What are your thoughts on Dempsey?

[josh609]

http://www.2cpu.com/review.php?id=110

http://www.tecchannel.de/server/hardware/432957/

Seems to be a good competitor to the opteron as far as i read. What are your thoughts?

 

[josh609]

does anybody have thoughts? I do, all the time.........

 

[dmens]

So much for the absolute necessity of an on-die memory controller even for a server platform.

 

[josh609]

Originally posted by: dmens
So much for the absolute necessity of an on-die memory controller even for a server platform.

Soo not true. Just wait until Intel wants to add more than 1-2 cores, we'll see who's faster.

 

[mamisano]

WHat happens when they move from DP to MP? How do they implement the dual bus? Any ideas?

 

[Furen]

2CPU's review is useless. Comparing a dual-core chip that will come out in 4-6 months to AMD's top-of-the-line single-core from 2 YEARS AGO is pretty bad. I know they claim that they didnt have access to dual-cores but at least they could have benched a single Dempsey as well, just for a closer comparison.

That said the performance of the Bensley platform (the CPUs are just higher-FSB 65nm Paxvilles, I believe) is pretty good. Quad-channel DDR2 533 is very helpful, as is having dual 1066MHz buses. That said, Opterons should see a speed bump or two (more likely only one) before Bensley comes into the market so, though Dempsey looks very competitive right now, things may change before its mass-market introduction. Another thing to look at is the scaling to quad-cpu systems. Though dual-independent busses help a lot in a 2 CPU system, there should be a pretty pronounced performance hit when going quad-CPU, then again, the higher FSB clock should help with this.

What I'm truly concerned about is price. Not just the CPU price, though that is important, but the price of the whole thing. The "rumors" and horror stories about FB-DIMMs seem to be well founded, considering that the modules are actively cooled, so I wonder how this affects power consumption (neither of the two reviews included this). Lastly, I'm guessing there wont be any faster-clocked Dempseys, so this should be the highest-performance performance offering until Woodcrest's introduction in 8-10 months (and who knows how that will perform). It's very nice that Woodcrest will be a drop-in replacement for these, though.

All in all, though I think there will be a few drawbacks (most notably price), this is what Paxville should have been.

 

[Gamingphreek]

Originally posted by: Furen
2CPU's review is useless. Comparing a dual-core chip that will come out in 4-6 months to AMD's top-of-the-line single-core from 2 YEARS AGO is pretty bad. I know they claim that they didnt have access to dual-cores but at least they could have benched a single Dempsey as well, just for a closer comparison.

That said the performance of the Bensley platform (the CPUs are just higher-FSB 65nm Paxvilles, I believe) is pretty good. Quad-channel DDR2 533 is very helpful, as is having dual 1066MHz buses. That said, Opterons should see a speed bump or two (more likely only one) before Bensley comes into the market so, though Dempsey looks very competitive right now, things may change before its mass-market introduction. Another thing to look at is the scaling to quad-cpu systems. Though dual-independent busses help a lot in a 2 CPU system, there should be a pretty pronounced performance hit when going quad-CPU, then again, the higher FSB clock should help with this.

What I'm truly concerned about is price. Not just the CPU price, though that is important, but the price of the whole thing. The "rumors" and horror stories about FB-DIMMs seem to be well founded, considering that the modules are actively cooled, so I wonder how this affects power consumption (neither of the two reviews included this). Lastly, I'm guessing there wont be any faster-clocked Dempseys, so this should be the highest-performance performance offering until Woodcrest's introduction in 8-10 months (and who knows how that will perform). It's very nice that Woodcrest will be a drop-in replacement for these, though.

All in all, though I think there will be a few drawbacks (most notably price), this is what Paxville should have been.

Good post. In addition there is the fact that the Xeon's consume nearly 2x the power.

Not only that but the 2CPU review is useless. They use all synthetic benchmarks, among everything one that is made by Intel!?

