Hardware will obviously catch up to this truly epic speed, but what about operating systems? Is it possible for RAM to reach a point when the OS throws its hands up in the air and says "SLOW DOOOOWWWN!" Or am I thinking this the wrong way?
Don't see how the OS would care. It simply does its job faster.
The question for this particular memory kit is whether anything can even use its available bandwidth. Whats the point of this memory kit if it doesn't add any tangible performance to your system.
I think you need to evaluate your premises. The memory isn't going to feed the CPU anything that's not requested, It'd be like wondering if a 2000 watt PSU will fry your system.
It depends entirely on the OS, eg. run IRIX on an Origin with 1024GB RAM, it'll hum along just fine. My mid-range system has 72GB.
Good hardware helps of course though. This is why professional systems generally use ECC/buffered RAM, for better reliability, and supporting those tasks which cannot tolerate errors (not so important for 3D games).
It depends entirely on the OS, eg. run IRIX on an Origin with 1024GB RAM, it'll hum along just fine. My mid-range system has 72GB.
Good hardware helps of course though. This is why professional systems generally use ECC/buffered RAM, for better reliability, and supporting those tasks which cannot tolerate errors (not so important for 3D games).
I'm inclined to agree. Only a minor & narrow example, but when I dropped my i7's RAM speed from DDR3/2000 to DDR3/1600, all individual subtests for 3DMark06 only dropped by 1%.
Are CPUs right now even bandwidth limited? With memory dividers and async CPU overclocking I don't really see the point paying $400 for a kit of RAM like I did for BH-5 back in 2003.
Well, anything outside L3 incurs a performance hit on access. Other than a fundamental architecture and layout change two parameters that would help access times are CAS (tCL) and tRCD, which is where DRAM typically makes the slowest advancements. With increased density there can be an improvement in some back to back reads (fewer different DIMM rank to rank delays), although these don't impact 24/7 systems to a discernible degree.
But these high speed RAM SKUs aren't improving latencies, they're trading latency for bandwidth, which is why I was asking if current CPUs are even bandwidth limited in the first place. What is that kit running, something insane like 9-9-9-24?
For an example of what I'm talking about, back when the K8s moved from 939 to AM2 and DDR to DDR2 as a result of that, performance actually decreased due to the increased latency involved with DDR2. 2-2-2-5 1T at 200MHz was better than 4-4-4-8 400MHz since the K8s weren't bandwidth dependent for the most part. At what point do the current CPU uarchs stop saturating the RAM controller? How latency sensitive are the same uarchs?
Spending the $$$ for top tier RAM has almost always been one of the lowest returns on investment when trying to squeeze more performance out, but now a days I see even less of a reason. If you wanted to run a 300MHz FSB for whatever reason, you pretty much had to splurge for some BH-5, CH-6 or some other high end RAM if you wanted to run 1:1 FSB ratios, and if you didn't run a 1:1 it wasn't worth overclocking to that point to begin with. So again, I'll ask the question, outside of a small subset of insane hardcore overclockers, what is the point? It's gotten so easy overclock the CPU independently of the memory and if they aren't bandwidth limited (I don't think they are considering the marginal benefits of tri-channel over dual channel) you're better off paying half the price for some DDR3 1600 with tighter timings.
I never said they were offering better latencies - merely pointed out that performance can be improved if tCL and tRCD are improved and also said that is where DRAM makes the slowest advancements.
Some pre-availability announcements are fine. But this is for a somewhat generic computer part that is not yet available, has no specs and no price, it's basically just a picture.
Am I the only one that seems to get alot of bad OCZ ram? You pay a premium price for it, you'd think each set is tested. But we get alot of DOA returns. I have much better luck buying Supertalent which has a lifetime warranty.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
19 Comments
Back to Article
quiksilvr - Tuesday, July 13, 2010 - link
Hardware will obviously catch up to this truly epic speed, but what about operating systems? Is it possible for RAM to reach a point when the OS throws its hands up in the air and says "SLOW DOOOOWWWN!" Or am I thinking this the wrong way?SandmanWN - Tuesday, July 13, 2010 - link
Don't see how the OS would care. It simply does its job faster.The question for this particular memory kit is whether anything can even use its available bandwidth. Whats the point of this memory kit if it doesn't add any tangible performance to your system.
