Advanced Micro Devices is working to ensure that no platform change will be required to move from its next-generation dual-core processors to quad-core processors. When AMD rolls out dual-core processors with built-in virtualization hooks midyear, the company also aims to show that quad-core processors can also run on the same socket on that platform, said Marty Seyer, senior vice president of AMD’s Commercial Business and Performance Computing, Microprocessor Solutions Sector unit.
very nice move from amd. same as with their dualcores (i hope to buy one once i have enough money
). it saves consumers money, and if they buy the latest M2 as soon as it is out, they’ll be able to upgrade it a few years later with a lot more power 😉
Quad core CPU’s…I will wait to be convinced when the perfomance reviews come out. My guess is that they won’t be that great..100%..80%..60%..40%
Like everything else in SMP, some applications types will show a decrease in performance while other will show linear (or better) scale in performance.
If you write office documents all day, quad core is not for you.
/If/ you use your machine for ray-tracing, quad core is definitely for you.
Don’t forget that this is a server solution (at least at first) and most server applications scale *very* nicely as the core/socket count increases.
Like everything else in SMP, some applications types will show a decrease in performance while other will show linear (or better) scale in performance.
If you write office documents all day, quad core is not for you.
/If/ you use your machine for ray-tracing, quad core is definitely for you.
What you’re overlooking (as I’ve seen many do concerning multicore/multi-cpu configs) is that the benefits aren’t just limited to the scalability of one application. You can achieve performance gains even from single-threaded apps by being able to run more of such apps at the same time and setting thread affinity so that each app gets its own CPU/core. There could also be benefits to the OS and runtime processes, which could yield more resources to applications. Most apps aren’t single-threaded these days however. They usually use one thread for UI updates and at least one thread for non-UI work.
”
What you’re overlooking (as I’ve seen many do concerning multicore/multi-cpu configs) is that the benefits aren’t just limited to the scalability of one application. You can achieve performance gains even from single-threaded apps by being able to run more of such apps at the same time and setting thread affinity so that each app gets its own CPU/core. There could also be benefits to the OS and runtime processes, which could yield more resources to applications. Most apps aren’t single-threaded these days however. They usually use one thread for UI updates and at least one thread for non-UI work.”
Which is pretty rare on servers; Most servers (and workstations) run a single type of software the was designed with SMP scaling in mind.
I doubt that neither Intel nor AMD will release quad-core for the desktop before multi-threading (and multi-processing) on the desktop matures a bit.
As for the merits of having SMP workstation, let me put this way: last non SMP workstation I had was a 486… Nuff said
(Since then: Dual P120, Dual P2/333, Dual P3/750, Dual Athlon XP/1700, Dual AthlonMP 2400 and now one Dual Opteron 248 and soon, Dual Opteron 275)
G.
P.S. AFAIR Windows does not support UI from a thread, forcing all UI work to be done from the message message handlers; Hopefully, by the time quad core on the desktop will be an option, they’ll fix this omission.
Which is pretty rare on servers; Most servers (and workstations) run a single type of software the was designed with SMP scaling in mind.
I doubt that neither Intel nor AMD will release quad-core for the desktop before multi-threading (and multi-processing) on the desktop matures a bit.
I think you’ll have quad-cores on desktops in 2 years max. Clockspeed isn’t scaling as much, and CPU vendors need new products. Plus desktops are taking on server roles in many cases. Centralized and peer-to-peer media processing will be big. Media processing in general (including games) will push more cores into desktop CPUs to handle more remote user tasks while keeping the local user’s tasks performant.
P.S. AFAIR Windows does not support UI from a thread, forcing all UI work to be done from the message message handlers; Hopefully, by the time quad core on the desktop will be an option, they’ll fix this omission.
I’m not implying that UI updates are handled from worker threads if that’s what you mean. My point is that many apps today perform non-UI operations, especially potentially long-running ones, on worker threads so as not to block the main application thread which processes UI messages, and thus provide the user with a better experience when using the app.
One thing that may be overlooked, however, is the need to change incorrectly written software. There exist a substantial class of multithreaded algorithms that, although not obvious, are correct for two threads of CPU control (shorthand: 2 CPUs) but not for greater than two CPUs. The use of such algorithms may result in substantial stability issues. Independently created algorithms (rather than those adapted from reliable sources and known to be valid for N>2 CPUs) are particularly prone to this.
