AMD Shanghai/Deneb Review Thread

[qurious63ss]

The Barton processors originally came out with a max 'rating' of 2600. A year later that was up to 3200. I don't know how much you understand about semiconductor manufacturing, but the process tuning to get yields of certain parts is not the same as the chip design. So if they want to shift yield balance, they might do minor tweaks to increase yields of higher speed parts. If, on the other hand, they see a bigger volume need for low power chips, and there is a market for more than the bin split gives, they may adjust the process to increase those yields.

I think AMD is more interested in tuning the process for low power than for high speed.

DFM, process tweaks, etc. is all good but my point is that their is no way that a 4ghz Phenom II will only dissipate 90W at load. If you are implying that this first batch of PII are "tweaked for low power" then they must also be "tweaked for higher speed parts" as well since they are running at 4ghz with 90W?? The poof is in the pudding, if this chip could run at that speed with that wattage then we would already of seen it released as an FX chip regardless of what their tuning their process to do. $$$$$$

[roofsniper]

DFM, process tweaks, etc. is all good but my point is that their is no way that a 4ghz Phenom II will only dissipate 90W at load. If you are implying that this first batch of PII are "tweaked for low power" then they must also be "tweaked for higher speed parts" as well since they are running at 4ghz with 90W?? The poof is in the pudding, if this chip could run at that speed with that wattage then we would already of seen it released as an FX chip regardless of what their tuning their process to do. $$$$$$

who said it would only be 90W at 4ghz?

edit: just read that review and it looks pretty good. im confused as to why they did the gaming tests at such a low resolution. because really whats the point of 1024x768? does anyone still run that? so that part of the review was just pointless imo. the power consumption looks good too. completely different from hwbox's results. interesting to see that the 920 and 940 use the same amount of power in standby and idle and are close in load. shows that maybe it won't heat up too bad when it overclocks. i also saw that they listed the 940 and 920 both as unlocked too so idk about that.

[qurious63ss]

edit: just read that review and it looks pretty good. im confused as to why they did the gaming tests at such a low resolution. because really whats the point of 1024x768? does anyone still run that? so that part of the review was just pointless imo.

They test at low resolution to remove the bottleneck from the GPU no?

[roofsniper]

They test at low resolution to remove the bottleneck from the GPU no?

maybe but still what does that say? if you want to know gaming results then you want to know what your cpu will get at normal resolutions. as we saw with the hwbox review at 1680x1050 deneb got about the same performance as competing intel cpus and sometimes even better. showing it at 1024x768 is a whole new story.

[qurious63ss]

maybe but still what does that say? if you want to know gaming results then you want to know what your cpu will get at normal resolutions. as we saw with the hwbox review at 1680x1050 deneb got about the same performance as competing intel cpus and sometimes even better. showing it at 1024x768 is a whole new story.

Yes, hwbox numbers were the same because the GPU became the bottleneck so those numbers actually say less about the CPU and more about the GPU. The low resolution numbers I think are more informative if you wan't to future proof your rig so that say you upgrade to a faster card in a year or so the chances of your CPU becoming the bottleneck are less likely.

[roofsniper]

Yes, hwbox numbers were the same because the GPU became the bottleneck so those numbers actually say less about the CPU and more about the GPU. The low resolution numbers I think are more informative if you wan't to future proof your rig so that say you upgrade to a faster card in a year or so the chances of your CPU becoming the bottleneck are less likely.

not necessarily. the lower resolution numbers don't test the same exact things as the higher ones do. it just seems that if you are going to post gaming benchmarks then you post what people game at. 1280x1024 was the highest i saw and does that make me want to buy it when im running 1650x1080? i want to know how it performs on the resolutions that i run and the resolutions that most others run as well. if they were making the gaming benchmarks to show how futureproof it would be you would think that they would show it at the high resolutions.

