Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Sorry. I'm underwhelmed. It's basically beaten by i7-6950X (also 10C/20T) several times. Power consumption is VERY disappointing. And temps are ridiculous, especially while overclocking ('soaring beyond 100ºC').
Not even mentioning that price tag.....
Sandy Bridge Core i7-2600K review:
http://www.inpai.com.cn/doc/hard/138458.htm
Its updated: Intel to detail 8-core server chip:
Updated at 10:00 a.m. PDT with correction about launch of Nehalem-EX processor.
Intel is expected to announce details of an 8-core processor for the high-end server market next week.
The chip itself will not actually ship in systems until late 2009 or early 2010
Intel Said to Face Fine, Rebate Ban in Eight-Year-Old EU Case:
April 22 (Bloomberg) -- Intel Corp., the world’s biggest computer-chip maker, faces a fine by the European Union and a ban on rebates on sales to computer makers, said two people who have seen a draft decision in an eight-year-old antitrust case.
The 500-page draft was circulated to 27 national competition authorities over the past few weeks in preparation for a ruling in the case, two people with direct knowledge of the document said. The people asked for anonymity because the document isn’t public.
Intel has been...
http://www.bloomberg.com/apps/news?pid=20601087&sid=aZt2fIUTfXKI&refer=home
Lots of IDC market data up to 4Q07 in AMD confidential presentation
marked 4/30/08 here (slides 4-8):
http://se.sun.com/sunnews/events/sunexpo/pdf/AMD_Meeting_demands.pdf
Want a good laugh at AMD credibility? Have a look at slide 16. Also
compare slide 13 to 58. And much too much more entertainment value
to mention specifically
Lots of different slides from different dates mixed together in that document. Slide 13 is one of those typical non-saying "rough timeline" slides you find back in every (AMD) presentation, and slide 58 still features 'Montreal' which was cancelled quite a while ago.
Only slides 4-8 contain some interesting stuff.
For actual roadmap data we just have to go back a month and a half:
http://news.cnet.com/8301-13579_3-9938372-37.html
It is about 18 months too soon for a dual socket Poulson
system, also that would be IA64 not Intel64, so my vote
is for a quad socket Dunnington. It is due to be released
in a few months last I heard.
I think so too. A quick search gives dunnington launch dates from anywhere from summer to 2nd half '08, but all in all it seems that the launch is probably not that far away (and that dunnington cpu's/systems can and actually should be expected out there).
Dunnington systems appear in the wild?
http://www.realworldtech.com/forums/index.cfm?action=detail&id=91605&threadid=91605&roomid=2
-Genuine Intel(R) CPU @ 2.13GHz
-Intel64 Family 6 Model 29 Stepping 0
-24 cores and 16 GB of ram
'Puma' notebooks appear in pricewatch:
http://www.heise.de/preisvergleich/a343198.html
http://geizhals.eu/a343198.html
http://h10010.www1.hp.com/wwpc/de/de/sm/WF06a/21259-282919-282919-282919-12434714-82135756.html
Expected delivery/shipment - 1 week:
http://www.notebook.de/index.php?section=shop&group=613&productid=13065
The allocated memory is 8,388,648 MB = 8MB. Sure Windows pads this out a bit more in usage but 8MB is the allocated memory. Also using memory should play into Nehalem's hands even more but it doesn't and it sure as hell isn't 'cache-bound' as you so falsely copied verbatim from Doug ! As a chip designer you shouldn't be relying on that clueless clown for your information without checking !
It's widely known that cache size effects SuperPI scores greatly. To deduct Nehalems single threaded performance from this simplistic piece of software without taking into account the differences in caches etc. is kinda pointless:
Here is what happens when cache-size increases:
Nehalem like performance improvements!
Or should I say penryn like performance improvements?
(L2 cache 4MB -> 6MB):
Even prescott, which was generally viewed upon as slower as its predecessor northwood, and also had some changes in its cache configurations(same case applies to Nehalem), was faster in SuperPI:
Pentium 4C L1-8KB, L2-512KB: 45s
Pentium 4E L1-16KB, L2-1MB: 43s
Pentium 4EE L1-8KB, L2-512KB, L3-2MB: 41s
Nehalem like performance improvements!
Anandtech - The Nehalem Preview: Intel Does It Again
First keep in mind that these performance numbers are early, and they were run on a partly crippled, very early platform. With that preface, the fact that Nehalem is still able to post these 20 - 50% performance gains says only one thing about Intel's tick-tock cadence: they did it.
