backtop


Print 42 comment(s) - last by GoatMonkey.. on Jun 30 at 11:01 AM


Performance-per-watt results, Revision "G" quad-core detailed

Performance results for dual-core Opterons

Memo containing the details on how to decode the new model numbers
Death to single core!

AMD is set introduce Next-Generation AMD Opteron processors on August 1st, 2006. The Next-Generation AMD Opteron processors were previously known as AMD Opteron revision F CPUs. Following the release of socket AM2 processors the Next-Generation AMD Opteron comes in a new socket -- Socket F. Socket F sports 1,207 pins in a land-grid array similar to Intel’s LGA775 and LGA771 sockets. The new Socket F will be used for dual, quad and eight way processor configurations with up to 16 cores at launch.

AMD and Intel are both pushing efficiency this year, and AMD is already taking a shot at Intel with its dual and quad-core systems.  AMD claims its quad-core Socket F processor, scheduled for launch in 2007, will use the exact same power efficiency as the dual-core processors available today.  According to AMD's documentation (right), 90nm 95W dual-core Opterons (scheduled for launch this August) will have the same power draw as 65nm quad-core processors. 

Socket F AMD Opteron processors add support for DDR2 and AMD Virtualization. DDR2 with speeds up to 667MHz is supported while the AM2 Athlon 64 X2 and FX chips support DDR2 800MHz. FB-DIMM memory will not initially be supported by Socket F processors. AMD intends to add FB-DIMM support to its processors in the 2008 with the K8L architecture -- nearly two years after Intel launched support. Upcoming revision "G" quad-core processors will also support the Socket F. Previous features such as the integrated memory controller, HyperTransport Technology and AMD64 Execution will continue to be standard features on Socket F processor. With a new socket comes a new naming system too.

This time around AMD has moved to a four digit model number system. Three AMD Opteron families will be available at launch: 1000, 2000 and 8000. 1000 series processors will be Socket AM2 based and replace the existing AMD Opteron 100 series. 2000 series will be Socket F based and aimed towards dual processor systems. 8000 series will be available for four and eight way processor configurations. All three AMD Opteron series will be available as dual-core only. Socket F marks the death of single-core AMD Opteron processors.

Decoding the new model numbers is not too different from the existing system. The addition of a fourth digit simply adds generation designation. The first digit coincides to the series while the second digit coincides with the processor generation. Clock speeds are determined by the last two digits of the model number starting with 10 and increasing in increments of 2. An example of how the new model number works would be the "AMD Opteron 8218" -- which is an eight-way capable processor that’s a second generation Opteron design and clocked at 2.6GHz.

AMD is expected to introduce a full lineup of Socket F processors similar to what it’s done with socket AM2 processors. Socket F processors will be available in models x210, x212, x214, x216, x218 and x220 which are 1.8, 2.0, 2.2, 2.4, 2.6 and 2.8GHz parts, respectively. Regular, HE and SE AMD Opteron models will be available, though clock speeds will vary depending on the Opteron model. Thermal data power for regular AMD Opteron processors will be around 95 watts and 55 watts for AMD Opteron HE processors. Flagship AMD Opteron Model 2220 SE and 8220 SE have a 120 watt TDP while the Socket AM2 AMD Opteron Model 1220 SE has a slightly higher 125 watt TDP. All Opteron processors will have a 1MB of L2 dedicated to each processor core.

Availability is expected July 17th, 2006 under embargo from approved distributors while retail availability is expected August 1st, 2006.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Something is wrong here...
By kilkennycat on 6/21/2006 10:17:37 PM , Rating: 2
"Essentially, two 90nm 95W dual-core Opterons (scheduled for launch this August) will have the same power draw as a single 65nm 190W quad-core processor."

This statement does not seem to be correct. It implies that there is zero power benefit in going to 65nm.




