Is it my idea or Penryn ---> Yorkifield TDP is really bad

[oscar3d]

Hey guys:

Just checked the news on the upcoming Yorkfield brand on Nov.12.

I was checking the TDP's on this things and until then I was looking foward to check the QX9650. But OMG did you guys see the TDP on that?

130W

I was really hoping to get my hands on it, but seriously, the price on that chip is +$999. I was hoping that a 45nm shrink was going to help on the temps, but this is absolutely insane and dissapointing. So what elso do you get for that price apart from the large 12mb cache and the SS4 instructions?

Nothing?

I have a hard time cooling a Netburst P4 3.8GHz in the past and I really don't want to have another "devil's waterbag" furnace inside my case.

Right now I have a Core 2 Extreme X6800 and the TDP on that thing is 75W which is half of the top Yorkfield and at 65nm. So seriously what is so good about the 45nm shrink?!

My second option would be the Yorkfield Q9550 @ 2.83Ghz with a TDP of 95W. And the price cut's down to +500 which makes sense.

I would be interested to know what you guys have in mind.

Also have you checked the Xeon Yorkfields? These are much more decent in tdp than the desktop processors. Look at this.

E5405 2.0 GHz 1333 MHz 80 W 2x 6 MB
L5410 2.33 GHz 1333 MHz 50 W 2x 6 MB
E5410 2.33 GHz 1333 MHz 80 W 2x 6 MB
E5420 2.5 GHz 1333 MHz 80 W 2x 6 MB
L5430 2.67 GHz 1333 MHz 50 W 2x 6 MB
E5430 2.67 GHz 1333 MHz 80 W 2x 6 MB
E5440 2.83 GHz 1333 MHz 80 W 2x 6 MB
E5450 3.0 GHz 1333 MHz 80 W 2x 6 MB
X5460 3.17 GHz 1333 MHz 120 W 2x 6 MB
E5462 2.8 GHz 1600 MHz 80 W 2x 6 MB
E5472 3.0 GHz 1600 MHz 80 W 2x 6 MB

Personally I would pick the E5450. With a TDP of 80W, makes much more sense and cooled just as close as my current X6800, and I get 3.00Ghz.

All of this makes me thing. What is really the difference betwen the QX6850 (the current top Intel Quadcore) and the upcoming QX9650. I really really don't see much difference, specially with the benches that have been run on it.

Any inputs, feedback or suggestions?

Thanks guys.

[Aris]

any CPU over $300 is too much.
quad core is over rated, most current software doesnt even fully utilize more than 2 cores right now anyhow.
any CPU with a TDP more than 75w is too much.

[smilingcrow]

The top bin chip usually has a disproportionately high TDP which doesn’t reflect accurately on what the whole range is capable of. The previews show Penryn consuming less power than Conroe across the board and the same for the Xeon flavours so I don’t see anything to be concerned about. It is strange to see the Xeon’s having a lower TDP than the desktop chips though!
The unanswered question for me is how well will the 45nm chips under-volt as the G0 Quads are undervolting monsters. The one I tested was stable at 1.05V in the BIOS which dropped power consumption under a heavy load by 34W. From what I’ve seen this is typical. Who needs Penryn when you can run a Q6600 at 3GHz for a system power consumption of 157W running Prime95.

According to Dailytech the 2.6GHz Phenom X4 has a TDP of 125W which is something that looks more of a concern as it’s not the top bin. Hopefully the top bin of 3GHz or whatever will have the same TDP.

[oscar3d]

Hi smiling crow:

So you basically say that a QX9650 TDp 130W would consume less power than my current Conroe X6800 TDp 75W?

How is that so? Excuse me for my ignorance but I'm really trying to undertand.

[smilingcrow]

So you basically say that a QX9650 TDp 130W would consume less power than my current Conroe X6800 TDp 75W?

How did you manage to draw that conclusion!

[oscar3d]

That's why I need clarification

Hehe, I guess the answer tomy prayers would be the PenrynQ9550 sitting at 95W TDP, just like the the Q6600.

Penryn will only gain in 12mb cache and SS4 instructions, but for the die shrink to 45nm we will actually get nothing.