SlyNine - Tuesday, July 13, 2010 - link
I think you need to evaluate your premises. The memory isn't going to feed the CPU anything that's not requested, It'd be like wondering if a 2000 watt PSU will fry your system.fausto412 - Tuesday, July 13, 2010 - link
on point there buddy!fepple - Wednesday, July 14, 2010 - link
Its conceivable (though unlikely) it could expose an existing race condition in the code :)fausto412 - Tuesday, July 13, 2010 - link
if trhe software doesn't take advantage of it then it's wasted moneymapesdhs - Monday, July 19, 2010 - link
It depends entirely on the OS, eg. run IRIX on an Origin with 1024GB RAM, it'll hum
along just fine. My mid-range system has 72GB.
Good hardware helps of course though. This is why professional systems generally
use ECC/buffered RAM, for better reliability, and supporting those tasks which cannot
tolerate errors (not so important for 3D games).
Ian.
mapesdhs - Monday, July 19, 2010 - link
It depends entirely on the OS, eg. run IRIX on an Origin with 1024GB RAM, it'll hum
along just fine. My mid-range system has 72GB.
Good hardware helps of course though. This is why professional systems generally
use ECC/buffered RAM, for better reliability, and supporting those tasks which cannot
tolerate errors (not so important for 3D games).
Ian.
fausto412 - Tuesday, July 13, 2010 - link
we need an article looking at how ddr3 performs in a high end lga775 system, i5 system and i7 system.i'm not convinced DDR3 is delivering yet in real world apps/games...and also we need an overclocking ddr3 article. make it happen anandtech
Rajinder Gill - Tuesday, July 13, 2010 - link
Take a look here for DDR3 performance on i7 - this answers most of what one needs:http://www.anandtech.com/show/2792
Nothing much has changed over the past year since this article went up, other than the slow introduction of 4GB modules.
Later
Raja
mapesdhs - Monday, July 19, 2010 - link
I'm inclined to agree. Only a minor & narrow example, but when I dropped my i7's
RAM speed from DDR3/2000 to DDR3/1600, all individual subtests for 3DMark06
only dropped by 1%.
Ian.
rscoot - Tuesday, July 13, 2010 - link
Are CPUs right now even bandwidth limited? With memory dividers and async CPU overclocking I don't really see the point paying $400 for a kit of RAM like I did for BH-5 back in 2003.Rajinder Gill - Tuesday, July 13, 2010 - link
Well, anything outside L3 incurs a performance hit on access. Other than a fundamental architecture and layout change two parameters that would help access times are CAS (tCL) and tRCD, which is where DRAM typically makes the slowest advancements. With increased density there can be an improvement in some back to back reads (fewer different DIMM rank to rank delays), although these don't impact 24/7 systems to a discernible degree.Later
Raja
rscoot - Wednesday, July 14, 2010 - link
But these high speed RAM SKUs aren't improving latencies, they're trading latency for bandwidth, which is why I was asking if current CPUs are even bandwidth limited in the first place. What is that kit running, something insane like 9-9-9-24?For an example of what I'm talking about, back when the K8s moved from 939 to AM2 and DDR to DDR2 as a result of that, performance actually decreased due to the increased latency involved with DDR2. 2-2-2-5 1T at 200MHz was better than 4-4-4-8 400MHz since the K8s weren't bandwidth dependent for the most part. At what point do the current CPU uarchs stop saturating the RAM controller? How latency sensitive are the same uarchs?
Spending the $$$ for top tier RAM has almost always been one of the lowest returns on investment when trying to squeeze more performance out, but now a days I see even less of a reason. If you wanted to run a 300MHz FSB for whatever reason, you pretty much had to splurge for some BH-5, CH-6 or some other high end RAM if you wanted to run 1:1 FSB ratios, and if you didn't run a 1:1 it wasn't worth overclocking to that point to begin with. So again, I'll ask the question, outside of a small subset of insane hardcore overclockers, what is the point? It's gotten so easy overclock the CPU independently of the memory and if they aren't bandwidth limited (I don't think they are considering the marginal benefits of tri-channel over dual channel) you're better off paying half the price for some DDR3 1600 with tighter timings.
Rajinder Gill - Thursday, July 15, 2010 - link
I never said they were offering better latencies - merely pointed out that performance can be improved if tCL and tRCD are improved and also said that is where DRAM makes the slowest advancements.Later
Raja
levelup - Wednesday, July 14, 2010 - link
It would be great to have some new ram.NeonFlak - Wednesday, July 14, 2010 - link
pleaseGTVic - Wednesday, July 14, 2010 - link
Some pre-availability announcements are fine. But this is for a somewhat generic computer part that is not yet available, has no specs and no price, it's basically just a picture.bwave - Thursday, July 15, 2010 - link
Am I the only one that seems to get alot of bad OCZ ram? You pay a premium price for it, you'd think each set is tested. But we get alot of DOA returns. I have much better luck buying Supertalent which has a lifetime warranty.