AMD Shanghai/Deneb Review Thread

[accord99]

that one thread he posted is the proof that all phenoms have a bad temp sensor and are not accurate.

Plus the additional info from the writer of Core Temp:

*K10 = Phenom (Agena), Opteron (Barcelona). The K10 reports a temperature value that is relative to a certain predefined value, it doesn't report the actual processor temperature! So take that into consideration.
**CPU Temp is because the Phenom\Opteron (K10) have only one sensor per package, meaning there is only one reading per processor.

[sundancerx]

well, depends if you say power is watts or power in ampere's.

lets say it like this.
It comsumes more energy.

i dont get it. anyway, 4 more days, we shall better things to discuss

[sundancerx]

no he is just saying how the temperature is measured. as electronics heat up the resistance changes so the probe measures the resistance and can tell what the cpu temp is because of that.

sorry, dont want to drag this on. i understood about the thermistor (did i get this right) being the source of measurement for core temp. what im trying to say is that, he was implying that there has to be wrong about how the core temp is measured inorder to get lower temp even though the voltage is higher. check his comments about core temps being lower than ambient. my point was that, even with accurate core temp measurements, lower temps can still be achieved with higher volts, thus the formula i gave. this is just my point, if it cant be understood, then i rest my case. no need to drag this further. again, sorry if cant express well my message.

[roofsniper]

i think he thinks power uses the units amps and watts when its watts = volts x amps. since the voltage on phenom II went up you can guarantee the amps went down.

[QuietIce]

The sensors themselves are fine as far as scaling goes (with a few, rare exceptions), it's just the index that is off - that always has been the issue on both sides of the aisle. Any monkey can index a temp sensor to within 1-2°C by simply changing the offset - whether it be an Intel or AMD chip, doesn't matter. The Coolest put adjustments in the *.ini file specifically for that purpose. If you're going through all the hours to OC your chip and monitor with Core Temp you'd be a fool not to at least glance at Core Temp at stock idle and do a fast comparison between the reading and your (CPU HS/rad) ambient temps to correct for any serious variation. The tool is there, it takes all of 5 seconds to index the sensor - what IS the big issue here? If you don't trust the author to be at least that conscientious (or maybe he just doesn't care because it's a moot point - a view I can relate to!) then you may as well throw out all his data and move on ...

[sundancerx]

anyone has a good pic of i7 and deneb stock cooler? could be a good indicator of how hot these chips can get.
edit:pic (side by side for good size comparison) or dimension i should say.

[QuietIce]

The AMD 4-pipe HS far exceeds minimum requirements so I'm not sure that would show much ...

[knopflerbruce]

I'm just wondering... why don't the thermal sensors work like they should? I mean... you manage to design a CPU, but something as simple as a temperature measuring device fails

What's funny is that the sensors were OK back in the 939 days, but with 65nm they got pretty much useless.

Not sure if this is OT, but it would definately be a good idea to find out if the Deneb sensors work properly. I HATE using everst to measure temps

[Astennu]

I ordered a Phenom II 940 BE yesterday. (they are in stock at some stores over here). I should get one Wednesday. I will be comparing it to my Q9450 @ 3,5 GHz. I expect they will be quite matched in therms of performance. Maybe the PHII needs 100-200 MHz more to be on par with the heavy crunching apps.

[jonspd]

I did the same with my q9450 and if your intel cpu is at 3.5 then you will need 4ghz or better out of the PII IMO to make it a close match.

my q9450 at 3.4 folds better then my PII at 3.7 by 200-400 ppd better.

cant wait to see some 940 BE's around the states...

[gallag]

My early point about volts, power and heat was not based on volts alone, In this review when all at at 3.7ghz deneb uses more power than i7 so it should be warmer, No? Or is there any good reason why the temp sensors on deneb report it to be radically cooler while using more power?

[Astennu]

I did the same with my q9450 and if your intel cpu is at 3.5 then you will need 4ghz or better out of the PII IMO to make it a close match.

my q9450 at 3.4 folds better then my PII at 3.7 by 200-400 ppd better.

cant wait to see some 940 BE's around the states...

