-
Join the MyBroadband community
South Africa’s biggest forum. Discuss, discover, and connect with thousands of members.
-
Question of the Day: What's the most interesting small town you've visited in South Africa?
AMD six-core Opteron processor
- Thread starter rpm
- Start date
scotty777
...doesn't know
Sweet, but does this mean AMD is copying Intel's route? 
I'm very curious if AMD still has Bulldozer still on the boards
I'm very curious if AMD still has Bulldozer still on the boards
Messugga
Honorary Master
scotty777
...doesn't know
Hmmm, well that is interesting, I guess AMD does have a decent socket to play with (it's as universal as you can get these days). And the ship itself has enough room for 2 more buddies
.extremist
Well-Known Member
I recall reading a study that showed that if the general architecture of a computer remains the same and we just increase the number of cores on the CPU, one actually starts to see poorer performance from 8 cores onwards.
I believe they showed that at 8 cores the CPU performs about as well as a 2 core CPU and at 16 cores it performed poorer than a single core CPU. It has something to do with the overhead of divvying up the resources between more and more independent processing components.
And now for spending the hours it takes to track down the link to said article/study/research. If anyone else has seen this or has some insight into the matter, please post!
I'm trying to find out what these guys' findings were for a "strange" number of cores like 6. But I see AMD has 8 and 16 core CPUs on the table as well...
I believe they showed that at 8 cores the CPU performs about as well as a 2 core CPU and at 16 cores it performed poorer than a single core CPU. It has something to do with the overhead of divvying up the resources between more and more independent processing components.
And now for spending the hours it takes to track down the link to said article/study/research. If anyone else has seen this or has some insight into the matter, please post!
I'm trying to find out what these guys' findings were for a "strange" number of cores like 6. But I see AMD has 8 and 16 core CPUs on the table as well...
scotty777
...doesn't know
hmmm... well, the only way I think they going to avoid the effect that you mentioning (which I'm assuming is caused by a limited number of transactions that can happen between cpu and ram, thus when you add more cores, you limit the amount of data going to each processor, so each processor is either kinda busy, or not working) is they bump up the data transfer to and from the ram. AMD doesn't have to worry about this because they have a graphics card company under their belt (the gpu of graphics cards nowadays are limited by the memory transfers per second. hence why DDR5 was so wow). So maybe in 2 years time we will have DDR5 for your mother board available.
hyperian
Expert Member
- Joined
- Apr 17, 2008
- Messages
- 2,030
- Reaction score
- 168
It's also got to do with the fact that most programs and operating systems can not fully utilize all cores all the time. The chip architectures are pretty much at the mercy of how well the software that runs on them is able to utilize multiple cores.
chickenbeef
Executive Member
Just read on overclock.net. Photos of core i9 have surfaced. They are going to be 32nm six core processors with 12MB l3 cache, hyperthreaded and socket LGA1366
More cores is there to please the market. People want more and more. Time will tell whether more than 6 cores is better or not. I'm sure smarter architectures will be created to remove whatever bottlenecks exist between cores, etc.
Just curious, but what is actually better: more cores or more CPUs?
Just curious, but what is actually better: more cores or more CPUs?
chickenbeef
Executive Member
I would guess more cores. The more cores close together in one package the faster the data is transferred then having many cpus far apart
It all depends on the bus between them. My bet would be on more cores vs cpus. Then again you could get a dual socket motherboard with multicore cpus to make sure you get the best
scotty777
...doesn't know
It all depends on the bus between them. My bet would be on more cores vs cpus. Then again you could get a dual socket motherboard with multicore cpus to make sure you get the best
AMD's done that, it didn't do so well. Simple reason being that they only allowed certain CPU's to be paired. Also it means the motherboard needs another place to put the other CPU, and this means bigger motherboard (fine for servers, not fine for your average joe with a R500 case).
Also, you must remember that the communication between all the cores is limited with the more cores you have.
Interesting note, if the kept the same architecture (assuming transistor size remains the same), does that mean a 16 core cpu cover 4 times the size of the area? lol, that would be one big cpu
I know I might get shot for this, but maybe Intel has used the bigger 1366 socket for future plans on 8 core plus cpu's (even at 32nm process, an 8 core cpu would take up a lot of area)
Last edited:
The AMD implementation of multicore is much better than Intel. I dunno if things have changed as of late though. Before the fanboys get their nickers in a knot I have a q6600.
scotty777
...doesn't know
Well, AMD has all 4 processors on the same die. the intel core 2 quads have 2 core 2 duo's slapped together and put on the same chip (ps, I have a Q8200)
Sweet, but does this mean AMD is copying Intel's route?
