Originally Posted by Measter I must assume that when you said that 64bit OSs can address 16TB of memory, you meant current 64bit OSs, because the memory limit of 64bit is 18,446,744,073,709,551,616 bytes (17,179,869,184 GB).
Anyway, great article, glad I was correct when I chose to go for 2x2GB.
I was wondering whether I was insane or not... 64-bit procs should be able to address 2^64 bytes of RAM (16EB?).
Nice read... what workstations are you using over there that are so magical?
Originally Posted by Smeggy "If you notice, the percentage memory use actually goes down from 31 percent to just 12 percent."
ummm... yeah, but if you work out what that percentage translates to in actual memory, 31% = 634mb, 19% = just over 700mb, 12% = nearly a gig! Of course the percentage is going down when you are adding more memory, 10% of 100 = 10, but 10% of 1000 = 100 (over simplified i know :P)
Yes but the percentage memory available is still higher, even though Vista continually adds more resources which make Vista-haters think you have less memory to use. I was just trying to clarify this point.
Great article Richard! I'm going to be upgrading soon from 2gb of ddr2 6400, and this really convinced me. Very insightful :)
Is there any chance you could repeat some of the benches with some faster memory kits? I'm really curious what the advantage between running at 1150, 1066 and running at 800 is (at similar timings).
Originally Posted by Xtrafresh Great article Richard! I'm going to be upgrading soon from 2gb of ddr2 6400, and this really convinced me. Very insightful :)
Is there any chance you could repeat some of the benches with some faster memory kits? I'm really curious what the advantage between running at 1150, 1066 and running at 800 is (at similar timings).
yeah you have good point here. I really want to see the different between 800 and 1066 with some benches. That will be really really good.
Because if you think on it. With the same amount of money you can get (4GB 1066) OR (8GB 800), So it will make a different.
Would have been nice to see some tests with a virus scanner and some updating going on in the back ground as this is what I find bogs down my workstation in the office running server 2008 and my desktop at home running VIsta 64
Originally Posted by wuyanxu Intel is different from AMD. when you want to see 800Mhz vs 1066Mhz, are you asking for AMD or Intel system?
if it's Intel. i can tell you now that the 1066Mhz bandwidth will make it produce faster result. simple
Benchmarks are not done to say which one is faster, obviously i'm aware that 1150 > 1066 > 800. A benchmark is taken to clarify how much difference there is between different setups.
This article has helped me decide between 2gb or 4gb, based on price premium vs. real world performance gain. I am simply requesting some help in a very related matter: the memory speed, as this article only covers PC6400 memory. I'm decided on 4gb, but i'm not quite sure if it's worth forking out the extra cash for memory that is actually faster.
the more the better, the faster the better. it's always been like this, and it will always be like this.
get 1066 if benchmarking is all you'll need. otherwise stay with 800Mhz for affordable yet performing RAM.
it's like 8GB vs 4GB. if you want the best, you have to spend 2x the money to get 10% increase in performance. 1066 vs 800 is the same (just not as extreme), it's up to you.
Originally Posted by Xtrafresh Is there any chance you could repeat some of the benches with some faster memory kits? I'm really curious what the advantage between running at 1150, 1066 and running at 800 is (at similar timings).
When the C2D was launched, the performance difference in real world tests between several RAM speeds was pratically non-existant above the DDR2-800 speed.
C2D architecture really doesnt need more RAM bandwidth. Nehalem might require it, though.
Originally Posted by impar Greetings!When the C2D was launched, the performance difference in real world tests between several RAM speeds was pratically non-existant above the DDR2-800 speed.
C2D architecture really doesnt need more RAM bandwidth.
I'm not refuting or agreeing this, nor am I saying that it's changed, but the effect of memory amount and bandwidth on the "C2D" or C2Q "architecture" is entirely dependent on the northbridge.
