I think it already is no meaningful difference in most games, even with the integrated GPU! As with all memory performance tests, the real world difference is so small that it makes no sense to throw money at higher speced memory. Almost in all circuimstances, perhaps except some special cases like they highlighted in an earlier article hear at anand. The biggest difference there was for Winrar64 compression (not the usual unpacking we normal user do almost daily) and that was still only 20% differece between 1333 and 2400.
I haven't seen any benchies with memory for compilers (programing) or for DAW (music making, like with cubse, sonar and so on), but I suspect it is the same old story of almost no difference.
Second Life at max settings is a notable exception. I recently upgraded from DDR3-1600 to DDR3-2000 on my i5-760 @3.8ghz with GTX670, and saw a jump of 5-6fps in minimum framerates.
I COMPLETELY agree. This article is BOGUS. Here's why:
If you're going to spend money on higher end memory then you may as well fork over a little for a discreet graphics card instead. It will make much more difference in games.
Intel integrated graphics are only good for office usage still. No real gamer or anyone doing heavy GPU calculations, cares about Intel IGPs.
Show us what this more costly memory does for a real gamer and we'll consider purchasing it. Three frames per second more on an IGP is not worth the money spent on this memory.
That's the thing - not everyone that has a PC uses it for gaming. As alluded to by Magnus, there are other things that do not need a discrete GPU but are still used by a large number of enthusiasts - VMs, compilations, even non-parallel scientific simulations. You whack in a stupidly large matrix into memory and it will bog down. Failing that, how about financial calculations? Now put all that inside a mITX chassis and board where you're limited to two memory slots. Sure 2x8GB 2400 C11 may not be your first port of call, but if saving an extra 3-5% time on whatever you do actually has a financial impact on your work portfolio, then the early investment could pay off in the long run. Then again, it may not and that 2x4GB 1333 C9 is looking a little sweeter.
No article is bogus, as you put it. Sure there are good things to review and some that are not so good. If we solely focused on the good, then we'd all be sitting around patting each other on the back for doing a service to the industry. I like reviewing all sorts - you get to see the niggles of the smaller companies that can't invest, or you can point out when a top company is just being stupid. Thus when you get a really good product that shines out from the rest, it is something special to behold.
Ian
PS There are plans for a compilation benchmark in future reviews. I'm trying to organise a decent one that I can strap a timer to without sitting in front of the screen for 20 minutes waiting for it to finish.
True, not everyone uses a PC for gaming. But do you really think the "frost white" Evo Veloce memory is targeted towards the financial or enterprise sector? As silly as it is, Enthusiasts with money than sense WILL pay a premium to raise their benchmark synthetics.
Sticks like these coupled with a X79 motherboard with 8 DIMMs populated are biting into server territory - I'm sure! I'm running a 3930K + 48GB (I will soon have to upgrade to 64) as a dedicated SQL server. How much would similar performance cost me if I'd built the system around a Xeon? Probably twice.
Anyone who has actually tested higher RAM frequencies or RAM quantities above 4 GB. in a typical modern desktop PC knows that frequencies above 1600 MHz. produce no tangible system performance gain in either Intel or AMD powered PCs. Tangible means a change in system performance that you can actually see and or feel. In addition more than 4 GB. of RAM produces minimal gains unless you operate your PC with many applications open or functioning concurrently, which most people don't. CAD, modeling, and some other business applications CAN benefit from more RAM but many consumer apps don't.
Be advised that RAM benches grosssly mis-represent the actual SYSTEM gains because they assume the RAM is saturated 100% of the time, which it is not. If you run real applications and compare 1333 MHz. to 2400 MHz. you will likely not even be able to tell the difference between the two because DDR3 RAM @ 1333+ MHZ. is not a system bottleneck.
The RAM mfgs. are doing everything they can to convince enthusiasts to buy high priced high frequency RAM because this is very profitable for them. When you test with real applications and see how tiny the system gains are, you will wish that you had bought 1600 or 8166 MHz. RAM at a reasonable price and used the cash for a faster CPU or GPU.