I'm very curious if AMD still has Bulldozer still on the boards
As far as i know its for 2011.
http://www.fudzilla.com/content/view/15567/35/
AMD is counting on it to give Intel a real go.
CPUs look bigger externally than they are internally.
The size is dependent on number of pins/connections that need to be made. If the cores all do off-die communication via the same pins then more cores could be packed into the same packaging at the expense of something having to manage that communication trunk (ie. scheduling, etc).With trying to squeeze all that onto a single package, it might be best to go with multiple instances of multiple core CPUs.
I wonder when they will change the bus architectures to something difference. We've moved from parallel to serial... I wonder what's next... osmosis?
scotty777
...doesn't know
CPUs look bigger externally than they are internally.
With trying to squeeze all that onto a single package, it might be best to go with multiple instances of multiple core CPUs.
I wonder when they will change the bus architectures to something difference. We've moved from parallel to serial... I wonder what's next... osmosis?
Well, the current Bus system still has a few more kicks left. However, i think AMD/Intel/IBM should start looking into a new bus system AS WELL AS a new cpu interconnection system. What would be ideal is a new system where the CPU's all communicate on another level, to a point where they can calculate a process at the same time (so a quad core can use all the cores to run a single core app). This may need something on the software level, but if we just look at graphics cards when they are in SLI (two or more processors working in harmony).
lol, osmosis, well, I kinda don't see that working out very fast truth be told
. ha ha. However, they may need to come up with another system that can manage very high transfers to and from the bus (ram to cpu, cpu to graphics card or cpu to ram), other wise you won't be able to scale up the CPU's without bandwidth bottle neck (think of graphics cards again, the HD 4850 and HD4870 aren't very different in GPU clock, but the bandwidth on the HD4870 is massive, and that launches the graphics card into another level (less bottle necking from bandwidth))Does no one else see the irony of the names between the AMD processor (codename: Bulldozer), and the Intel processor (codename: Sandy Bridge).
I really hope that bulldozer does own the shyte out of Intel, you can see that if AMD can't keep up, then Intel is going to keep the i7 architecture for a very long time (they planning on taking this current architecture to 22nm, wtf?)
killadoob
Honorary Master
People say the core implementation is better but amd quadcore are still slower so explain to me how intel has implemented it poorly?
I understand amd have single cores making up the quadcore and intel use dual cores but why does this make the amd better? In what way does it make the amd quadcore better? In all tests vs intel they are behind so i am confused. Sure you can overclock each core but i still fail to see what makes it so much better than intel. It is not faster so how is it better? How is it implemented better?
That idea of yours is called threading. Program still needs to be built with it in mind. I understand modern OS do swap processes from core to core, but I see where you going with it.
If you think about it, it's like taking an instruction and making it into little ones which can all run at the same time. But then that's called RISC.
There've been a number of other bus and transport architectures over the years, most of them relegated to the likes of supercomputers. Things like Fibre Channel, Infiniband and NUMALink. When you see the stats on that stuff you're going to wonder why we have PCI-e.
Unfortunately, machines with those exotic architectures cost more than the average home PC so we make do with technology that is easy to mass produce.
Yeah, funny on the bulldozer vs sandy bridge thing. Let's see where the 22nm process goes. At those sizes there is plenty of power saving to be had, but understand that comes with some other nightmares.
As for AMD still sitting on the 45nm process; are they trying to find a manufacturing plant that can take things smaller?
Then again ... 45nm ... 22nm... meh.... it don't bug me none... I can barely see this 10pt font on my screen.
Program still needs to be built with it in mind. I understand modern OS do swap processes from core to core, but I see where you going with it.If you think about it, it's like taking an instruction and making it into little ones which can all run at the same time. But then that's called RISC.
There've been a number of other bus and transport architectures over the years, most of them relegated to the likes of supercomputers. Things like Fibre Channel, Infiniband and NUMALink. When you see the stats on that stuff you're going to wonder why we have PCI-e.
Unfortunately, machines with those exotic architectures cost more than the average home PC so we make do with technology that is easy to mass produce.
Yeah, funny on the bulldozer vs sandy bridge thing.
Let's see where the 22nm process goes. At those sizes there is plenty of power saving to be had, but understand that comes with some other nightmares.As for AMD still sitting on the 45nm process; are they trying to find a manufacturing plant that can take things smaller?
Then again ... 45nm ... 22nm... meh.... it don't bug me none... I can barely see this 10pt font on my screen.
I just bought a AMD quad. It's nippy enough for me.
Wish I had the spare cash for the competing Intel to see what the real difference is, but notepad.exe seems to open at the same speed on all PCs I use during the day. Which quads are you comparing there? I've seen tests with Intel leading AMD, then AMD leading Intel.