We've been told to expect a 20 - 30% overall advantage over Penryn and it looks like Intel is on track to delivering just that in Q4. At 2.66GHz, Nehalem is already faster than the fastest 3.2GHz Penryns on the market today. At 3.2GHz, I'd feel comfortable calling it baby Skulltrail in all but the most heavily threaded benchmarks. This thing is fast and this is on a very early platform, keep in mind that Nehalem doesn't launch until Q4 of this year.
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3326
Re: *** More AMD Execs Leaving
The funny thing is, that the people in that list were all very "fresh" AMD/ATi employees and/or were never part of (or responsible for) previous successes and/or failures from AMD.
In a contributional sense they probably scored higher on the income list than on the "I was the key to success" list.
I think AMD benefits more from smart engineering talent than those guys from "run your mouth" department.
If one has to suffer some losses it could be much worse.
Furthermore it seems to me (but that may just be my imagination) that some K7 involved (lead)people are placed in the vacant spots.
Re: Given all the new features you mentioned, Griffin seemed to be optimized for lower idle power. It's quite possible that the problem they are having is with TDP power.
Yes, it seems like Griffin (Puma) saves power whenever and wherever it can, when full performance isn't needed ('typical usage' I guess). IIRC AMD said puma will extend battery life with at least an hour on average. On the point that griffin under full load exceeds its predecessor although specifically designed to be a better alternative - I'll guess we'll find out soon enough.
Given that Brisbane is already about 40% behind Core 2 Duo in terms of IPC, it would take substantial effort to close that gap. Even K10 didn't close the gap, so I can't see Griffin doing any better.
Agreed.
AMD also said the Griffin CPU was an unchanged K8.
Well, I believe AMD said K8 was the basis for Griffin and made the following changes:
The only performance related aspects I can find back are (so yes there probably will be an IPC increase):
-DRAM prefetcher
-Improvements in DRAM efficiency
-Larger caches
-Improved HT bandwith.
But the bulk of the changes are done towards the reduction of power consumption.
If the slight increase in IPC whipes out any advantage in powerconsumption then I guess they've made some strange trade-offs.
This article claims higher power than previous Turion cores.
If you think about characteristics and features a mobile platform requires, the answer always boils down to power consumption. And according to our sources, power consumption is exactly AMD’s problem with Puma. It seems that AMD is in trouble and created a processor that actually consumes more power than its predecessor.
Another striking line:
An AMD source told us that "Griffin will not be able to touch Core 2 Duo", which is quite a bombastic statement coming from a senior source.
So..
You take a K8 and:
-Greatly improve multi-core power management (independent voltages and frequencies for each core).
-Optimize the memory controller for power efficiency (seperate power plane, lower voltage, deeper sleep etc. etc.).
-Add powermanagement to Hypertransport.
-Add better thermal control (throttle memory frequency etc.).
-etc.
Or in summary: You do everything you can, to get the power consumption down.
Starting point: the 65nm 2,4GHz K8x2 @ 35W (TL-68)
But when you finish you end up with a part that consumes more?
If that indeed would be true then they really f#ck#d up something.
The other alternative is ofcourse that the article is just filled with FUD from various sources:
-Article: ..take a guess what the main problem will be when AMD is trying to be getting those mobile design wins?
-Reality: ..Puma has garnered over 100 design wins from original equipment manufacturers around the world and expects to have systems available for launch in Q2 2008.
On the performance part it is/was already clear that griffin won't come near Core 2 performance. (Although griffin was said to increase IPC during a cebit interview with AMD).
Correction on Nehalem sample speeds:
It seems that Hexus.net mistakenly took a Harpertown system -running at IDF in shanghai- for a Nehalem box:
http://www.realworldtech.com/forums/index.cfm?action=detail&id=89005&threadid=88864&roomid=2
The speed that the Nehalem samples were running at, is 2.13GHz:
Probably still enough ;)
http://www.hwupgrade.it/articoli/cpu/1938/intel-developer-forum-spring-2008-day-1_2.html
AMD to cut 10% of work force:
Advanced Micro Devices Inc. said Monday it plans to cut its work force by about 10 percent by the end of the third quarter.
Sunnyvale-based AMD said it doesn't yet know how much it will take in restructuring charges because the reorganization details are still being finalized. AMD has about 16,420 employees.
The company also said it expects revenue for the first quarter to be about $1.5 billion, a 22 percent increase compared to the first quarter of 2007 and 15 percent decrease compared to the fourth quarter of 2007.
The decrease is due to lower than expected sales across all business segments, AMD said. The company previously anticipated first quarter revenue to decline in line with seasonality.
http://sanjose.bizjournals.com/sanjose/stories/2008/04/07/daily17.html
AMD to cut 10% of work force:
Advanced Micro Devices Inc. said Monday it plans to cut its work force by about 10 percent by the end of the third quarter.