Uhmm to bad. But i think gaming performance will be quite on par with a few exceptions.

I will wait a bit before i will try a i7 for comparison. The platform is to expensive. I'm gonna wait for the cheaper P55 boards. In some ocations i7's are slower then Penryn or a PHII. i dont know what this is. Low res they are faster or on par with the competition but sometimes they are slower. The Performance is not good enough if you look at the price. And PHII is a nice upgrade. My second pc already has a 780g board. So i only need to upgrade the bios and change the Phenom X3 8450 for a PHII 940 and we are done. If i want a big boost from my intel rig i will need to change almost everything. Of i will have to buy a way to expensive QX9650 to get above the 4.0 GHz.

But i like low power usage so i keep my Q9450 very close to stock volts. I'm gonna do some efficiency tests with the PHII aswell. Like i did with my 5600+ and Q6600. With the Q6600 i needed 0.25v exta to get from 3.2 to 3.550. But then the power i pulled from the outlet jumped 80 watt. With a efficiency of 83% on my PSU the cpu was pulling about 65 watt extra.

Of that same reason i was running my Opteron 170 on 1.27v with 2.8 GHz. If i wanted to do 3.0 i needed 1.35v and performance per watt went down really fast from that point.

I'm gonna do the same tests with the PHII. Its to bad my Abit 780g has no options to lower the voltage (at least on my X3 witch is running 1.25v). There is a chance that i can run the 940 on 1.25v but there is also a chance the bios will adjust the minimum voltage to 1.35v.

[Uncle Jimbo]

My early point about volts, power and heat was not based on volts alone, In this review when all at at 3.7ghz deneb uses more power than i7 so it should be warmer, No? Or is there any good reason why the temp sensors on deneb report it to be radically cooler while using more power?

To figure out the power, we need some independent measurement of the heat output. I have done some testing on an Intel i7 965 with a heatsink calibrated against a resistive load at .12C/W. That's a pretty good cooler by the way. Testing was done with the thermal throttling disabled. At 3.2G, 1.15V, rise was 20C (45C in 25C ambient), which gives 166W heat out. At 3.5G, 1.4V , we saw a rise of 31C (56C in 25C ambient). That says power generated by the chip is 31/.12 or 258W. So adding the volts and clock increased the heat load by 56%.

The average of peak internal thermal sensors for the 56C temp was 68C. No real way to tell if the internal temp is 'accurate' but it's in the ball park and is repeatable.

I have not done similar testing on Phenom II but hope to run some tests in the next few weeks. A friend who has one measured (at stock 1.35V and 3G clock, 30 min full load) the base of his sink at 38C, internal temp showed 43C in 25C ambient. His cooler is not calibrated but it is in the .15 C/W range, so that would be around 85W. Only time (and a calibrated cooler) will tell how temp goes up with increased voltage and clock.

[iocedmyself]

The only ONLY place i've ever seen inaccurate temps with either Phenom 1s or phenom II's has been in AOD, and that depends largely on the version.

As phenom II's are not out yet however, nor are the motherboards for which they are intended (which will feature the enhanced C&Q and allow you to set P states manually) Bios support isn't 100% which is kinda pivitol in accurate temp readings as well as power estimations.

But again having used thermal sensors, several pieces of software, bios temp and IR thermometer as far as the Asus 790GX board goes temps report accurately and the chips run very very very cool.

[Viss]

I ordered a Phenom II 940 BE yesterday. (they are in stock at some stores over here). I should get one Wednesday. I will be comparing it to my Q9450 @ 3,5 GHz. I expect they will be quite matched in therms of performance. Maybe the PHII needs 100-200 MHz more to be on par with the heavy crunching apps.

Same here mate Testing i920 @ 4ghz as we speak for compare with Deneb

[Astennu]

Nice. Let me know what apps you are gonna test and your setup. Maybe we can do some cross compare.

[TL1000S]

idk if he was talking to you because it doesn't have to do with what you said. he might of been responding to people in this thread that have said because of voltage is higher it should consume more power which isn't true.