Hmmm

[jackylman]

I know the transistors are smaller, but 12MB of cache has to increase the transistor count to something ridiculously high.

[smilingcrow]

Penryn will only gain in 12mb cache and SS4 instructions, but for the die shrink to 45nm we will actually get nothing.

The preview at Techreport states:

‘The Stoakley/Harpertown pairing brings a drastic drop in power draw versus the Xeon X5365s on Bensley. In fact, the Stoakley/Harpertown combo at 3GHz draws less power than Bensley/Clovertown pairing at 2.33GHz. Notably, the Xeon E5472 system also consumes less power than the Opteron 2360 SE-based one.’

[austinbike]

That article is somewhat misleading. They talk about how the Penryn product is 30% lower in power, but they neglect to mention that they are comparing an 80W E processor to a 120W X processor. Of course it is going to draw less power.

The real crime is that in going to 45nm the products are actually not going to have a different set of TDPs, servers are 50, 80 & 120, just like their 65nm counterparts.

In a different review, they saw penryn platforms at slightly higher power at idle and slightly lower under load:

http://www.hardware.info/en-UK/articles ... pertown/19

I also take all of the POV-ray data with a grain of salt. Intel added some new instructions to their processors and this happens to be one of the few applications that take advantage (doing things in silicon is a lot faster than doing it in sw). Seeing performance differences in web, database, email, etc is more realistic. The Java benchmark data is interesting, but, java loves cache.

[smilingcrow]

That article is somewhat misleading. They talk about how the Penryn product is 30% lower in power, but they neglect to mention that they are comparing an 80W E processor to a 120W X processor. Of course it is going to draw less power.

The real crime is that in going to 45nm the products are actually not going to have a different set of TDPs, servers are 50, 80 & 120, just like their 65nm counterparts.

You seem to be obsessing about TDP so why not focus on the raw power consumption data and the performance per watt data and not get hung up on TDP. Perhaps someone here can point you in the right direction to help you see TDP in a fuller context.

I also take all of the POV-ray data with a grain of salt. Intel added some new instructions to their processors and this happens to be one of the few applications that take advantage (doing things in silicon is a lot faster than doing it in sw).

You are mistaken about PovRay and possibly confusing it with a beta of the DivX codec which supports SSE4. There isn’t much of a difference between the 45 and 65 nm Xeons in PovRay which supports this, whereas Penryn performance with the SSE4 enhanced DivX codec almost doubles.

[austinbike]

Actually, its less about TDP because that is only an engineering number.

The key is power draw at the wall. If you check that link you'll see the comparison at the wall of the penryn vs. clovertown. Penryn is slightly higher at idle, slightly lower at load. Definitely not the 30% number that people are tossing around.

[smilingcrow]

The key is power draw at the wall. If you check that link you'll see the comparison at the wall of the penryn vs. clovertown. Penryn is slightly higher at idle, slightly lower at load. Definitely not the 30% number that people are tossing around.

I looked at the link you posted and noticed that the two systems had different specifications so aren’t comparable; they had different disk sub-systems and the 45nm system had twice as much RAM and a faster CPU.
If you look at comparative reviews that use more evenly matched systems it will give a better indication of relative performance and power consumption.
The Techreport preview shows some very significant gains for Intel for the new platform. As for the 30% number that you mentioned there are just too many varying Server configurations for an average figure to have much meaning. The first graph below shows the amount of energy needed to complete a particular task so is equivalent to performance per watt. I‘m not suggesting that this is typical but it’s the only such comparison that I’ve seen.

System configurations, all with dual CPUs:

E5472 – 3GHz, 1600 FSB, 45nm, 16GB RAM (8 sticks)
X5365 – 3GHz, 1333 FSB, 65nm, 8GB RAM (8 sticks)
E5345 – 2.33Hz, 1333 FSB, 65nm, 8GB RAM (8 sticks)
L5335 – 2GHz, 1333 FSB, 65nm, 8GB RAM (8 sticks)




The 3GHz 45nm system consumes 1W less at idle compared to the 2GHz 65nm low voltage chip even though it has twice as much RAM and a faster FSB. The idle figures are still appalling though compared to AMD.


The 3GHz 45nm system consumes 9W less at load compared to the 2.33GHz 65nm chip.