Sunnyvale-based AMD said it doesn't yet know how much it will take in restructuring charges because the reorganization details are still being finalized. AMD has about 16,420 employees.
The company also said it expects revenue for the first quarter to be about $1.5 billion, a 22 percent increase compared to the first quarter of 2007 and 15 percent decrease compared to the fourth quarter of 2007.
The decrease is due to lower than expected sales across all business segments, AMD said. The company previously anticipated first quarter revenue to decline in line with seasonality.
http://sanjose.bizjournals.com/sanjose/stories/2008/04/07/daily17.html
Re: Is that an 8-core chip or 2 fours
1 socket quad-core, with SMT (aka "Hyperthreading") in action.
Nehalem samples already chugging along at 3.2GHz:
Launch frequenties possibly at up to 3.6GHz
Vinod Dham interview:
When Speed Was King: Vinod Dham and the Birth of the Pentium:
http://knowledge.wharton.upenn.edu/india/article.cfm?articleid=4270
AMD Appoints Nigel Dessau as Chief Marketing Officer
The new Henri Richard:
SUNNYVALE, Calif.--(BUSINESS WIRE)--
AMD (NYSE: AMD) today announced the appointment of Nigel Dessau as chief marketing officer (CMO). He brings 22 years of technology sales and marketing leadership experience at IBM and Sun, and will be responsible for AMD's global marketing, image and campaign strategies.
http://www.ad-hoc-news.de/CorporateNews/en/15991879/AMD+Appoints+Nigel+Dessau+as+Chief+Marketing+Officer
AMD makes massive layoffs
I guess Charlie in essence picked up something vague that was planned all along, added with some of his great vision ("Whispers also say") as a bonus.
Wasn't it to be expected that once AMD acquired ATI that there were people that had to go?
The first round was essentialy during the take-over (nov 06), this is the one after evaluation.
BTW, first round;
Charlie: AMD-ATI to cut 2,000 jobs: http://www.theinquirer.net/en/inquirer/news/2006/10/12/amd-ati-to-cut-2000-jobs
Outcome: AMD cutting 375 jobs: http://www.pcworld.com/article/id,127946-pg,1-RSS,RSS/article.html
HP too embarrassed to mention either AMD or Opteron by name
on the web page introducing DL785, its new 8 socket Barc box:
What a non-story;
Introducing the HP ProLiant DL785 G5:
The HP ProLiant DL785 G5, the newest and most expandable additon to the award winning HP ProLiant line, is an 8-socket x86 server, supporting up to 8 AMD Opteron Quad Core processors, 256GB of memory and 11 PCI-e I/O slots. With this highly scalable feature set the DL785 G5, is an ideal choice for the growing enterprise class database, consolidation and virtualization environments seeking to improve server utilization and reduce server sprawl, while continuing to leverage all the familiar and easy to use ProLiant management tools and options.
http://h18004.www1.hp.com/products/servers/proliantdl785g5/index.html
it looks like the Shanghai die is approximately 300/21 by
300/16 or 14 x 19 mm or 266 mm2 in area. That is only
7% smaller than Barcelona, a suprisingly small reduction
considering how little of Barcelona was occupied by the
L3 cache which was tripled in Shanghai.
Barcelona 2MB: 283 mm2
Potential Barcelona 6MB: 328 mm2
Shanghai 6MB: 266 mm2
Barcelona-6MB > Shanghai-6MB reduction: ~19%, at an "equal" number of transistors. That would obviously mean a very poor shrink, so they probably stuffed in some more transistors in there? Core?
IMO, 4-8GB is still too small. After you take into account OS and application overhead, you get maybe 1GB left in the case of the 4GB drive.
What do you think they run on that thing as an OS? Embedded (linux etc.) OS's may only take up as much as a few dozen MB's (even with applications).
I've heard it was a very poorly written integer application.It may actually do its numerics byte by byte with decimal digits in ASCII form but I might be misremembering that about a different program. Other implementations of PI
calculation work an order of magnitude or more faster
than SuperPI.
Thanks. It seems that the original 'programmer' (Kanada) is still quite into PI calculations:
ftp://pi.super-computing.org/
http://www.super-computing.org/
No. The x87 FPU does not support a 3.32 million bit FP data type.
Chipguy, do you have an idea at what part of the CPU it exactly stresses than? Apart from the fact that SuperPI uses the Gauss-Legendre algorithm, there's not that much more info to be found.
SuperPi 1M is a very strange kind of workload
Isn't SuperPI purely x87 FPU based?
so I guess the voltage reading is correct
Or deliberate? By that same logic?
CPU-Z voltage readings are known to fluctuate a lot, or to simply show it plain wrong.