[Hornet331]

not necessarily. the lower resolution numbers don't test the same exact things as the higher ones do. it just seems that if you are going to post gaming benchmarks then you post what people game at. 1280x1024 was the highest i saw and does that make me want to buy it when im running 1650x1080? i want to know how it performs on the resolutions that i run and the resolutions that most others run as well. if they were making the gaming benchmarks to show how futureproof it would be you would think that they would show it at the high resolutions.

There are to many resolution to test, and such stuff usually is part of GFX tests and not cpu tests.
E.g. Im only interessted in 1920x1200, but since at that resolution most cards just reach its limit it wont tell you how good the cpu is and what you can expect if you later upgrade your gpu. (thats just my viewpoint on that topic)

[qurious63ss]

not necessarily. the lower resolution numbers don't test the same exact things as the higher ones do.

What do you mean by that? Again, I could be wrong but at higher resolutions you actually testing the GPU and not the CPU since the GPU becomes the bottleneck.

[Uncle Jimbo]

DFM, process tweaks, etc. is all good but my point is that their is no way that a 4ghz Phenom II will only dissipate 90W at load. If you are implying that this first batch of PII are "tweaked for low power" then they must also be "tweaked for higher speed parts" as well since they are running at 4ghz with 90W?? The poof is in the pudding, if this chip could run at that speed with that wattage then we would already of seen it released as an FX chip regardless of what their tuning their process to do. $$$$$$

I really don't know what they are tweaking for. I was pointing out that historically, AMD brings out parts at the 'design center', and as they tweak the process, they bring out both lower power and higher speed parts. In the server world, they already announced that there will be a high speed 'SE' and a low power 'LE' added to the Shanghai line in 2009. So my assumption is the same will happen with Phenom II.

As far as the wattage, the methods many reviewers use to measure power are measuring total board power or even total system power. Those might be interesting if you are concerned about total energy draw, but are pretty much useless for determining processor wattage, since they include chipset, RAM, and anything else on the board.

The method I use measures actual heat out from the processor. Once I get a 940 and test it. I'll be able to say for sure what the results are. Based on the few tests that gave enough information to do the calculation, the estimate of under 100W at 4G seems reasonable.

[gallag]

the estimate of under 100W at 4G seems reasonable.

lol, You gotta be kidding. It will be closer to double that.

[Uncle Jimbo]

lol, You gotta be kidding. It will be closer to double that.

you know that bumper sticker that says 'Gravity - not just a good idea, it's the law'? That's kind of what you are doing. Not to be arrogant, but this is physics, not someone's opinion.

A heat load that produces a 10C rise on a .1 C/W calibrated sink is 100W. Doesn't matter if it is a resistor, a light bulb, or a CPU. Since I don't have one of the Phenom II parts, I have not done any direct testing. But I do know the thermal coefficients of many popular high performance coolers, and based on the heat numbers we have seen from a number of posters, it looks like the Phenom II parts are running pretty cool with decent overclocks.

The wattage of CPUs under various load conditions is one of the areas where even experienced techies get it completely wrong - like the guy who measured 170W by taking the amp draw on the secondary MB connector and multiplying by 12. However, at least on his 3.85G run, he was using a XIGMATEK HDT S983 V2, a good cooler with .13 C/W performance. Unfortunately, he did not measure temp at the base of the cooler, and give ambient, which would have allowed us to tell what his real CPU wattage was.

So you can think what you like - but the data out on the web points to significantly lower wattage than previous phenom, and probably in the ball park of the better C2D parts (which are also very impressive). The fact that most OC requires jacking up the NB and RAM voltage (which increases their dissipation) means that getting a CPU to CPU comparison is impossible using the methods most reviewers use. Measuring power draw at the wall is easy, but also not very useful for anything other than total system load. I have verified the accuracy of the thermal gradient measurement by comparing results with instrumented VRMs, and both give the same results.