I've recently added 4GB of Corsair (2x2GB) memory to my existing 2GB (2x1GB) of Corsair, totalling 6GB. They run in Dual Channel mode at the same timings and the extra 4GB of space has made Vista x64 run like a dream. Multiple Virtual O/S on top of Vista, with plenty of potential space. I bought the original 2GB a couple of years ago for £195. Today's 4GB cost a little over £60, which shows how affordable memory has become. One last note, Half Life 2 only consumes ~450MB of the 6GB.
so im just wondering from a speed point of view, right now i have 2x1gb of 800mhz ocz stuff. i wasthinking with the cheap prices should i get a 2x2gb kit and add it to what i have now making 6gb but different sizes or just leave it with the 4gb. btw it'd all be the same brand and speeds. since its worthless to sell i was under the impression i may aswell keep what i have and add 4gb to it.
However, the game variety in Sup Com is massive because you can have an obscenely large number of units on screen, and it's well documented that those with smaller memory footprints have simply run out of space when the system had to handle thousands and thousands of units.
Does that mean that because memory usage became so large as to spill over in to the page file the OS started to prefer to give the RAM to other less greedy processes?
Quote:
Originally Posted by Bindibadgi Because I just didn't have time unfortunately, even though I did plan to - this set of tests took a solid week to complete.
Fair enough. Does that mean this is part 1 of 2?
You tested with a Yorkfield (2x6MB L2) quad core on the P45.
Does testing on the P45 invalidate the results as everybody here will most likely be using the P35 or older?
On the Task switching (World in Conflict) test you are using a GeForce 9800 GTX with 512MB of RAM, maybe the time difference is because the 512MB of video RAM can't hold all of the data from both the game and Windows & Firefox? It's not unlikely that Superfetch has already filled the RAM on the 2GB system but as it isn't used to 4GB or 8GB it would not know what to do with such a large amount of RAM. The hard drive has a transfer rate of somewhere in the region of 100MB/s, if 512MB is written to it then 7.2 seconds is not unreasonable.
Does the Raptor win due to a superior access speed or transfer rate?
Which would win: a VelociRaptor or Samsung Spinpoint F1s in RAID 0/RAID 5?
Finally why on earth are you using quad core Xenons? For ECC support?
Originally Posted by K20 Does that mean that because memory usage became so large as to spill over in to the page file the OS started to prefer to give the RAM to other less greedy processes?
No, Sup Com used to simply crash but they've patched it now.
Quote:
Fair enough. Does that mean this is part 1 of 2?
No, I have other features to write. It would simply be the same conclusion with slightly different numbers I'd imagine. What we are showing is how Vista, programs and games use memory. It would have been nice for inter-platform comparisons, but it just wasn't feasible.
Quote:
You tested with a Yorkfield (2x6MB L2) quad core on the P45.
Does testing on the P45 invalidate the results as everybody here will most likely be using the P35 or older?
No, again, the performance platform is kept consistent, the only variable is memory size. P45 is merely an evolution of P35 etc, and the results would in theory be reproducible it's unfeasible to test every possible platform under the sun.
Quote:
On the Task switching (World in Conflict) test you are using a GeForce 9800 GTX with 512MB of RAM, maybe the time difference is because the 512MB of video RAM can't hold all of the data from both the game and Windows & Firefox? It's not unlikely that Superfetch has already filled the RAM as it on the 2GB system but as it isn't used to 4GB or 8GB it would not know what to do with such a large amount of RAM. The hard drive has a transfer rate of somewhere in the region of 100MB/s, if 512MB is written to it then 7.2 seconds is not unreasonable.
The video memory would be dedicated to the DirectX overlay, yes, so the system has to disable this, fetch Firefox from memory and load that into memory. Graphics memory and the PCI-Express bus is faster than memory access in general, which is again faster than hard drive. Since Firefox was open before we started World in Conflict it will just reside in memory unless the system runs out of memory and pagefiles it because it deems it low priority since the game it playing.
Quote:
Does the Raptor win due to a superior access speed or transfer rate?