I’m not suggesting that this data is useful as a means of comparing Xeon with Barcelona but it is useful for comparing the two Intel platforms.

[Flandry]

Bear in mind that the first stepping of a chip fabbed at a new process node will not be as highly optimized as the final stepping of the previous generation at a larger node. This is true of every generation i can think of. Regardless of what the actual situation is now, give it six months and you'll see noticeable improvements in power draw across the boards of the 45 nm vs. the 65 nm parts.

However, taking the top-binned chip is always asking for a wildcard when it comes to power consumption, as was mentioned. Finally, as was also mentioned, TDP isn't a particularly meaningful number...

[austinbike]

Yes, the systems had different amounts of memory, but the number of DIMMs was the same (8 1GB vs. 8 2GB.)

This was basically "apples to apples" in power because power measurements are not generally impacted by the amount of memory, but rather by the # of DIMMs.

However, on the performance side they should be ashamed to publish benchmarks based on 2 systems with radically different memory configurations.

I'll be interested to see what the final shipping products look like.

[austinbike]

Bear in mind that the first stepping of a chip fabbed at a new process node will not be as highly optimized as the final stepping of the previous generation at a larger node. This is true of every generation i can think of. Regardless of what the actual situation is now, give it six months and you'll see noticeable improvements in power draw across the boards of the 45 nm vs. the 65 nm parts.

However, taking the top-binned chip is always asking for a wildcard when it comes to power consumption, as was mentioned. Finally, as was also mentioned, TDP isn't a particularly meaningful number...

Absolutely true. However, the Opteron was not the final stepping either. It will be interesting to see the final results on both systems.

[smilingcrow]

Yes, the systems had different amounts of memory, but the number of DIMMs was the same (8 1GB vs. 8 2GB.)
This was basically "apples to apples" in power because power measurements are not generally impacted by the amount of memory, but rather by the # of DIMMs.

The power consumption of FB-DIMMs is mainly affected by the number of RAM sticks but the clock speed and memory size is also a factor but certainly a lesser one. The specs for the two systems in the review you linked to are:

65nm server - two Intel Xeon X5355, 2.67 GHz, 1333 MHz FSB, 8GB (8x1) DDR2-667 FB-DIMM, a RAID 5 array of three Fujitsu Allegro MAX3073RC harddisks.

45nm server - two Intel Xeon E5472, 3.0 GHz, 1600 MHz FSB, 16GB (8x2) DDR2-800 FB-DIMM, two Seagate Cheetah 15k.5 146 GB harddisks in a fast RAID 0 array.

The only thing they have in common is the number of RAM sticks; not my idea of Apples to Apples.

However, on the performance side they should be ashamed to publish benchmarks based on 2 systems with radically different memory configurations..

It’s not an issue if the applications being tested are happy with ‘only’ 8GB of RAM. I very much doubt that the PovRay performance data above is affected by this and that’s the only significant data that I’ve seen with regard to performance per watt so I think it stands.

[ronrem]

Actually, its less about TDP because that is only an engineering number.

The key is power draw at the wall. If you check that link you'll see the comparison at the wall of the penryn vs. clovertown. Penryn is slightly higher at idle, slightly lower at load. Definitely not the 30% number that people are tossing around.

Yeah....Figuring how hot/noisy/power hungry a chip (not yet out in the real world) will be is like basing your bet on how fast a horse runs on how deep of a mark his hoofprints leave. There MIGHT be some minimal...and relative info...or not.

So far.......nobody has come up with a yardstick and run tests so we get a true view of watts from the wall or specific heat output. Once these chips are available....30+ websites will run pretty much the same benchmarks...how far will it overclock,how fast does it make an MP3...how quick is it in Doom or Quake or whatever game.

I'd like to know if it UNDERVOLTS well....can you fry chicken on it or can it be cooled passive....how many REAL watts will it add to the bill.