If you have money burning a hole in your pocket then by all means buy the fastest RAM you can find. Then you can brag at how cool it is... even if you can't actually use the highest frequency in your PC. Understand that your CPU may not be able to run the RAM anywhere near it's rated frequency. Typically Deneb CPUs top out around 1600 MHz., Thubans around 1800 MHz., Zambezi at 2000+ MHz. and Vishera around 2400 MHz. based on initial testing for Vishera. No one cares about Intel CPUs so I won't post on them... <LOL>
Of course, if you have an APU-based system, the faster memory does indeed make a difference... though I agree, it's the exception rather than the norm.
But then its totally contrary to one of the main reasons behind having an APU -- penny pinching.
These kits cost twice the DDR3-1333 going rate, so that's $75 you could have put into a GPU. Can't speak for everyone, but I'd probably choose an i3 with DDR3-1333 + a 7750 over an A10-5800k with DDR3-2400.
Anandtech did a series of memory frequency tests like a year ago (I forget exactly). While they found that 1333 to 1600 didn't offer much in terms of average FPS gains in gaming, it had a clearer impact on minimum frame rates. I'm not saying it's worth it either way here, but I'd like people to give some attention to minimum frame rates when talking about the benefits of bumps in memory frequency.
That said, 2400 is obviously overkill here, but that should be obvious to anyone who wants to spend their money efficiently.
The article the did a year ago (with Sandy Bridge in mind) says absolutely nothing about minimum frame rates vs average... I don't even see how faster memory could have such an effect with a dedicated GPU.
pseudoking what are u talking about? there is virtually NO effect on minimum frames on a dedicated GPU system. Ever since the memory controller moved to the CPU, the RAM timings have become ALOT a less important component in the system. The only way it shows a difference is when you go to all kinds of outlandish scenerios that involve isolating the GPU and CPU to situations that show some difference between RAM, but in a real world setting those situations are so rare that it becomes pointless to even entertain them.
But add running virtual machines to your list of reasons why a lot of memory might be good. When working from home I've actually typically got the host machine where I'm doing most of my actual work plus at least two virtual machines running, each VPN'ed into a different remote network. So it isn't too uncommon for me to see about 90% of my 16 gb in use at any one time. And I do occasionally hit times when I have to shut down one VM in order to start another. So I wouldn't actually mind having 32 GB.
On the other hand, while I need a large quantity of RAM, my 1600 MHz G-Skill works just fine performance wise so I don't need speed - I need quantity.
Actually, yes you can tell the difference depending on the applications you run. In fact, the distributed OpenCL GPU computing projects I use over at BOINC, the biological research, requires extreme frequencies. The higher the processor frequency, the higher your GPU load and ability to install more GPUs for the project becomes. If I run my RAM at 1333Mhz with my 7950, I need to run 8 work units in order to get that memory to get 80% efficiency. After overclocking my RAM to 1800Mhz I was able to get the efficiency up to 95%. With faster memory, I could run less work units, and install a second graphics cards, although I would likely need 2133Mhz quad channel memory in order to saturate two 7950s in the HCC project. In POEM, I would actually need 4266Mhz quad channel DDR3 in order to saturate a single 7950 more than likely.
Another scenario is AMD APUs, where FPS almost scales linearly based on RAM frequency because it uses system RAM as VRAM. GPUs are very memory intense, which is why GPUs have 256-384 bit memory interfaces. With an APU, the GPU on the die is restricted to the lame 64 bit memory interface we have with our system RAM.
RAM kits like these are sold to people who need them, if you don't need them, buy the lower frequency RAM.
What's with the cost analysis of all these high speed $150 kits? Would you actually shell out that amount of money for a run of the mill gaming/enthusiast system when there's a ton of 1600 4x4GB or even 2x8GB kits on Newegg selling for $75-85?
I've wishlisted like half a dozen G.Skill kits within that range AND with blue spreaders specifically... If I width the search there's tons more obviously, just hoping one of those goes on sale during Thanksgiving for like $50-60, even tho I'm in no dire need to upgrade from 8GB.