My answer to the previous post was *obviously* about temp sensors and numbers people like to think is real temps.
This regards CPU's from both "camps".
Fortunately in the Intel camp there are people realizing this and try to do something about it (Unclewebb).
This should also be the case for AMD users.
Personally I use this numbers as an indication/representation for my system, AMD or Intel.

Edit:
Uncle Jimbo has by far the most scientific approach to this issue. Same with Unclewebb (RealTemp author).

[Calmatory]

I just don't get how it can use more volts and power but still be far cooler.

The 45nm process allows higher voltage than Intel's equivalent. SOI + Immersion lithography do the magic.

Edit: clocks.. Voltage! Too tired today.

[Astennu]

The first people on my dutch forum got one. First result with SB600 and a Artic Freezer 64 Pro was 3.8 GHz 1.5v. Nothing special there. I was hoping to see 3.8 near stock volts but i guess you need a SB750 for that.

He did have his HTT bus set to 1800. If he lowers it to 1000 He Might get a bit higher.

[jonspd]

Uhmm to bad. But i think gaming performance will be quite on par with a few exceptions.

I will wait a bit before i will try a i7 for comparison. The platform is to expensive. I'm gonna wait for the cheaper P55 boards. In some ocations i7's are slower then Penryn or a PHII. i dont know what this is. Low res they are faster or on par with the competition but sometimes they are slower. The Performance is not good enough if you look at the price. And PHII is a nice upgrade. My second pc already has a 780g board. So i only need to upgrade the bios and change the Phenom X3 8450 for a PHII 940 and we are done. If i want a big boost from my intel rig i will need to change almost everything. Of i will have to buy a way to expensive QX9650 to get above the 4.0 GHz.

But i like low power usage so i keep my Q9450 very close to stock volts. I'm gonna do some efficiency tests with the PHII aswell. Like i did with my 5600+ and Q6600. With the Q6600 i needed 0.25v exta to get from 3.2 to 3.550. But then the power i pulled from the outlet jumped 80 watt. With a efficiency of 83% on my PSU the cpu was pulling about 65 watt extra.

Of that same reason i was running my Opteron 170 on 1.27v with 2.8 GHz. If i wanted to do 3.0 i needed 1.35v and performance per watt went down really fast from that point.

I'm gonna do the same tests with the PHII. Its to bad my Abit 780g has no options to lower the voltage (at least on my X3 witch is running 1.25v). There is a chance that i can run the 940 on 1.25v but there is also a chance the bios will adjust the minimum voltage to 1.35v.

Yea I'm hoping to pick up i7 around april (b-day) when the am3 cpus start to hit.

I'm currently on a sb600 and all of my testing was done at 1800ht also as you talked about maybe I should try 1k or lower...

[SkullCracka]

1k-ht is only for extreme cooling such as ln2.

[iocedmyself]

The artic freezer 64 pro is not exactly beefy..especially when pushing voltage to 1.5.

Unless you are using water or cold going past 1.45v is less beneficial the higher the voltage increase...when going for high clocks on air a $28 heatsink isn't what i'd use as a accurate measurement of potential, though considering the that is the heatsink and voltage is that high, the clock is rather impressive.

Yes lowering the HT can be helpful, but mainly to prevent coldbug, or when going for clocks in excess of 4ghz. Check the Cpuz banner below,

245x16 3932mhz, 2457 HT and NB clock on air ACC disabled @ 1.475v i believe, granted it's on a 790GX 750SB board, but still was able to pull 4ghz on SB600 790FX without breaking much of a sweat and HT speed at default.

[Macadamia]

I'm not sure, but IF AMD is demo-ing the Phenom II in CES, I heard that no more <1k HTT in the new revision is needed. I think Sampsa whispered that quietly in the Xtreme 3D section...


Probably a new stepping.

[leoy]

That would be nice for some 3d damage.

[imsochobo]

Lhe and high ht would never work by my guess.
If youre phenom 2 doesnt exceed Core 2 clock for clock in games.
Turn up HT link. grants you some extra % ~5 with good enough graphics cards.

Excited about the rumoured amd Lhe run @ CES!!! seriously... i cannot wait.