M ANSARI wrote:It says more than 2 sockets ... how many sockets more I wonder? If 4 sockets then that is 32 cores sharing memory right there.
It will be available for 3 and 4 sockets, so 32 max.
Wow,I don't know about you and Majd,but I am already feeling dizzy
Dr.D
Without saying too much, I have seen some 100M nps numbers from a quad beckton prototype box. This thing looks very good. The Nehalem has 3 connections from a chip, which means a quad-socket will be a bit more efficient than a quad-socket AMD since their HT only has two connections and you end up with a "box" with each socket at one corner and the processor on the opposite corner is 2 hops away. nehalem has one 0-hop local memory and 3 1-hop remote memory configuration which is better. Not sure what they will do to go beyond 4 sockets however.
Do you think you could send me a protoype Bob?
Almost all of my playing with such boxes is done remotely. Only prototype I have had locally was the 2x nehalem box I was playing with a month ago to figure out a BIOS bug before we bought a small cluster using those nodes.
I'll take that one, as long as it doesn't come preloaded with Vista!
M ANSARI wrote:It says more than 2 sockets ... how many sockets more I wonder? If 4 sockets then that is 32 cores sharing memory right there.
It will be available for 3 and 4 sockets, so 32 max.
Wow,I don't know about you and Majd,but I am already feeling dizzy
Dr.D
Without saying too much, I have seen some 100M nps numbers from a quad beckton prototype box. This thing looks very good. The Nehalem has 3 connections from a chip, which means a quad-socket will be a bit more efficient than a quad-socket AMD since their HT only has two connections and you end up with a "box" with each socket at one corner and the processor on the opposite corner is 2 hops away. nehalem has one 0-hop local memory and 3 1-hop remote memory configuration which is better. Not sure what they will do to go beyond 4 sockets however.
Do you think you could send me a protoype Bob?
Almost all of my playing with such boxes is done remotely. Only prototype I have had locally was the 2x nehalem box I was playing with a month ago to figure out a BIOS bug before we bought a small cluster using those nodes.
I'll take that one, as long as it doesn't come preloaded with Vista!
Yeah...I do already have a Vista machine.
we don't buy any HPC (high performance computing) machines with windows, all of them run Linux.
M ANSARI wrote:It says more than 2 sockets ... how many sockets more I wonder? If 4 sockets then that is 32 cores sharing memory right there.
It will be available for 3 and 4 sockets, so 32 max.
Wow,I don't know about you and Majd,but I am already feeling dizzy
Dr.D
Without saying too much, I have seen some 100M nps numbers from a quad beckton prototype box. This thing looks very good. The Nehalem has 3 connections from a chip, which means a quad-socket will be a bit more efficient than a quad-socket AMD since their HT only has two connections and you end up with a "box" with each socket at one corner and the processor on the opposite corner is 2 hops away. nehalem has one 0-hop local memory and 3 1-hop remote memory configuration which is better. Not sure what they will do to go beyond 4 sockets however.
Do you think you could send me a protoype Bob?
Almost all of my playing with such boxes is done remotely. Only prototype I have had locally was the 2x nehalem box I was playing with a month ago to figure out a BIOS bug before we bought a small cluster using those nodes.
I'll take that one, as long as it doesn't come preloaded with Vista!
Yeah...I do already have a Vista machine.
we don't buy any HPC (high performance computing) machines with windows, all of them run Linux.
Well, I guess it's about time I learned Linux! I'll take it!
M ANSARI wrote:It says more than 2 sockets ... how many sockets more I wonder? If 4 sockets then that is 32 cores sharing memory right there.
It will be available for 3 and 4 sockets, so 32 max.
Wow,I don't know about you and Majd,but I am already feeling dizzy
Dr.D
Without saying too much, I have seen some 100M nps numbers from a quad beckton prototype box. This thing looks very good. The Nehalem has 3 connections from a chip, which means a quad-socket will be a bit more efficient than a quad-socket AMD since their HT only has two connections and you end up with a "box" with each socket at one corner and the processor on the opposite corner is 2 hops away. nehalem has one 0-hop local memory and 3 1-hop remote memory configuration which is better. Not sure what they will do to go beyond 4 sockets however.
M ANSARI wrote:It says more than 2 sockets ... how many sockets more I wonder? If 4 sockets then that is 32 cores sharing memory right there.
It will be available for 3 and 4 sockets, so 32 max.
Wow,I don't know about you and Majd,but I am already feeling dizzy
Dr.D
Without saying too much, I have seen some 100M nps numbers from a quad beckton prototype box. This thing looks very good. The Nehalem has 3 connections from a chip, which means a quad-socket will be a bit more efficient than a quad-socket AMD since their HT only has two connections and you end up with a "box" with each socket at one corner and the processor on the opposite corner is 2 hops away. nehalem has one 0-hop local memory and 3 1-hop remote memory configuration which is better. Not sure what they will do to go beyond 4 sockets however.
Hey Bob, any word about clockspeeds?
These were not anything "special" since we are talking prototypes. The last Dell Nehalem we had here was overclocking itself by 266mhz incorrectly, and I was seeing 24-25M nodes per second on a dual-socket quad-core system. It was nominally a 2.37ghz chip, but was overclocking when it should not. If we see becktons at 3ghz and up, 100M for Crafty will be on the low side of estimates...
I spent quite a bit of time on the 32 core testing trying to get popcnt to work. The gnu syntax is a pain to remember if you don't do it often. Turns out it was a percent or so improvement anyway as we have already optimized our use of popcnt().
I'll take it! 