This graphic at Wikipedia shows the big failure that is Itanium (a bit old):
And an very informative article at Arstechnica about the new SPARK and POWER 7 designs:
http://arstechnica.com/business/news...-niagara-3.ars
This graphic at Wikipedia shows the big failure that is Itanium (a bit old):
And an very informative article at Arstechnica about the new SPARK and POWER 7 designs:
http://arstechnica.com/business/news...-niagara-3.ars
which of these two were bigger fails? larrabee or itanic?
i mean that in total project cost.
intel is bleeding dollars, from failed tech to antitrust. i would even say they still pay off oems with "marketing" dollars just to keep amd products less appealing technically/aesthetically(keeping amd at budget segments only) or lower volume.
might intel be doing enron-type accounting?
are you really really sure you want to start talking about power consumption.... (my dear old friend shintel tried that a few times) there are very good tools for that you know once you work with large OEM's. Review sites are never 100% identical or fail to really match the same load settings etc, OEM don't because they can be sued for thatand they need to be made available for rack power calculation.
Lets take one of the major players in server world: HP
go to there site and start calculating the power, since they have tools for that:http://h18004.www1.hp.com/products/s...sor/index.html
Now lets put something together, one of the most sold server is still the INTEL DL380 and AMD DL385 (equal to the dell 29xx series) now lets take a look at power:
config DL380: 2 x QC E5530, 4 x 4GB RAM LP (yes the tri channel is only a marketing hype), P410 512mb, 2 x72GB SAS 15K, 2 x 460W power redundant.
idle 148,4W
50% 209,3W
100% 272,3W
config DL385: 2 x HC 2435, 4 x 4GB RAM LP , P410 512mb, 2 x72GB SAS 15K, 2 x 460W power redundant.
idle 130,7W
50% 220,8W
100% 314,6W
now lets take the actual retail price equal to E5530:
config DL385: 2 x HC 2431, 4 x 4GB RAM LP , P410 512mb, 2 x72GB SAS 15K, 2 x 460W power redundant.
idle 130,7W
50% 214,5W
100% 293,1W
or the low power config
config DL380: 2 x QC L5520, 4 x 4GB RAM LP, P410 512mb, 2 x72GB SAS 15K, 2 x 460W power redundant.
idle 138,6W
50% 190,5W
100% 244,4W
config DL385: 2 x HC 2425HE, 4 x 4GB RAM LP , P410 512mb, 2 x72GB SAS 10K, 2 x 460W power redundant.
idle 126,7W
50% 185,2W
100% 245,9W
conclusion, idle and 50% load AMD is the better pick, only in 100% load the advantage goes to Intel and all equal in the power savings platform (but intel has the clear performance gain here). When AMD launches 4000-6000 series the power consumption will improve more thx to the new c-states, apml 2.0, stepping and soi and ddr3, intel will have to counter that with 32nm.
You are also comparing Intel's 2009 platform to AMD's 2006 platform.
You'll see a lot of innovation around power efficiency in our 2010 platforms.
Looking at absolute power consumption is not that useful because if that were the case, running an Atom processor as a server would give you the best power bills. You have to look at performance per watt. When you're buying a server, chances you're not just buying one, but multiple servers. If 2 Nehalems can do the work of 3 Opterons, which one would you buy?
So, Johan, you referred above, is not authority in the measurement of power consumption any more?
And instead of measuring the real power consumption you put some virtual calculator from HP, which somehow believes that the E5530 uses 95W at 100% load (although its TDP is 80W).
But wait ... Wasn't that HP who submited these results:
http://www.spec.org/power_ssj2008/re...909-00197.html
http://www.spec.org/power_ssj2008/re...811-00184.html
http://www.spec.org/power_ssj2008/re...407-00143.html
ProLiant DL385 G6 (2.6GHz, AMD Opteron 2435 processor) SPECpower_ssj2008 = 1,394 overall ssj_ops/watt, Active idle = 120W
ProLiant DL360 G6 (2.93GHz, Intel Xeon processor X5570) SPECpower_ssj2008 = 1,586 overall ssj_ops/watt, Active idle = 88.4W
ProLiant DL380 G6 (Xeon L5530) SPECpower_ssj2008 = 2,012 overall ssj_ops/watt, Active idle = 63.7W
Last edited by kl0012; 02-11-2010 at 09:14 PM.
first you talk about idle performance, now you drag spec into it, typical, it is well known (at least if you know a bit in server world that spec ain't just that all to take). And it is for sure known that spec benches favor the hyperthreading a lot when daily workflow is quite a bit different.
And no although i find anandtech reviews quite good, they are in no means able to compare platform power, i will trust my OEM supplier for that, afterall that is what you make rack power calculations on, not some spec or mixed environment like anandtech which is btw not the same baseline i provided.
btw i already told you that the L series intel is the best solution if you go for low power/performance.
and your platforms are not the same, different mem, different discs and most of all different sw used, so unable to compare.
Bookmarks