Anyone else considering upgrading from q6600?

[Idontcare]

I likely won't upgrade any of my 65nm Kentsfields to 45nm Yorkfields until well after Nehalem is out and the 9550 become basement bargains.

I am more likely to add Nehalem-based systems to my collection before upgrading the existing Q6600's.

The payoff for me just isn't there until I can get Yorkfields <$200 that clock >3.3GHz without requiring new mobo and ram.

Nehalem is special (to me) because the opportunity to double the computing power in a commodity box is there thanks to SMT.

If I can run 8 threads in a box with system level power consumption fitting inside the footprint of my existing 4-thread Q6600 boxes then I consider that a worthwhile investment. Until then I will just run-out the depreciation on my Q6600 boxes.

 

[bryanW1995]

Yeah, but how much faster than a highly overclocked Q6600 is a midlevel nehalem going to be? Based on everything I've seen, it appears that nehalem will have major differences between midrange and high-end.

 

[Idontcare]

Originally posted by: bryanW1995
Yeah, but how much faster than a highly overclocked Q6600 is a midlevel nehalem going to be? Based on everything I've seen, it appears that nehalem will have major differences between midrange and high-end.

For me it's a value-proposition. My "highly" overclocked Q6600's all run at 3.3GHz.

Any higher and I start pissing $ away nearly needlessly to the power company and my financial models barely complete any sooner.

So if I can overclock a Nehalem to deliver approximately the same single-thread performance as a 3.3GHz Q6600 (which would be what, maybe a 2.5GHz Nehalem?) while having 2X the threads and same or lower power consumption (it matters because I'd buy a minimum of 5 systems) then I'd be happy if the mobo + 4GB DDR3 + Nehalem chip came to <$1k.

I imagine just about every cruncher out there is making a similiar assessment. Power may not be such a factor to most though as most folks don't have 6 quads sucking up the electrons.

Gamer's will obviously be another story as GPU budget trade-offs enter into the equation.

 

[bryanW1995]

as an example of crunching, I used to have a p4 w/ ht processor on seti. It looked like 2 cpus in boinc, but only got about 10% more actual work done than a non-ht cpu at similar speeds. If your models completely tax the cpu then I wouldn't be surprised to see a similarly small improvement with ht unless they've significantly improved it in the past few years. Also, when are the actual octo-cores supposed to come out? Are they just going to be the enthusiast cpus, are they getting delayed to 32nm, etc?

 

[Idontcare]

HT was a technique to attempt to put some idle paths to work in the P4 due to it's uber long stages.

SMT is reportedly more like the multi-threading that exists on SUN Niagara and IBM Power6 series wherein the hardware truly exists within each core to support >1 simulatenous thread at full speed.

I wouldn't use P4 HT as a data point when attempting to extrapolate SMT performance of Nehalem. The HT situation was truly an outlier of the industry, even at the time it debuted.

Intel has gone on record forcasting Nehalem multi-threaded performance to be 2X that of Yorkfield but single-threaded performance is only 15% higher...the only way you go from 15% to 200% is if Intel is saying SMT delivers nearly perfect performance scaling.

Price and timeline are the only unresolved questions in my mind. Looking at Yorkfield I just don't see Nehalem being an option on the desktop until late 2009. (excepting for the extreme chip, Bloomfield)

Octo-core Nehalem is still reportedly on track per Intel for 45nm...should be no more of a challenge to manufacture than Dunnington or Tukwila (all these chips will be ~450mm^2) but yield challenges will no doubt exist so prices will be sick high and volume will be paltry low, but no doubt that is planned/intended as such things always are.

 

[bryanW1995]

thank you, good info on smt. does that stand for "simultaneous multi-tasking"? Nehalem should be a cruncher's dream in that case.

You really don't see nehalem on the desktop until LATE 2009? I thought that it was originally planned for late 08 then pushed back to Q1 09. Is this more evidence of amd's incompetence or is intel actually having problems getting nehalem to market in a timely manner?

 

[lopri]

I have a feeling that Intel will try to move on fast to 32nm with Nehalem. I know that performance per clock (a la IPC) will improve, but some folks might be disappointed with its clocking. I predict initial Nehalem will be hot and won't clock very well.

 

[Idontcare]

Originally posted by: lopri
I have a feeling that Intel will try to move on fast to 32nm with Nehalem. I know that performance per clock (a la IPC) will improve, but some folks might be disappointed with its clocking. I predict initial Nehalem will be hot and won't clock very well.

I agree on all accounts.

Doing 2X the work of a yorkfield (8 threads vs 4) will require nearly 2X the power consumption...nothing magical about Nehalems 45nm xtors relative to Yorkfield's 45nm xtors.

And being 2X the die-size of a wolfdale...nehalem is going to more difficult to yield relative to yielding two wolfdales in terms of both defectivity and speedbins.

You can't outrun the same physics that AMD's Phenom is snarled in except for going to a node-shrink.