RE: Something is wrong here...
By MDme on 6/21/2006 10:22:35 PM , Rating: 2
I was wondering about that, I would have thought that AMD may have implied that quad-core cpus would use the same TDP. but by saying same power efficiency, then it would mean that there is no improvement in TDP i.e. quad core cpu is no better than 2 dual core cpus in terms of total TDP.

Also, I hope AMD stops using "next-generation" as a buzz word for these CPUs since they're not really next gen. People might think that these "next gen" cpus compete with intel's next gen which is bad considering the prelim benches of kentsfield.


By KristopherKubicki (blog) on 6/21/2006 10:57:55 PM , Rating: 2
quote:
This statement does not seem to be correct. It implies that there is zero power benefit in going to 65nm.

Sorry, I clarified the text a bit. It actually implies that 2S systems with 65nm quad-core processors will have the same power efficiency as 2S systems with 90nm dual-core processors. Good news!


RE: Something is wrong here...
By hstewarth on 6/21/2006 11:30:30 PM , Rating: 2
agree - this is very missing leading information, no references to exact models. Where to they get 83Watt's for Woodcrest - the 3Ghz is 80Watt, the others are 65W and the low one is 40Watt.

It sounds like information to delay people purchasing the Woodcrests. It likely the same cpu in new socket and ddr2 - but same old technology.


RE: Something is wrong here...
By PrinceGaz on 6/22/2006 1:03:55 AM , Rating: 2
It clearly shows that 83 watts is for the memory for Woodcrest and other Intel CPUs, the two Woodcrest processors together use 180 watts (so 90 watts each). We all know Intel use a considerably more optimistic maximum power-consumption figure than AMD so they probably got the 90 watts for each Woodcrest bt taking Intel's "usual maximum" figure and converted it to AMD's "absolute maximum".

When both processors are using the same power-measurement methodology, it would seem Woodcrest is only marginally (~5%) more efficient than next-gen AM2 Opterons, and that the Opterons are actually more efficient once the additional power draw of the necessarily more complicated northbridge for Intel CPUs is taken into account. I'm not sure why the memory power consumption figures are so different between Intel and AMD as they're both using DDR2, but even excluding that it's clear AMD offers more performance per watt.

Looking further ahead the picture is even more bleak for Intel as their quad-core Clovertown is less efficient than the dual-core Woodcrest, probably in part because of the FSB bottleneck whereas AMD's quad-core Opterons are just as efficient as their dual-core models. It looks like AMD's future in the highly profitable server market will remain bright for the next few years whereas Intel will have to rely on marketing to minimise their fall in market share.

Intel may have a temporary lead in raw desktop performance with the fastest C2D models, but it's clear that AMD will continue to lead the way in servers and high-end workstations.


RE: Something is wrong here...
By coldpower27 on 6/22/2006 3:33:12 AM , Rating: 2

This are slides made by AMD to show AMD in the best light. Who knows what kind of FUD they are spreading...

Form TechReport's numbers a Woodcrest 3.0GHZ at full load consumes 59W, a far cry from the 80W TDP intel specifies for this model.


RE: Something is wrong here...
By coldpower27 on 6/22/2006 3:42:54 AM , Rating: 2

And somehow here AMD is saying that Woodcrest has a 90W TDP...ok....

Not to mention they are claiming 83W for the FB-DIMM's Intel uses, and I assume the 35W is for generic Registered ECC DDR2.

Very Very fishy to say the least.


RE: Something is wrong here...
By Viditor on 6/22/2006 4:17:46 AM , Rating: 2
quote:
Form TechReport's numbers a Woodcrest 3.0GHZ at full load consumes 59W, a far cry from the 80W TDP intel specifies for this model

This is why you shouldn't use TDP to guage power use...
What AMD is probably doing here is leveling the playing field and transposing TDP numbers...
For example, the A64 Venice 3800 has a TDP of 89w, but runs at 30w under full load.
http://www.lostcircuits.com/cpu/amd_venice/8.shtml

AMD appears to be trying to display a comparitive TDP, which I grant you is a mistake, but understandable in this case. TDP means nothing for actual power use, but most people just don't get that...so AMD needs to 'splain it to them.