Interesting results nonetheless though. It goes to show how good Intel still is

-Kevin

 

[Viditor]

This one really has me scratching my head...
1. Intel has ended their NDA on Dempsey earlier than ever before (by a long shot). While I agree that the review is not a very good one, just the fact that they released a Dempsey for review is wierd. This chip won't be available for a year!
2. Paxville is available TODAY, and Intel hasn't released samples to the usual reviewers...
3. For those who are familiar with the term Osborne Effect, I haven't seen such a clear case in quite awhile! Even the most rabid Intel fan would be an idiot to buy a Paxville at this point. Set aside how much better the Opteron is, now the Paxville has to contend with the fact that it will be obsolete and far outstripped, even BY IT'S OWN COMPANY within a year!

I wonder how many Xeons Intel will sell over this next year?

 

[Accord99]

Dempsey and the Bensley platform is scheduled for release in Q1. It's Woodcrest with Bensley that's scheduled for Q3 2006.

 

[imported_michaelpatrick33]

Originally posted by: Accord99
Dempsey and the Bensley platform is scheduled for release in Q1. It's Woodcrest with Bensley that's scheduled for Q3 2006.

Yes, I thought Dempsey was slated for around the end of March 06. It still is a pretty desparate appearing comparison since we will just about have Socket 940 DDR2 Opterons and higher speed grade about that time to have a real comparison. Testing a processr six months out to a middle of the road speed processor competitor is humorous.

A 246 single core Opteron with synthetic benchmarks as a comparison? Ooooooooooook

Edit: two 246 single core Opterons

 

[Viditor]

Originally posted by: Accord99
Dempsey and the Bensley platform is scheduled for release in Q1. It's Woodcrest with Bensley that's scheduled for Q3 2006.

Thanks for the correction Accord! Then it becomes even less clear why they developed Paxville...who's going to buy it?
Do you think that they are trying to keep clients from switching to Opteron by promising a better Intel platform early next year (at the expense of their own Paxville, mind you...)?
If Dempsey doesn't really measure up when it's released, will they have lost even more credibility (vis a vis The Boy Who Cried Wolf)?

I have had nothing but respect for the abilities of Intel's marketing, but this just doesn't seem to have been a very smart play...

 

[Viditor]

Originally posted by: michaelpatrick33
Yes, I thought Dempsey was slated for around the end of March 06. It still is a pretty desparate appearing comparison since we will just about have Socket 940 DDR2 Opterons and higher speed grade about that time to have a real comparison. Testing a processr six months out to a middle of the road speed processor competitor is humorous.

A 246 single core Opteron with synthetic benchmarks as a comparison? Ooooooooooook

Edit: two 246 single core Opterons

What will be interesting is whether or not AMD can get the Socket F Opterons out within a quarter of that...

 

[dmens]

The current dempsey silicon is still in very early steppings, and the FBD has not been stressed. The word is it will perform on par with contemporary opterons... I guess that is credible since FBD really helps. Also, the reason paxville is in the market is to fill a short gap... it did not take a lot of effort to shove it into the market. It will be off the market as quickly as it popped out.

 

[Viditor]

Originally posted by: dmens
The current dempsey silicon is still in very early steppings, and the FBD has not been stressed. The word is it will perform on par with contemporary opterons... I guess that is credible since FBD really helps.

I had guessed as much...that's why I was mentioning the Socket F Opterons (which are supposed to also be FBD).
An interesting thing I've noted about that though, the FBDs appear to require active cooling.
This will be a big issue for both AMD and Intel in the 1u/2u and blade servers...

Also, the reason paxville is in the market is to fill a short gap... it did not take a lot of effort to shove it into the market. It will be off the market as quickly as it popped out

By "short gap", do you mean "appease Dell"?

 

[dmens]

Oh yeah, FBD is quite toasty. That is why it is better to have it both ways...

I don't know if the main reason of the paxville project was to appease dell, but the term used when it was first announced internally was "premier customers". I guess dell is a big part of that. I don't really care either way because paxville got me a lot of free food and drink over the past year.