

graphics performance is no doubt restricted by low HT speed. try to achieve at least 1200-1500 MHz

Hyper transport is more for the HDD and low subsystems. .

higher core speeds will over come this anyways

[mAJORD]

Better than Agena by some 12% per clock,but still lower than C2Qs.SuperPi is one of those little programs that prefers intel's low latency legacy fp instructions+huge cache mix.Oh and superPi is really slow at calculating Pi .

dont say superpi stuff, its just a total waste for comparison amd-intel.

Superpi likes, much, fast low latency cache, which phenoms doesnt exactly excell at either.
Their L3 cache isnt exactly fast to say it mildly, it got boosted quite abit in the phenom 2.

hehe sorry, I don't buy the cache argument for why AMD K8-K10 gets its butt handed to it in super pi..

Nehalem has very similar cache structure to K10.. though lower latency/faster it does NOT explain the massive IPC difference in this proggie.. esp since i7 is notably faster than conroe/wolf ..

there's obviously some other reason.. Love to know what it is exactly. Hell even 512kb L2 Celeron beats Deneb at it with no IMC not to mention Yonah even with 2mb cache, a 3 issue core and no marco-ops fusion Cache obviously has a lot to do with the scaling of the benchmark, as evident by celeron (conroe-L) vs 4Mb conroe, but it's how the cache is utilized that has a large bearing on it.. perhaps in combination with another unrelated factor

Almost worth starting a new topic over IMO.. It's an intriguing topic due to the populariy of the bench..

[demonkevy666]

hehe sorry, I don't buy the cache argument for why AMD K8-K10 gets its butt handed to it in super pi..

Nehalem has very similar cache structure to K10.. though lower latency/faster it does NOT explain the massive IPC difference in this proggie.. esp since i7 is notably faster than conroe/wolf ..

there's obviously some other reason.. Love to know what it is exactly. Hell even 512kb L2 Celeron beats Deneb at it with no IMC not to mention Yonah even with 2mb cache, a 3 issue core and no marco-ops fusion Cache obviously has a lot to do with the scaling of the benchmark, as evident by celeron (conroe-L) vs 4Mb conroe, but it's how the cache is utilized that has a large bearing on it.. perhaps in combination with another unrelated factor

Almost worth starting a new topic over IMO.. It's an intriguing topic due to the populariy of the bench..

try cache ways mostly, sizing plays a small role.
AMD cache hasn't changed since K10.

IHO personally would recommend they either make it bigger and/or make more associations. threw me a bone here but there also a lack of SSE1,2,3,4 pre-core.

[mAJORD]

associativity is usually a product of cache size though.. larger cache, higher associativity is required.. So it's not really a 'seperate' factor to cache size..

If you look at 512kb K8 Vs 1Mb K8, Pi times increase by approximatly 5%..

If you took the L3 away from K10 completely, (propus.. we need you for this discussion [that belongs elsewhere.. sorry ] ) and bumped its L2 to 2Mb, it would still fall well behind Yonah, which for the record, does spi 1M in about 23 sec @ 2.7ghz despite all the building blocks of the architectures being either equal or superior for the K10

[demonkevy666]

associativity is usually a product of cache size though.. larger cache, higher associativity is required.. So it's not really a 'seperate' factor to cache size..

If you look at 512kb K8 Vs 1Mb K8, Pi times increase by approximatly 5%..

If you took the L3 away from K10 completely, (propus.. we need you for this discussion [that belongs elsewhere.. sorry ] ) and bumped its L2 to 2Mb, it would still fall well behind Yonah, which for the record, does spi 1M in about 23 sec @ 2.7ghz despite all the building blocks of the architectures being either equal or superior for the K10

5% is small

4 way 32K cache vs 2 way 64K cache.

[MAS]

Hyper transport is more for the HDD and low subsystems. .

HT is an interconnect between pci-e controller and cpu

[imsochobo]

speed have been bumped slightly from phenom1-phenom2.

if phenom 2 had 6 mb of L2 it would be much better than the current config, just in superpi and single threaded applications.

numbercrounching isnt exactly what we need, all the reasons why amd perform bad in superpi is unknown for me, but i dont care, most useless benchmark by my oppinion.

and that HT link affect hdd and all that, ERR wrong, its between PCI-E and chipset->cpu vise versa, and there is huuuge amounbts of traffic.
cpu cores communicate with eachother. so the ht link isnt some fluffy thing for some subsystem performance.

[dinos22]

there's obviously some other reason.. Love to know what it is exactly. Hell even 512kb L2 Celeron beats Deneb

how many more GHz will AMD need to beat this Celery....erm yeah

[mongoled]

Come on guys please stay on topic

[drizzt5]

Better than Agena by some 12% per clock,but still lower than C2Qs.SuperPi is one of those little programs that prefers intel's low latency legacy fp instructions+huge cache mix.Oh and superPi is really slow at calculating Pi .

Thx, 12% clock for clock right? This doesn't take into account how well they OC I hope.

dont say superpi stuff, its just a total waste for comparison amd-intel.

I know that, but I thought that it would be a good comparison to past amd cpu's.


And I've asked this before I think but why don't they make a super pi equally suited for both AMD and Intel? Or is that not possible.

[demonkevy666]

how many more GHz will AMD need to beat this Celery....erm yeah

post one with the cache way association screen too

btw what ram are you using 1400 mhz DDR3 ?

limited to ddr2 on AMD 1,200mhz wouldn't not be hard to do with some low voltage sticks.

I've got evo one's will do 1,100mhz @ 1.98 volts( 6-6-6-24-(31) ), on my phenom 9850. my sticks won't do 1,200mhz though

tight timings doesn't help K10 at all.

[chew*]

Dino that was 13 minutes of your life you will never get back