Deneb preview

[harpoon84]

http://nl.babelfish.yahoo.com/...t_en&btnTrUrl=Vertalen (translated version)

In short, roughly 5 - 10% faster than Agena per clock, and lower power consumption, especially under load - at 2.3GHz, it consumes 57W vs 104W when compared with a Phenom 9650. All in all, a solid job from a die shrink, couldn't have asked for much more. :thumbsup:

 

[nbowman]

looks good, hopefully we will be seeing these in quantity soon.

OT, but babelfish translations make me smile, particularly "Pirate ship DDR2-800" and "Seagate Barracuda (hopes the Czechoslovakia cool fish) "

 

[harpoon84]

I kind of ignored the translation (if you can call it that) and moved right onto the graphs...

 

[Kuzi]

Wow, power consumption numbers for the low clocked Deneb @ 2.3Ghz are amazing, much better than expected.

Performance seems to have increased by only 5-9%, lower than expected, still it's better than nothing I guess. But that means it will still be a bit slower than Yorkfield.

 

[Kuzi]

The Deneb ES samples (like the one tested here) are running the IMC at 1.8Ghz. I hope retail Deneb will have the IMC running higher than that, especially for the faster clocked Black Edition CPUs. It will help improve performance for sure.

 

[harpoon84]

I guess if Deneb can get the IMC/L3 in sync with the core clock, performance could conceivably go up another ~5% or so, which should be enough to give it per clock parity with Yorkfield.

 

[thilanliyan]

When does this come out? Also, will it be compatible with current AM2+ motherboards?

I don't feel like spending an arm and a leg to get Nehalem so I think for my next build I'll go to AMD...I don't need 8 logical cores right now anyway.

 

[taltamir]

did they go over to a smaller process? cause this is a MASSIVE decrease in power consumption. it consumes only 54% of the power...

This think is sweet. sweet enough to own.

 

[OCGuy]

Originally posted by: thilan29
When does this come out? Also, will it be compatible with current AM2+ motherboards?

I don't feel like spending an arm and a leg to get Nehalem so I think for my next build I'll go to AMD...I don't need 8 logical cores right now anyway.

The choices for your next build wouldnt be AMD or Nehalem, it would be AMD or Yorkie, which...well..see above.

 

[nyker96]

the reviews translated are hilarious, c2d => cool wisdom 2 series. I love these translted reviews :] thanks!

looks good that they are able to clock to 3.2Ghz now with the 45nm transition. I still say AMD needs to change the core to new revisions that can allow higher clocks. Intel chips already doing 4 at 45nm. But still 3.2 quad on AMD is excellent and the power consumption is like much less compare to the 65nm parts. Good to see AMD making progress, we need this competition!

 

[VirtualLarry]

Sounds good to me, hopefully AMD can turn a profit with these babies.

 

[Cookie Monster]

Its impressive seeing at this is a C0 stepping. C2 is probably the retail version i believe so i think people can expect further increases in performance.

 

[nerp]

Wow.

I might be building a Deneb box down the road after all.

 

[ArchAngel777]

If they can just get the clock speeds up, it might not be a bad alternative to Intel. I always did like AMD, but the performance of Barcelona was just flat out terrible. I think the X2 did better clock for clock, or at least on par. That is pretty sad when that happens.

I am hoping that AMD can succeed here as I would love to have an AMD based rig again. come to think of it, the rig I have now is the first Intel rig I had in like... 5-6 years.

 

[thilanliyan]

Deneb WILL be AM2+ compatible apparently:
http://www.fudzilla.com/index....=view&id=8236&Itemid=1

Well hopefully fudzilla is right...yes yes I know it's fudzilla.

 

[Sylvanas]

Yes, Deneb is 45nm and it has been around for a while (even on AMD's roadmap) that it will be compatible with current AM2+ boards. It seems that 6mb L3 cache is going to come in handy, this is looking awesome especially with the advent of the 790GX and SB750 .

 

[taltamir]

they should call it the phenom 2 quad

 

[Nathelion]

Originally posted by: taltamir
they should call it the phenom 2 quad

I wonder how that would turn out in the translator. "Splendiferous prescience 2 series" maybe?

 

[bryanW1995]

yeah, um, this looks kinda cool, I guess....soooo, anybody know when the yorkie price cuts kick in???

 

[Comdrpopnfresh]

http://en.hardspell.com/doc/showcont.asp?news_id=3858
here is similar, already translated link

 

[Extelleron]

Originally posted by: bryanW1995
yeah, um, this looks kinda cool, I guess....soooo, anybody know when the yorkie price cuts kick in???

Should be sometime this month, but if I were you I wouldn't be jumping on a Yorkfield at this point.

The Q9550 at $316 isn't such a bad deal, but the cheapest Yorkfield I would be interested in is the Q9650 with the 9x multiplier, and that is still $530. In a few months, you will be able to buy a 2.66GHz Nehalem for $284 that will blow that $530 Q9650 away in performance. And with Deneb coming out and being pretty competitive with Yorkfield, you can expect prices to drop considerably.

So IMO at least, if you are looking at a Yorkfield right now..... wait. Wolfdales are the best deal right now.