First Silverthorne benchmark:
http://www.computerbase.de/news/hardware/prozessoren/intel/2008/maerz/erster_benchmark_intels_silverthorne/
Amazing. Finally one journalist gets it right.
"As X-bit Labs points out..."
So besides:
Ed Stroligo
George Ou
We now also have pinned down Anton Shilov as a confirmed ihub/si/aces/rwt/ars (etc.) scanner?
Re: Well, it's hard to deny, after hearing it from the horse's mouth! Perhaps you are linking to the final few days worth of these systems being available online. But either way, it has gone from rumor to confirmed fact, via Duke's article
From the horse's mouth:
Dell, the second-largest PC maker after Hewlett-Packard Co (HPQ.N), is still selling one AMD-based consumer desktop computer on its Web site, and the Web site will continue to offer AMD-based systems to business customers, spokesman David Frink said.
We regularly change the way we merchandise our product portfolio," Frink said. "Currently, this is the mix we've put in place for our consumer systems of all types."
Frink said the change affects "a few" Dell laptop and desktop computers offered on its Web site."
http://www.washingtonpost.com/wp-dyn/content/article/2008/02/08/AR2008020802431.html
Also: "About 80 percent of Dell's sales are made to business, government and education customers, and about 20 percent to consumers."
Dell is shifting its attention to the retail market in the consumer section, thereby having AMD in mind it seems. Focusing on the cheaper systems, and leaving the -much less sold- more expensive systems there to be purchased online.
In Dell's own words they called the developments: — "not even all that interesting." And adding to that: "We are committed to the AMD product lines as a long-term partner to provide the maximum choice for our customers."
http://direct2dell.com/one2one/archive/2008/02/08/45012.aspx
While I agree that taking 4 or 5 systems off of the website might make some nice blog headlines, I think it will barely have an impact on the cheques AMD receives from Dell.
Re: Dell Dumps Online Sales Of AMD Consumer PCs
Not even that is true:
What it seems like that Dell has done, is that it moved some (the sub $700 systems it seems) of it's consumer systems to retail outlets. Or as Dell describes it: "We adjust product offerings regularly"
Why would Dell want to sell prebuilt (added cost) systems? Because they think they can sell them.
Aren't the majority of systems sold the low(er) cost systems anyway?
When Dell started selling in stores they described it as: "our first step" into global retail. It looks like they've chosen to do that for a large part with (the lower cost) AMD systems.
Choose any AMD system -> Dell Recommends -> (Customize ->) Review & Add to cart.
Yes you can SHOP FOR THEM ONLINE, but there doesn't appear to be a way to ORDER THEM ONLINE. Such as, for instance, ADD TO CART, or CHECKOUT.
What are you talking about?
Here, done just a minute ago:
Dell Stops Selling AMD Online
Wow, what a lot of research have those guys done:
http://search.dell.com/results.aspx?s=gen&c=us&l=en&cs=&k=AMD+&cat=all
HP readying AMD quad-core 'Barcelona' servers:
"Hewlett-Packard has committed to Advanced Micro Devices' quad-core Opteron "Barcelona" processors in its Proliant DL585 servers, according to company documents.
Specifications for the DL582 G2 architecture include AMD Opteron 8200 series dual-core or 8300-series quad-core processors, Nvidia nForce Professional 2200 and 2050 chipsets, and the AMD 8132 chipset. Also: two 100MHz PCI-X slots, four PCI Express x4 slots, three PCI Express x8 slots, and two embedded multifunction gigabit network adapters."
http://www.news.com/8301-10784_3-9866698-7.html?tag=head
The post you linked to described an article on Barcelona.
It doesn't mention anything about AMD's 45 nm process
or any related changes in circuit techniques for its 45 nm
shrink of K8L.
Well, you responded to the following:
AMD's 45nm processors are expected to see core frequencies greatly increased, while L3 cache sizes will grow from 2MB to 6MB. Power consumption of the cache will also see a significant drop, noted the sources.
With the answer:
....Cache power consumption tends to be a pretty small
part of overall device power consumption so reducing it, even
significantly, is not a major return on investment.
So I linked to a post that essentially says (or better: reads like), that AMD had to do some concessions with respect to the cache design. (Which -as you pointed out- isn't really a surprise because of time constraints).
In summary: it seems that the "significant drop" in cache power consumption is the expression of something that should have been there all along without the alien timeschedule.
So "return on investment" doesn't seem to be the major issue here.
Asking it more clearly: Do you think that the things mentioned in the post I linked to (not the article, but the post. Like "or perhaps they can get the voltage lower than 0.8V on the cache now") are a plausible explanation for the digitimes claim "power consumption of the cache will also see a significant drop"?