(and yeah, looking at blue purely for aesthetic reasons obviously!)
I would because it would cut my research rate in half. I run OpenCL GPU projects on my systems, where if I had 2400Mhz memory I could output twice as many work units with my graphics cards per day in Help Conquer Cancer.
That's how I feel. I used to care about the color of my computer case, etc but I realized that its more about longevity and ease of installation for me. It's nice to go with a color theme, but the extra cost each time I upgrade just isn't worth it. So now that I've matured a bit I've decided on mostly black boxes and hardware as its plentiful and price competitive.
Memory companies will continue to put out insanely spec'ed pieces so they get reviewed by Hardware sites, because it's the only way to get their name out there. Otherwise, memory as a whole is a pretty boring component. Why else do they also need to have interesting names and brightly colored heatsinks? Marketing at its best (or is it worst?)
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.
Facebook
Twitter
RSS
30 Comments
Back to Article
Tchamber - Wednesday, October 24, 2012 - link
It would be nice to see how memory impacts gaming with a graphics card. Or does the difference get so small that there's no meaningful difference?Magnus101 - Wednesday, October 24, 2012 - link
I think it already is no meaningful difference in most games, even with the integrated GPU!As with all memory performance tests, the real world difference is so small that it makes no sense to throw money at higher speced memory.
Almost in all circuimstances, perhaps except some special cases like they highlighted in an earlier article hear at anand.
The biggest difference there was for Winrar64 compression (not the usual unpacking we normal user do almost daily) and that was still only 20% differece between 1333 and 2400.
I haven't seen any benchies with memory for compilers (programing) or for DAW (music making, like with cubse, sonar and so on), but I suspect it is the same old story of almost no difference.
tekphnx - Thursday, October 25, 2012 - link
Second Life at max settings is a notable exception. I recently upgraded from DDR3-1600 to DDR3-2000 on my i5-760 @3.8ghz with GTX670, and saw a jump of 5-6fps in minimum framerates.JonnyDough - Friday, October 26, 2012 - link
I COMPLETELY agree. This article is BOGUS. Here's why:If you're going to spend money on higher end memory then you may as well fork over a little for a discreet graphics card instead. It will make much more difference in games.
Intel integrated graphics are only good for office usage still. No real gamer or anyone doing heavy GPU calculations, cares about Intel IGPs.
Show us what this more costly memory does for a real gamer and we'll consider purchasing it. Three frames per second more on an IGP is not worth the money spent on this memory.
IanCutress - Friday, October 26, 2012 - link
That's the thing - not everyone that has a PC uses it for gaming. As alluded to by Magnus, there are other things that do not need a discrete GPU but are still used by a large number of enthusiasts - VMs, compilations, even non-parallel scientific simulations. You whack in a stupidly large matrix into memory and it will bog down. Failing that, how about financial calculations? Now put all that inside a mITX chassis and board where you're limited to two memory slots. Sure 2x8GB 2400 C11 may not be your first port of call, but if saving an extra 3-5% time on whatever you do actually has a financial impact on your work portfolio, then the early investment could pay off in the long run. Then again, it may not and that 2x4GB 1333 C9 is looking a little sweeter.No article is bogus, as you put it. Sure there are good things to review and some that are not so good. If we solely focused on the good, then we'd all be sitting around patting each other on the back for doing a service to the industry. I like reviewing all sorts - you get to see the niggles of the smaller companies that can't invest, or you can point out when a top company is just being stupid. Thus when you get a really good product that shines out from the rest, it is something special to behold.
Ian
PS There are plans for a compilation benchmark in future reviews. I'm trying to organise a decent one that I can strap a timer to without sitting in front of the screen for 20 minutes waiting for it to finish.