 

[allies]

Hopefully this is a sign of good things to come from AMD's CPU division. I'm probably not building again until late '09 early '10... when's the 'Dozer coming through?

 

[Idontcare]

Originally posted by: Kuzi
Wow, power consumption numbers for the low clocked Deneb @ 2.3Ghz are amazing, much better than expected.

Performance seems to have increased by only 5-9%, lower than expected, still it's better than nothing I guess. But that means it will still be a bit slower than Yorkfield.

The concern I have here is that lowering clocks to pointlessly low values is guaranteed to produce equally pointlessly low power consumption values.

A graph of power consumption versus operating frequency across >3 data points would have been helpful in actually aiding any attempts to make conclusions of any kind here. (see http://www.anandtech.com/cpuch...howdoc.aspx?i=3184&p=2 for example of what I mean)

The second concern I have is that they include 3.2GHz Deneb in all the performance benches but specifically exclude it from the power consumption test. Uh that wierd feeling in your head when you think about it is what we call a "big red flag". Surely they could have tested the power consumption of their system when it was setup to run thru those 3.2GHz benches and I have no doubt that they did this. Why they chose to "exclude" the 3.2GHz results from their graphs on power consumption is leads me to unhappy thoughts.

 

[Extelleron]

Originally posted by: Idontcare

Originally posted by: Kuzi
Wow, power consumption numbers for the low clocked Deneb @ 2.3Ghz are amazing, much better than expected.

Performance seems to have increased by only 5-9%, lower than expected, still it's better than nothing I guess. But that means it will still be a bit slower than Yorkfield.

The concern I have here is that lowering clocks to pointlessly low values is guaranteed to produce equally pointlessly low power consumption values.

A graph of power consumption versus operating frequency across >3 data points would have been helpful in actually aiding any attempts to make conclusions of any kind here. (see http://www.anandtech.com/cpuch...howdoc.aspx?i=3184&p=2 for example of what I mean)

The second concern I have is that they include 3.2GHz Deneb in all the performance benches but specifically exclude it from the power consumption test. Uh that wierd feeling in your head when you think about it is what we call a "big red flag". Surely they could have tested the power consumption of their system when it was setup to run thru those 3.2GHz benches and I have no doubt that they did this. Why they chose to "exclude" the 3.2GHz results from their graphs on power consumption is leads me to unhappy thoughts.

The 3.2GHz was an overclocked result at very high voltage and really has no bearing on what we can expect from retail Deneb parts. Remember this a C0 chip, first revision out of the fab and is not even close to what you will see come 4Q.

In general we can assume that, as long as all parts of the chip are functional and operating as they should be (which seems to be the case given performance) that power consumption @ a certain frequency/voltage will be very similar to power consumption of the retail part.

But should we see a 3.2GHz Deneb CPU, the voltage will hardly be ~1.5V...... I would expect something around ~1.25V or so would be the max voltage seen on shipping 45nm CPUs, given that Agena CPUs are run at 1.312V max. Showing power results at that voltage would make the CPU look bad, and its not really fair to give people a bad impression of a CPU based on the first revision. So long as the power consumption results seen in this preview are similar to the retail chips, I don't think 3.0-3.2GHz at a 125W/140W thermal envelope is going to be out of the reach of the retail chips. We probably won't seen those speeds until 2009 though; perhaps such frequencies will coincide with the launch of DDR3 motherboards. Should AMD actually move to a HKMG process in 2009 and improve the performance of their process significantly, then should speeds should definitely be possible.

 

[Kuzi]

Originally posted by: Idontcare

Originally posted by: Kuzi
Wow, power consumption numbers for the low clocked Deneb @ 2.3Ghz are amazing, much better than expected.

Performance seems to have increased by only 5-9%, lower than expected, still it's better than nothing I guess. But that means it will still be a bit slower than Yorkfield.

The concern I have here is that lowering clocks to pointlessly low values is guaranteed to produce equally pointlessly low power consumption values.

A graph of power consumption versus operating frequency across >3 data points would have been helpful in actually aiding any attempts to make conclusions of any kind here. (see http://www.anandtech.com/cpuch...howdoc.aspx?i=3184&p=2 for example of what I mean)

The second concern I have is that they include 3.2GHz Deneb in all the performance benches but specifically exclude it from the power consumption test. Uh that wierd feeling in your head when you think about it is what we call a "big red flag". Surely they could have tested the power consumption of their system when it was setup to run thru those 3.2GHz benches and I have no doubt that they did this. Why they chose to "exclude" the 3.2GHz results from their graphs on power consumption is leads me to unhappy thoughts.

You are probably right, that the power consumption @ 3.2GHz for the tested Deneb sample may have been very high. But again as Extelleron mentioned it's an early C0 sample produced some months back.

Many people are able to OC their Phenom 9850/9950 to 2.9-3.1GHz on stock voltage, so it's believable to assume that retail Deneb (45nm) can OC to around 3.0-3.2GHz on stock voltage. I actually won't be surprised if some Deneb samples OC to 3.8-4.0GHz with some exotic cooling.

AMD has retail Deneb ready now, I just hope it does a bit better than the ES samples, especially it's performance, this is my big concern now. Maybe we can expect a Deneb preview from Anand soon?