Which would win: a VelociRaptor or Samsung Spinpoint F1s in RAID 0/RAID 5?
That's a question for another article - the point is a faster harddrive provides a significant performance increase, even if more memory is being used because Photoshop is dependant as much on the scratch disk as it is memory.
Quote:
Finally why on earth are you using quad core Xenons? For ECC support?
Because I can :p because it's awesomely stable, never misses a beat and doesn't flinch when I fill all the memory slots. It's designed as a workstation for a reason.
You tested with a Yorkfield (2x6MB L2) quad core on the P45.
Does testing on the P45 invalidate the results as everybody here will most likely be using the P35 or older?
Older chipsets are generally worse with memory. That's certainly the case with older performance chipsets disliking different mismatched memory etc. My P35 is currently running 2 very different pairs of 1GB DDR-800 for example. However my P5K Premium has huge problems with 4 identical DDR-1066 modules.
So it's best to use something in reviews that can cope with whatever memory you can stick in it. That's why it's so frustrating to find out that the P5Q Deluxe is unstable with 8GB...
In RAM terms can someone please explain what the difference between 32 and 64 bit compatibility, as in, why does 32 bit not go well with over 2 gig of RAM.
Great article and it confirms what i have seeing when i switch over from 2 to 4GB (3.5GB) under XP
I understand the decision to do only Vista 64 but i still think its a shame you dint do 2 vs 4GB under XP 32bit as it will stay for a long time a OS that's gone be used by a lot of people because
1) it delivers noting extra for most people so why upgrade to Vista,
2) there are people that don't like Vista period,
3) lattes technet meeting i was for the 2008 server role out there was not even one admin that was even thinking about upgrading to Vista all ware waiting on windows 7 and hoping it will be better then Vista.
Personally i still lean toward 2 as i like XP still better then Vista but will properly make the switch to Vista just because of DX 10 support for non gamers there is no reason to switch to Vista and for office PC's it even a big no no because of the cost of a vista role-out and license and hardware upgrade cost as there is no RL benefit over XP
I also missed a lot of background programs running on the test machine like a resource hog like Norton Internet security and MSN, Logitech key/mouse driver ect ect
Because a system whit only 17 task running is imho almost impossible for real live use.
Even do i understand that more background programs will pollute the test results a bid i still think it would be useful for RL usage Nr's
Comments 26 to 50 of 98
I was wondering whether I was insane or not... 64-bit procs should be able to address 2^64 bytes of RAM (16EB?).
Nice read... what workstations are you using over there that are so magical?
Yes but the percentage memory available is still higher, even though Vista continually adds more resources which make Vista-haters think you have less memory to use. I was just trying to clarify this point.
That's ACTUALLY a word now!
Woot!
Is there any chance you could repeat some of the benches with some faster memory kits? I'm really curious what the advantage between running at 1150, 1066 and running at 800 is (at similar timings).
yeah you have good point here. I really want to see the different between 800 and 1066 with some benches. That will be really really good.
Because if you think on it. With the same amount of money you can get (4GB 1066) OR (8GB 800), So it will make a different.
Thanks Xtrafresh to clear this out
if it's Intel. i can tell you now that the 1066Mhz bandwidth will make it produce faster result. simple
This article has helped me decide between 2gb or 4gb, based on price premium vs. real world performance gain. I am simply requesting some help in a very related matter: the memory speed, as this article only covers PC6400 memory. I'm decided on 4gb, but i'm not quite sure if it's worth forking out the extra cash for memory that is actually faster.
get 1066 if benchmarking is all you'll need. otherwise stay with 800Mhz for affordable yet performing RAM.
it's like 8GB vs 4GB. if you want the best, you have to spend 2x the money to get 10% increase in performance. 1066 vs 800 is the same (just not as extreme), it's up to you.
C2D architecture really doesnt need more RAM bandwidth. Nehalem might require it, though.
I'm not refuting or agreeing this, nor am I saying that it's changed, but the effect of memory amount and bandwidth on the "C2D" or C2Q "architecture" is entirely dependent on the northbridge.