 

[Idontcare]

Originally posted by: bryanW1995
thank you, good info on smt. does that stand for "simultaneous multi-tasking"? Nehalem should be a cruncher's dream in that case.

You really don't see nehalem on the desktop until LATE 2009? I thought that it was originally planned for late 08 then pushed back to Q1 09. Is this more evidence of amd's incompetence or is intel actually having problems getting nehalem to market in a timely manner?

SMT = Simultaneous multithreading

http://en.wikipedia.org/wiki/S...taneous_multithreading

Regarding timeline...look at the gaping hole between QX9650 release versus Q9450 availablility. It's going to be nearly 6 months by the time Intel is done having their way with the enthusiast consumer.

AMD isn't gaining speed, if anything Nehalem will have even less pressure than Penryn did. So yes I will be AMAZED if we can buy quad-core eight-thread Nehalems for <$500 before July 2009, even Sept 2009 wouldn't surprise me with the new Intel.

 

[Ratman6161]

Well I just upgraded TO a Q6600 and I'm happy with it! Next year at this time when the 9450 is old news, that will be the time to consider it.

 

[lopri]

Originally posted by: Idontcare

Originally posted by: lopri
I have a feeling that Intel will try to move on fast to 32nm with Nehalem. I know that performance per clock (a la IPC) will improve, but some folks might be disappointed with its clocking. I predict initial Nehalem will be hot and won't clock very well.

I agree on all accounts.

Doing 2X the work of a yorkfield (8 threads vs 4) will require nearly 2X the power consumption...nothing magical about Nehalems 45nm xtors relative to Yorkfield's 45nm xtors.

And being 2X the die-size of a wolfdale...nehalem is going to more difficult to yield relative to yielding two wolfdales in terms of both defectivity and speedbins.

You can't outrun the same physics that AMD's Phenom is snarled in except for going to a node-shrink.

Don't forget the memory controller, plus PCIe lanes that are supposedly built into high-end Nehalem. Both of them are known to increase heat/power.

 

[Tempered81]

Originally posted by: sgrinavi
Thinking about going from a q6600 to a 9450 for the quad system in my sig. Used for 3dsMAX & AutoCAD...

Think it's worth it?

Originally posted by: theprodigalrebel
I'd say no but here's a review for your own...err...review

"We have already complained that quad-core processors support is not being adopted by gaming applications fast enough. The result is clearly seen on the diagrams: Core 2 Quad Q9300 is very often falling behind its dual-core competitor at default frequencies as well as during overclocking.

This situation is completely different from what we saw with previous generation processors on 65nm cores. Since Conroe and Kentsfield both overclocked to pretty similar frequencies, overclocked Core 2 Quad Q6600 ran neck and neck with overclocked Core 2 Duo E6850 even in games not optimized for quad-core. Now Core 2 Quad Q9300 cannot catch up with the overclocked Core 2 Duo E8500, so that overclocked have to face a really hard choice between a dual- and a quad-core processor.

Moreover, gaming benchmarks results reveal one more problem. During overclocking Core 2 Quad Q9300 sometimes falls not only behind its dual-core counterpart, but also behind its predecessor running at 3.6GHz frequency. This makes things even more confusing."



I'd say no as well.

 

[Nanobaud]

Anything out there yet indicating power? Plenty of tests showing 9300 OC-Loaded power consumption being well below 6600, but always with the caveat that it has less cache. I tend to run jobs that load-up the computer for long periods of time, and if the 9450 ends-up burning 50W less power then 6600 at similar performance, I might switch just for the peace-of-mind. My 6600 @ 3.0G peaks temp in high-40's on long-term (multiple-day) load, but I can see that creeping up a few more degrees as summer rolls around and then I start to get nervous about the quality of the computation (or I could just back-off the overclock for the important stuff).

 

[Idontcare]

Originally posted by: Nanobaud
Anything out there yet indicating power? Plenty of tests showing 9300 OC-Loaded power consumption being well below 6600, but always with the caveat that it has less cache. I tend to run jobs that load-up the computer for long periods of time, and if the 9450 ends-up burning 50W less power then 6600 at similar performance, I might switch just for the peace-of-mind. My 6600 @ 3.0G peaks temp in high-40's on long-term (multiple-day) load, but I can see that creeping up a few more degrees as summer rolls around and then I start to get nervous about the quality of the computation (or I could just back-off the overclock for the important stuff).

For what its worth you could use the following Anandtech power consumption chart generated by underclocking (reduced multiplier) a QX9650:

http://www.anandtech.com/cpuch...howdoc.aspx?i=3184&p=2

The specific chart is http://images.anandtech.com/gr..._12040751228/16135.png

At 3GHz, the Q6600 number you spec'ed above, Anand reports their QX9650 consumes 54W when "fully loaded".

They don't compare this to anything, so who knows what "fully loaded" means...but I would bet at 3GHz a Q6600 comes in at more than 104W "fully loaded".