RE: Something is wrong here...
By Viditor on 6/22/2006 4:28:18 AM , Rating: 2
Actually, if you read down to the bottom of the slice, it says it's based on max wattage, NOT on TDP.
Based on that, it seems to be correct.


RE: Something is wrong here...
By coldpower27 on 6/22/2006 5:11:23 AM , Rating: 2

So it's an unrealistic comparison. Hence of very limited use.


RE: Something is wrong here...
By Viditor on 6/23/2006 9:43:26 AM , Rating: 2
quote:
So it's an unrealistic comparison. Hence of very limited use

Actually, it's a better comparison than TDP is...
At least this way, the TDP is measured on the same scale (Max CPU Voltage).


RE: Something is wrong here...
By coldpower27 on 6/22/2006 5:24:51 AM , Rating: 2

Lost Circuits numbers are seriously in question.

http://www.lostcircuits.com/cpu/intel_p4-955/11.sh...

They put a Presler XE as hot as the Smithfield XE. That is completely not the case.

http://www.xbitlabs.com/articles/cpu/display/presl...

http://techreport.com/reviews/2006q1/fx60-vs-955xe...

Go to the tab with power consumption figures.
http://hothardware.com/viewarticle.aspx?articleid=...



RE: Something is wrong here...
By Viditor on 6/23/2006 9:39:45 AM , Rating: 2
quote:
Lost Circuits numbers are seriously in question

Actually, they are the only reviewers to actually test the chips directly by probing the rails themselves. The other reviews are overall guestimates...
BTW, those are power draw numbers, not temperatures.


RE: Something is wrong here...
By MartinT on 6/22/2006 2:21:41 AM , Rating: 2
Yeah, the whole first slide is idiotic to the extreme, bogus (estimated = made up) power numbers for the Intel systems and not a single f'in performance number on a slide explaining "A new standard in performance-per-Watt".

That is repulsive marketing BS from a company deeply in trouble.


RE: Something is wrong here...
By MDme on 6/22/2006 10:41:32 AM , Rating: 2
actually, they post the source of their numbers at the bottom.

Also, it is right for AMD to use the northbridge and the corrected memory power consumption because the platforms are different. Intel's systems use FB-DIMMS which are hotter and consume more power than DDR2 systems. Also, the northbridge on intel's systems have power consumption too. AMD's northbridge is integrated into the CPU.

The TDP numbers of intel are actually based on "usual usage scenarios, while AMD's TDP numbers are based on "maximum load" secenarios.

I'm not saying these numbers are accurate but you should look at them in the right context.


RE: Something is wrong here...
By ShapeGSX on 6/22/2006 10:44:57 AM , Rating: 2
I wonder how many DIMMs they gave each machine. I seem to remember previous AMD comparisons giving Woodcrest twice as many DIMMs as their opteron system. And why doesn't the Opteron have any chipset power listed? Despite having an onboard memory controller, Opteron still needs to use a chipset.


RE: Something is wrong here...
By Viditor on 6/23/2006 11:47:56 AM , Rating: 2
quote:
Despite having an onboard memory controller, Opteron still needs to use a chipset

But not a complete chipset...
The Northbridge for the Woodcrest includes the dual FSB, and the transmitter to the southbridge. Those are included in the Opteron already (cHT, HT). The power they're listing is for all of that combined at max power...


Low Power DDR2
By Questar on 6/21/2006 11:46:17 PM , Rating: 2
Since when did AMD produce memory?




RE: Low Power DDR2
By Clauzii on 6/22/2006 12:08:09 AM , Rating: 2
What gives You the idea that they do?

(Aside from flash-mem that is, if they still make that, not shure...)