Nobody will get around to that for MONTHS

[smilingcrow]

So far.......nobody has come up with a yardstick and run tests so we get a true view of watts from the wall or specific heat output. Once these chips are available....30+ websites will run pretty much the same benchmarks...how far will it overclock, how fast does it make an MP3...how quick is it in Doom or Quake or whatever game.
I'd like to know if it UNDERVOLTS well....can you fry chicken on it or can it be cooled passive....how many REAL watts will it add to the bill.
Nobody will get around to that for MONTHS

I’m not clear from your tirade (not a bad thing) what data you are specifically looking for.
Many of the reputable (as far as I’m concerned) mainstream websites include power data these days and have done so for a while now. I personally prefer it when they measure system power at the wall rather than ‘trying’ to determine how much the CPU itself actually consumes. I say ‘trying’ because the data varies so much between reviews that it doesn’t seem reliable.

SPCR seems to have lost interest in such reviews after a promising initial look at the desktop way back in time.
There has been some recent forum led analysis of under-volting a Q6600 that really caught my attention on OCForums; 1.05V at 2.4GHz for a quad core at 65nm is impressive, especially when you see the power consumption.
The reviews linked to above already show a comparison of Xeon at 45 v 65nm.
The initial data on over-clocking Penryn on the desktop at Xtremesystems is showing some people hitting 4.5GHz on air. That has to bode well for lower power consumption and temperatures.

Reading between the lines I’d say that Penryn is looking very good and the only unknown quantity is how well it undervolts. I wish we had such a clear picture of K10 which is already released!

[Tzupy]

This review at Xbitlabs shows about 30% less power consumption for the 45 nm QX9650 (Yorkfield) compared to the QX6850 (Kentsfield).
That was at load, when idle it's less than half! http://www.xbitlabs.com/articles/cpu/di ... 50_13.html
Unfortunately the price of the QX9650 is going to be prohibitive for the next few months.

[smilingcrow]

After looking at the reviews of the first Penryn desktop chip I’m gob-smacked by just how low the power consumption is. I started to think that the comparisons weren’t fair as these are the $1,000 Xtreme Edition chips so they will be cherry picked, but then I remembered that they are been compared to the 65nm Xtreme Editions in many reviews which is a like for like comparison.

Intel have been claiming for ages that they have massively reduced the static power leakage with their 45nm process and when you look at the idle power consumption for these chips you can see what they mean. Intel’s claims with regard to pre-release products over the last 18 months have proven to be very reliable which makes the astounding claims that they make for Nehalem even more worthy of note.

Data from Anandtech:
At idle, the QX9650 draws an impressive 34W less than the QX6850 - there's 45nm high-k + metal gate transistors in action for you.
Under load the power advantage is even more impressive with a 47W delta the QX9650 under load uses only 11W more than its predecessor at idle.



Here’s some data from Techreport:

Power Consumption





Performance per watt


Over-clocking data:

Anandtech – 4GHz (333x12), 1.40V (stock = 1.25V)
Techreport - 3.66GHz (333x11) 1.2875V (stock = 1.25V)
XBitlabs – 3.41GHz (341x10) 1.2V (stock = 1.2V)
XBitlabs – 4.07GHz (339x12) 1.5V (stock = 1.2V)
Hothardware – 3.9GHz (355x11) 1.4V (stock = 1.232V!)

All results were achieved with air cooling and at least some used the stock Intel cooler.
As a comparison my Q6600 G0 which impressed me managed 3GHz at its stock 1.275V.

I suppose the big unknown is how well these chips undervolt, but this seems almost an irrelevance when you look how good they are at stock voltage.

I don’t mean to come across as an Intel fanboy but after years of promoting and exclusively building AMD systems it’s only fair to promote whatever is best at any given time.
Looking at the above data it’s hard to see why this CPU has a TDP of 130W! That is until you remember that Intel isn’t especially concerned about promoting TDP as a meaningful way of deducing power efficiency in the way that AMD is.

Conclusion – No pressure on AMD then with the pending Phenom X4 release!

[JazzJackRabbit]

Yes, it amazing that quad core 9650 basically ties with dual core 6850 for energy consumption.

[merlin]

It think penryn hammers home one major point. TDP DOES NOT MEASURE POWER USAGE. It only measures maximum possible thermal power that must be cooled in possibly the entire family of chips. Considering we have a TDP of 130w and sites measuring LOAD power usage of 76w for the chip, the only real answer for power usage is a measured value. Intel TDP are almost worthless when finding true power usage.