Why wasn't this also tested on AM2(+)? Possibly with these http://www.ocztechnology.com/products/memory/ocz_ddr2_pc2_5400_am2_special_high_density_kit?
Because I just didn't have time unfortunately, even though I did plan to - this set of tests took a solid week to complete.
Does that mean that because memory usage became so large as to spill over in to the page file the OS started to prefer to give the RAM to other less greedy processes?
Fair enough. Does that mean this is part 1 of 2?
You tested with a Yorkfield (2x6MB L2) quad core on the P45.
Does testing on the P45 invalidate the results as everybody here will most likely be using the P35 or older?
On the Task switching (World in Conflict) test you are using a GeForce 9800 GTX with 512MB of RAM, maybe the time difference is because the 512MB of video RAM can't hold all of the data from both the game and Windows & Firefox? It's not unlikely that Superfetch has already filled the RAM on the 2GB system but as it isn't used to 4GB or 8GB it would not know what to do with such a large amount of RAM. The hard drive has a transfer rate of somewhere in the region of 100MB/s, if 512MB is written to it then 7.2 seconds is not unreasonable.
Does the Raptor win due to a superior access speed or transfer rate?
Which would win: a VelociRaptor or Samsung Spinpoint F1s in RAID 0/RAID 5?
Finally why on earth are you using quad core Xenons? For ECC support?
No, Sup Com used to simply crash but they've patched it now.
No, I have other features to write. It would simply be the same conclusion with slightly different numbers I'd imagine. What we are showing is how Vista, programs and games use memory. It would have been nice for inter-platform comparisons, but it just wasn't feasible.
No, again, the performance platform is kept consistent, the only variable is memory size. P45 is merely an evolution of P35 etc, and the results would in theory be reproducible it's unfeasible to test every possible platform under the sun.
The video memory would be dedicated to the DirectX overlay, yes, so the system has to disable this, fetch Firefox from memory and load that into memory. Graphics memory and the PCI-Express bus is faster than memory access in general, which is again faster than hard drive. Since Firefox was open before we started World in Conflict it will just reside in memory unless the system runs out of memory and pagefiles it because it deems it low priority since the game it playing.
That's a question for another article - the point is a faster harddrive provides a significant performance increase, even if more memory is being used because Photoshop is dependant as much on the scratch disk as it is memory.
Because I can :p because it's awesomely stable, never misses a beat and doesn't flinch when I fill all the memory slots. It's designed as a workstation for a reason.
Older chipsets are generally worse with memory. That's certainly the case with older performance chipsets disliking different mismatched memory etc. My P35 is currently running 2 very different pairs of 1GB DDR-800 for example. However my P5K Premium has huge problems with 4 identical DDR-1066 modules.
So it's best to use something in reviews that can cope with whatever memory you can stick in it. That's why it's so frustrating to find out that the P5Q Deluxe is unstable with 8GB...
I understand the decision to do only Vista 64 but i still think its a shame you dint do 2 vs 4GB under XP 32bit as it will stay for a long time a OS that's gone be used by a lot of people because
1) it delivers noting extra for most people so why upgrade to Vista,
2) there are people that don't like Vista period,
3) lattes technet meeting i was for the 2008 server role out there was not even one admin that was even thinking about upgrading to Vista all ware waiting on windows 7 and hoping it will be better then Vista.
Personally i still lean toward 2 as i like XP still better then Vista but will properly make the switch to Vista just because of DX 10 support for non gamers there is no reason to switch to Vista and for office PC's it even a big no no because of the cost of a vista role-out and license and hardware upgrade cost as there is no RL benefit over XP
I also missed a lot of background programs running on the test machine like a resource hog like Norton Internet security and MSN, Logitech key/mouse driver ect ect
Because a system whit only 17 task running is imho almost impossible for real live use.
Even do i understand that more background programs will pollute the test results a bid i still think it would be useful for RL usage Nr's