RE: Low Power DDR2
By Questar on 6/22/2006 12:23:28 AM , Rating: 2
I was just pointing out that one of their marketing slides show they give better performance per watt because they use low power DDR2. Well, since they don't make memory this is kind of false. And there's nothing stopping anyone from putting low power memory in any system.


RE: Low Power DDR2
By RaynorWolfcastle on 6/22/2006 12:49:35 AM , Rating: 2
The top slide is awful in that it's totally unclear about what is being explained.

A) They're comparing DDR2 with FB-DIMMS with the light blue and green blocks, which is why they have lower power dissipation (this figure seems ball-parkish right, if a little exaggerated).

B) Obviously they're tacking on Intel's northbridge's power draw with the white blocks

C) Comparing CPUs directly to each other (2 socket systems, though their numbers are high even for that).

So AMD claims that a quad-core 65nm Opteron will have a TDP of just 95W. Compare this to current dual core CPUs which have a TDP of 95W and you'll see that this is a monumental achievement if based on just a die shrink and a few relatively minor architectural changes. I'll be interested in seeing whether they can actually pull it off.


RE: Low Power DDR2
By ChronoReverse on 6/22/2006 12:51:36 AM , Rating: 2
I'm being facetious but maybe they'll just switch over to the Intel method of measuring TDP.


RE: Low Power DDR2
By Griswold on 6/22/2006 5:05:10 AM , Rating: 2
They actually should. Makes their product look better. On the other hand, it would also spark some confusion when you compare the new figures to old ones on the same processor.


RE: Low Power DDR2
By Clauzii on 6/22/2006 1:09:20 AM , Rating: 2
I really don't hope we see a new generation of 'iWaterheaters' then :)


RE: Low Power DDR2
By Clauzii on 6/22/2006 1:17:08 AM , Rating: 2
Ahhh, You mean it looks like it's "AMD's DDR2 technoligy" that made the powerjump, right?


AMD's stupidity
By cnimativ on 6/22/2006 11:17:49 AM , Rating: 1
AMD talked so much trash about FB-DIMM, and yet they are going to implement it in 2008. Hypocrisy at its zenith.

Strip away the memories, AMD system consumes more power at a lower performance.

The RSA features of FB-DIMM is unmatched by ECC DDR2. Period.




RE: AMD's stupidity
By formulav8 on 6/22/2006 12:01:45 PM , Rating: 2
The only trash AMD talked about in regards to FB-Dimm is that Intel is FORCING it upon its customers. AMD is going to give the customer a CHOICE. They can either use DDR2 or FB-Dimms. AMD's has always had plans to use FB-Dimms, just down the road. And FB-Dimms run hotter than DDR2 and IIRC uses more power. Thats all I remember them saying anything bad about FB-Dimms.


Jason


RE: AMD's stupidity
By hans007 on 6/24/2006 6:46:07 PM , Rating: 2
you're the one that is a super fanboy apparently.

people actually want fb-dimm for one thing. amd doesnt give you A CHOICE either as none of their cpus support it. if you max out an opteron with tons of ram, since theres is no fb-dimm you have to run the ram at lower clocks.

that and intel could release a non fb-dimm chipset if they wanted to seeing as the non server based conroe cpus have perfectly fine ddr2 chipsets.

people in IT seem to like fb-dimm and besides fb-dimm will start using less power once the buffer chips are more refined and that will change considerably in the near future.


RE: AMD's stupidity
By Viditor on 6/24/2006 10:06:22 PM , Rating: 2
Actually, judging by the timing I'd say that AMD is planning on starting with FBD2 using DDR3.


RE: AMD's stupidity
By hstewarth on 6/22/2006 1:12:33 PM , Rating: 2
Yes there is a lot of stupid things against FB-Dimm, IE the following site even has FB-Dimm at 2008 even though FB-Dimm have been release.

http://endian.net/details.aspx?ItemNo=4185

It really stupid for AMD to trash FB-Dimm, when with this technology it will allow to take more advantage of 64 bit - because FB-Dimm's allow more memory on system.