[qurious63ss]

The method I use measures actual heat out from the processor. Once I get a 940 and test it. I'll be able to say for sure what the results are. Based on the few tests that gave enough information to do the calculation, the estimate of under 100W at 4G seems reasonable.

I see. Well the method I was talking about uses the formula P=I*E. So Watts is a function of current and voltage. The way you are getting wattage numbers is by reading how much heat a chip generates?

[Uncle Jimbo]

I see. Well the method I was talking about uses the formula P=I*E. So Watts is a function of current and voltage. The way you are getting wattage numbers is by reading how much heat a chip generates?

Correct - the 'I' you are measuring may or may not be going to the CPU, and likewise you don't know what 'E' actually gets supplied internally, but there is no question about the heat load.

Here's a good write-up on heatsink characterization by some guys who actually know what they are doing:
http://www.frostytech.com/testmethod_mk2.cfm

[The PyroPath]

Just got back home with my sweet little puppy 940

I'll installing it for the next hour or so (adding some more stuff while I'm at it) yoo bad my new motherboard wasn't in yet... my Gigabyte GA-MA790X-DS4 will have to do for another week or so (my dreaded friend asked me to order a graphics card from the same shop and now that one has gone from in stock to backorder all of the sudden)

So let's see what this baby can do, I'll report back on the wattage by the way, I have a Zalman fancontroller with watt's reading for the complete system. Isn't overly accurate but I can compare it to my Phenom 9500 I use ATM...

[qurious63ss]

Correct - the 'I' you are measuring may or may not be going to the CPU, and likewise you don't know what 'E' actually gets supplied internally, but there is no question about the heat load.

Here's a good write-up on heatsink characterization by some guys who actually know what they are doing:
http://www.frostytech.com/testmethod_mk2.cfm

So you measure temp at the base of heatsink? So that means the better your heatskink is at dissipating the heat, the lower the temp at the base of the heatsink should be, correct? So how does heat translate into the PIE formula? Power is still a function of current and voltage so how would lowering the temp of the heatsink affect power draw?

[Uncle Jimbo]

So you measure temp at the base of heatsink? So that means the better your heatskink is at dissipating the heat, the lower the temp at the base of the heatsink should be, correct? So how does heat translate into the PIE formula? Power is still a function of current and voltage so how would lowering the temp of the heatsink affect power draw?

A CPU is not a resistive load, and actual draw varies from one microsecond to the next. There is also no way to determine 'I' for a CPU unless you instrument the VRM. So it's not practical for most people to use P=I*E since they can only guess at 'I'. We do know that a CPU does no mechanical work, so all of the input power is turned into heat. Basically the heatsink method gives that power directly.

Since we know (or can determine) the thermal resistance of the heatsink, we can determine the power dissipated by applying the thermal resistance Tr (in C/W) where C is the difference between the base and ambient temperature. The formula is Crise / Tr = Wload. The actual temperature at the base, along with ambient, is only used to determine the rise.

Lowering the temperature at the base will have some effect on total power, but not very much unless the change is large. It's always best to use a big cooler, but the goal is to get rid of the heat, not reduce power.

If for some reason you want to calculate amps input, and you know power (from the heat load) and E (from the Vcore) you could use I=P/E to find amps... but unless you are designing a VRM, I'm not sure why you would do that.

[Zucker2k]

Snip

Uncle Jimbo what do you make of this?

I am certain that the reason so many overclocks by reviewers were so low was the sheer lack of experience with the systems. I was afraid of blowing up my sample before the launch date so I stuck with a conservative 1.45vCPU which yielded a paltry 3800MHz. Now that the article is live, I'm blasting 1.55vCPU through this chip at 4200MHz and climbing

The biggest issue I have though is the lack of temperature readings, I'm currently using a thermalprobe wedged up against the IHS of the actual CPU to get a ballpark figure :-\ It's reading around 55C at idle while the thermal diodes on the chip still read 19C

From here: http://www.xtremesystems.org/forums/...=213691&page=5