Projects I’ve worked on in the past have been bitten by this once or twice.
The future good news, however, is that algorithms that work for N CPUs but not N+1 CPUs where N > 2 are vastly easier to recognize as such (i.e., obvious).
The article is accurate, but might lead to overconfidence.
I would say that those who will have the most problems are those which rely on that certain piece of software that runs 90% of the worlds computers – for the Solaris crowd – Solaris is a thread monster, and believe me, it won’t have any problem letting it all hang out when it comes to quad cores coming on stream.
What will be the decider will be whether vendors are willing to get off their behind and start soaking some of those massive profits they make off their software back into the foundation of their software – and whilst they’re at it, maybe they should do some house cleaning.
I agree, and certainly the big Unix vendors (Solaris, IBM, and HP) have the proper foundation. I’m certain that Linux does as well, and probably some of the BSDs.
Thus, the applications which run on these OSs and use the OS (and library) -provided synchronization mechanisms will likely run without problems.
However, there are two particular areas outside this, specifically those based on OSs which were not designed to support larger numbers of CPUs (which you certainly alluded to), and those applications which bypsass the OS to handle concurrency to achieve higher performance (in which the software I work with — a main-memory DBMS – falls).
There are two groups that need this warning; those that design and develop software in the two areas I mentioned above, and end users who may naively think that they may simply swap in a new AMD CPU and not see any stability problems as a result.
“can also run on the same socket on that platform, said Marty Seyer”.
Unlike Intel, which change the sockets, the chipsets, the steppings, the BIOS support every few moths, rendering users big loosers.
Both AMD and Intel have done some annoying socket changes.
Indeed, but socket changes on the AMD is not as often as with Intel 🙂
Indeed. That’s one of the reasons I stopped buying Intel products. That and the difficulty with remaining competitive in performance, power consumption, and price.
Can you say: “2 hour gentoo installation”?
(full installation with kde, X, and Firefox)
Can’t wait for the quad core cpus to come out, although I think they will hover at a high price point for quite some time because most consumers will be content with dual cores for a while
When I can get one for under a grand, I’ll buy one.
Edited 2006-01-28 00:59
before one core used two memory controllers
then two cores used two memory controllers
and now four cores on two memcontrollers
ofcourse this would still be better than the older smp systems were the cpus shared the same bus and one memcontroller.
but still memory must start to be an issue
in a server that has 16 gigs?
dude, having 4 processes running with simultaneity with 16 gigs of memory should not be a huge issue.
before one core used two memory controllers
then two cores used two memory controllers
and now four cores on two memcontrollers
ofcourse this would still be better than the older smp systems were the cpus shared the same bus and one memcontroller.
but still memory must start to be an issue
Direct link for this comment
AFAIK, AMD’s Socket F will doubt the memory channels per socket. I assume that AMD does this with quad-core in mind.
G.
So AMD has finally discovered the G-Spot, but seriously are there any OSes that can take adavantage of the multi-core design, after all thats what maters.
So AMD has finally discovered the G-Spot, but seriously are there any OSes that can take adavantage of the multi-core design, after all thats what maters.
Close to all – of them.
Beside big-iron Unixes (that are designed for Huge-SMP setups), both Linux and Windows (2K, 2K3) scale very nicely as the socket/core count increases. (Though, at least in my experience, Windows is less effective at handling >= 8 core setups, mostly because it relays more on multithreading performance and less on the more-NUMA friendly multi-processing, which is long time strong point of both UNIX and Linux.)
AFAIK BSD has mad tremendous work in making the 6.x tree more SMP friendly and improve both thread and process efficiency, which puts it in a honorable 3’rd place.
Sadly enough, OSX’s kernel is the less scalable of them all; seems the Apple still uses a huge lock on kernel operation making it the least SMP friendly of the lot.
In short, if you are using Linux, Windows or BSD. You should be safe on the OS side. (If you are using big-iron UNIX, you are mostly likely running it on a 16 core machine right now, so this discussion shouldn’t interest you too much
)
G.
I’ll be curious to see what Intel releases in response to this. They are still getting the crap kicked out of them by AMD so it should be interesting.