Also people sure realized that FB-Dimm's use the same DDR2 and future DDR3 chips that current DDR2 Dimm's used.


RE: AMD's stupidity
By Jellodyne on 6/27/2006 4:57:17 PM , Rating: 2
Another way to allow more memory on systems is to tack a dual channel memory controller onto each processor, which gives you the memory scalability when you need it, ie in large scale systems.


RE: AMD's stupidity
By Viditor on 6/23/2006 11:51:53 AM , Rating: 2
quote:
The RSA features of FB-DIMM is unmatched by ECC DDR2. Period

I assume you mean the RAS features...they are only better if you have hooks for them. At this point, Intel does not.


Estimated performance
By hstewarth on 6/21/2006 11:34:08 PM , Rating: 2
I read the line that performance values are estimated value on DDR2-667 performance. What are they doing here, estimating performance.. if this chip is coming out they should know what the performance is not have to estimate it.

Very confusing.




RE: Estimated performance
By Clauzii on 6/22/2006 12:06:23 AM , Rating: 2
Right word might be 'expecting', since they can't predict/test each and every CPU/Motherboard combination.

But then - in danish, estimere (estimate) means "what we know and expect it to be" - so that might be exactly what they mean. It's just funny they don't seem more sure so close to launch :S


Some clarification please...
By Guuts on 6/22/2006 12:59:34 AM , Rating: 2
So the 1000 series will be the same AM2 socket that AMD just recently switched to....meaning that it will support DDR2-800 and not just DDR2-667 like the Socket F ones will? Is that correct?




RE: Some clarification please...
By Vysion on 6/22/2006 3:08:17 AM , Rating: 2
It actually looks like the article is saying that the Socket F AMD Opterons will only support up to DDR2-667 whereas the AM2 desktop processors (X2, FX, etc...) will support up to DDR2-800.

From what I understand AMD doesn't like that overclocking enthusiasts are buying up the Socket 939 Opterons for building desktop PC's. Maybe this is thier way of stopping this? If looking for the best performance for overclocking desktop rigs then we would be "forced" to buy X2's or FX's... <dunno>


AMD's CEO states 2008 for next generation cpus
By hstewarth on 6/23/2006 10:41:45 AM , Rating: 2
Slightly confused AMD's CEO states 2008 for next generation cpus - so what is up with this stuff.

http://www.reghardware.co.uk/2006/06/23/amd_ceo_sa...




By KristopherKubicki (blog) on 6/23/2006 10:53:03 AM , Rating: 2
There is "next-generation" Opteron, stuff that is slated for August 1 launch. Then there is "next-generation" architecture, which is what Ruiz is talking about.


Hm wait
By Regs on 6/21/2006 11:05:46 PM , Rating: 1
For a minute there I almost thought I was going to read something exciting.




RE: Hm wait
By Chiisuchianu on 6/21/2006 11:17:59 PM , Rating: 2
im no expert but i think this would be good news for over clockers :) :) very very good news... i think?


2x Quad Core = 2x Dual Core
By OCedHrt on 6/21/2006 10:48:25 PM , Rating: 2
I think AMD is saying that their quad cores will use the same power as a single dual core. If you look at the fineprint in the chart, "wattage based on 2P systems." Meaning, the quad cores are also 2P, so basically 8 cores at 65nm uses the same power as 4 cores at 90nm.




Registered memory
By GoatMonkey on 6/30/2006 11:01:13 AM , Rating: 2
How did I miss this topic? Anyway, I'm just wondering if AMD has said whether Registered memory is still required for Opterons. If not this would be a good alternative to the 4x4 platform.




"We shipped it on Saturday. Then on Sunday, we rested." -- Steve Jobs on the iPad launch

Related Articles
LGA Opteron F Spy Shots
June 9, 2006, 8:02 PM
Introducing AMD's AM2 Platform
May 22, 2006, 11:41 AM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki