Originally Posted by article 'Intel to ship 1M quad-cores before AMD ships first It will be a while before Intel moves to a native quad-core architecture and the jury is still out on the Kentsfield processor. The two pairs of cores will communicate via the northbridge, so that raises one potential performance bottleneck.
This may be a dumb idea/question, as I don't understand the way a processor works as well as many others in the bit-tech forums, but:
In a regular Core2duo, are the two cores able to communicate directly with each other?
If they are, then I wonder if there is a way with Kentsfield to send threads that may want to share data/communicate with each other to a pair of cores that CAN communicate directly? For example, if I'm running some code that can only really run on 2 cores, then would that be confined to one of the core2duo's inside the kentsfield....or could it presumably send one thread to each of the core2duo cores and result in a slowdown? Would this be a matter of the OS knowing the difference between the 4 cores in the processor, or would it be some logic embedded within the Kentsfield package?
Originally Posted by EQC On the other hand, even today I think a gigabyte of Flash RAM costs around $15-20 (I believe they've said that Vista is optimized for 1GB of this technology). As long as they were smart about how the Flash integrated with the motherboard (ie: some sort of standard card, and not some sort of chip that is both proprietary and varies from motherboard to motherboard), then replacing the flash every 2-3 years wouldn't be too much of a burden. In 2 or 3 years, a gigabyte will probably run you $2-3...and by then, with OS Service packs and what not, you might be wanting to upgrade to 2GB anyway.
true but the socket etc will add cost and probably won't be on many budget boards - who'll notice before the waranty is up? I like the idea and am hoping for a socket too.
Originally Posted by EQC This may be a dumb idea/question, as I don't understand the way a processor works as well as many others in the bit-tech forums, but:
In a regular Core2duo, are the two cores able to communicate directly with each other?
If they are, then I wonder if there is a way with Kentsfield to send threads that may want to share data/communicate with each other to a pair of cores that CAN communicate directly? For example, if I'm running some code that can only really run on 2 cores, then would that be confined to one of the core2duo's inside the kentsfield....or could it presumably send one thread to each of the core2duo cores and result in a slowdown? Would this be a matter of the OS knowing the difference between the 4 cores in the processor, or would it be some logic embedded within the Kentsfield package?
Core 2 Duo / Conroe has a shared L2 cache with a bus arbitration unit that allows the two cores to communicate with each other via the high speed L2 cache.
With Preslersaurus (dual core via two Cedar Mill dies on one package), the cores communicated via the front side bus, meaning that there were situations where the front side bus was saturated. Kentsfield works in a similar way, in that there are two CPU dies on one package and they communicate via the northbridge/front side bus. However, unlike Presler, each of the dies has two cores.
I am assuming that if you are running code that is optimised for two cores, it will run on one half of a Kentsfield CPU (i.e. one of the two Conroe dies). If you are communicating between the two Conroe dies, you will need to go via the front side bus. However, with the introduction of Advanced Smart Cache and Smart Memory Access technologies, Intel managed to make memory access and cache management more efficient on a dual core chip. This means that there is less traffic traveling across the front side bus and leaves more room for the two pairs of cores to talk to eachother.
From my understanding, there are going to be very few applications where the front side bus is going to be saturated in the same way that it was with the Preslersaurus.
Ive always thought that having a gfx card in the form of a gpu socket (like cpu's) would be acool idea. Then you would only have to buy gpu's to upgrade.
Originally Posted by bubsterboo Ive always thought that having a gfx card in the form of a gpu socket (like cpu's) would be acool idea. Then you would only have to buy gpu's to upgrade.
I'm still wondering how the RAM will work for the extra socket in either Intel's or AMD's ideas...as graphics ram is much faster than system ram (and just about the same magnitude in many systems). Will the extra socket have it's own RAM slots that'll take GDDR3 (or 4) somehow? Or is the more direct connection with the CPU going to pull in enough slack that the faster RAM isn't even necessary?
But its not a natural number, computer numbers should all be powers of 2, and of course 0s and 1s.
That is a good point, The chip makers could make these with terascale throughput, but it is the other hardware that has become the limiting factor. Terabytes per second is going to be very difficult for Hard drives, so I guess we will be seeing alot more flash based and hybrid designs.
Originally Posted by Why do we need teraflop computing? One example was that if you are a financial trader, you rely on complex models of markets and data being computed, often overnight. With tera-scale computing, you could have those models generated for you in real time, allowing you to predict and simulate the market in an instant.
Until some hacker hacks this software, & since all the traders would need to use this or loose out: world wide market manipulation mommy, i'm scared
64 cores? Lets just start with 64 bit software.
Quote:
The cores themselves will probably be less complex than they are today, we were told, whilst retaining full compatibility with the x86 instruction set. The cores will be optimised for threaded performance at the expense of scalar performance.
Doesn't anyone at Intel remember the Pentium Pro? (It did poorly because it had great 32bit performance but poor @ 16bit (Windows was 16bit at the time))
Is not the robson memory going to be more similar to Volitile memory instead of non-volitile. IIRC ram does not wear out for several years if run properly. So I do not think you will ever need to replace the stuff? Unless I am completely wrong. But I do not see any reason for the built on stuff to be non-volitile at all.
Whatever the technology turns out to be. Vista will need it, When I run it on my rig the HD activity light never seems to go off for more than 3 seconds. With that kind of power usage, what are we to expect from vista on mobile computers? I get better battery life in 2k than xp, and with vista I am glad they are deciding for the built in memory.
Seriously, when it comes, we might be able to consider it seriously, because of the processing power boost. Not too sure about Robson, as Warrior rocker says, Flash wears out with time and use. (I know it's the same for my HDD, but meh)
Tim S, when Alan Wake comes out, would you be so kind to test it on the max setting with full AA and AF to compare between the performance of single core CPU, dual core CPUs and quad core CPUs?
Comments 51 to 74 of 74
This may be a dumb idea/question, as I don't understand the way a processor works as well as many others in the bit-tech forums, but:
In a regular Core2duo, are the two cores able to communicate directly with each other?
If they are, then I wonder if there is a way with Kentsfield to send threads that may want to share data/communicate with each other to a pair of cores that CAN communicate directly? For example, if I'm running some code that can only really run on 2 cores, then would that be confined to one of the core2duo's inside the kentsfield....or could it presumably send one thread to each of the core2duo cores and result in a slowdown? Would this be a matter of the OS knowing the difference between the 4 cores in the processor, or would it be some logic embedded within the Kentsfield package?
With Preslersaurus (dual core via two Cedar Mill dies on one package), the cores communicated via the front side bus, meaning that there were situations where the front side bus was saturated. Kentsfield works in a similar way, in that there are two CPU dies on one package and they communicate via the northbridge/front side bus. However, unlike Presler, each of the dies has two cores.
I am assuming that if you are running code that is optimised for two cores, it will run on one half of a Kentsfield CPU (i.e. one of the two Conroe dies). If you are communicating between the two Conroe dies, you will need to go via the front side bus. However, with the introduction of Advanced Smart Cache and Smart Memory Access technologies, Intel managed to make memory access and cache management more efficient on a dual core chip. This means that there is less traffic traveling across the front side bus and leaves more room for the two pairs of cores to talk to eachother.
From my understanding, there are going to be very few applications where the front side bus is going to be saturated in the same way that it was with the Preslersaurus.
I'm still wondering how the RAM will work for the extra socket in either Intel's or AMD's ideas...as graphics ram is much faster than system ram (and just about the same magnitude in many systems). Will the extra socket have it's own RAM slots that'll take GDDR3 (or 4) somehow? Or is the more direct connection with the CPU going to pull in enough slack that the faster RAM isn't even necessary?
:?
EDIT: NM, I now see we cant... :'(
If the Wii can play videos over the network, we have our solution.
Meh, I have no care for that, aslong as its sleek and stylish like the Wii.
That is a good point, The chip makers could make these with terascale throughput, but it is the other hardware that has become the limiting factor. Terabytes per second is going to be very difficult for Hard drives, so I guess we will be seeing alot more flash based and hybrid designs.
EDIT: Woops double post
Looks nice but good grief is 320p harsh for that poor game! The tornado got the point across well enough though.
64 cores? Lets just start with 64 bit software.
Whatever the technology turns out to be. Vista will need it, When I run it on my rig the HD activity light never seems to go off for more than 3 seconds. With that kind of power usage, what are we to expect from vista on mobile computers? I get better battery life in 2k than xp, and with vista I am glad they are deciding for the built in memory.
Teraflop computing : raytracing, watch out !
Seriously, when it comes, we might be able to consider it seriously, because of the processing power boost. Not too sure about Robson, as Warrior rocker says, Flash wears out with time and use. (I know it's the same for my HDD, but meh)