Runamok81 - Friday, October 26, 2012 - link
True, not everyone uses a PC for gaming. But do you really think the "frost white" Evo Veloce memory is targeted towards the financial or enterprise sector? As silly as it is, Enthusiasts with money than sense WILL pay a premium to raise their benchmark synthetics.ytoledano - Wednesday, October 24, 2012 - link
Sticks like these coupled with a X79 motherboard with 8 DIMMs populated are biting into server territory - I'm sure! I'm running a 3930K + 48GB (I will soon have to upgrade to 64) as a dedicated SQL server. How much would similar performance cost me if I'd built the system around a Xeon? Probably twice.Blibbax - Wednesday, October 24, 2012 - link
Roughly £400 extra for the Xeon equivalent of the 3930K.quixver - Tuesday, November 20, 2012 - link
You are missing out on ECC though.Beenthere - Wednesday, October 24, 2012 - link
Anyone who has actually tested higher RAM frequencies or RAM quantities above 4 GB. in a typical modern desktop PC knows that frequencies above 1600 MHz. produce no tangible system performance gain in either Intel or AMD powered PCs. Tangible means a change in system performance that you can actually see and or feel. In addition more than 4 GB. of RAM produces minimal gains unless you operate your PC with many applications open or functioning concurrently, which most people don't. CAD, modeling, and some other business applications CAN benefit from more RAM but many consumer apps don't.Be advised that RAM benches grosssly mis-represent the actual SYSTEM gains because they assume the RAM is saturated 100% of the time, which it is not. If you run real applications and compare 1333 MHz. to 2400 MHz. you will likely not even be able to tell the difference between the two because DDR3 RAM @ 1333+ MHZ. is not a system bottleneck.
The RAM mfgs. are doing everything they can to convince enthusiasts to buy high priced high frequency RAM because this is very profitable for them. When you test with real applications and see how tiny the system gains are, you will wish that you had bought 1600 or 8166 MHz. RAM at a reasonable price and used the cash for a faster CPU or GPU.
If you have money burning a hole in your pocket then by all means buy the fastest RAM you can find. Then you can brag at how cool it is... even if you can't actually use the highest frequency in your PC. Understand that your CPU may not be able to run the RAM anywhere near it's rated frequency. Typically Deneb CPUs top out around 1600 MHz., Thubans around 1800 MHz., Zambezi at 2000+ MHz. and Vishera around 2400 MHz. based on initial testing for Vishera. No one cares about Intel CPUs so I won't post on them... <LOL>
Beenthere - Wednesday, October 24, 2012 - link
Can't change the type from 8166 MHz. to the proper 1866 MHz. but most folks should be able to figure it out...silverblue - Thursday, October 25, 2012 - link
Of course, if you have an APU-based system, the faster memory does indeed make a difference... though I agree, it's the exception rather than the norm.JlHADJOE - Thursday, October 25, 2012 - link
But then its totally contrary to one of the main reasons behind having an APU -- penny pinching.These kits cost twice the DDR3-1333 going rate, so that's $75 you could have put into a GPU. Can't speak for everyone, but I'd probably choose an i3 with DDR3-1333 + a 7750 over an A10-5800k with DDR3-2400.
JohnMD1022 - Wednesday, October 24, 2012 - link
My thoughts exactly.1600 seems to be the sweet spot on price and performance.
PseudoKnight - Wednesday, October 24, 2012 - link
Anandtech did a series of memory frequency tests like a year ago (I forget exactly). While they found that 1333 to 1600 didn't offer much in terms of average FPS gains in gaming, it had a clearer impact on minimum frame rates. I'm not saying it's worth it either way here, but I'd like people to give some attention to minimum frame rates when talking about the benefits of bumps in memory frequency.That said, 2400 is obviously overkill here, but that should be obvious to anyone who wants to spend their money efficiently.
Impulses - Thursday, October 25, 2012 - link
The article the did a year ago (with Sandy Bridge in mind) says absolutely nothing about minimum frame rates vs average... I don't even see how faster memory could have such an effect with a dedicated GPU.Impulses - Thursday, October 25, 2012 - link
*theyJlHADJOE - Thursday, October 25, 2012 - link
It might have been techreport. They're the guys who usually do those frame-time measurements.poohbear - Thursday, October 25, 2012 - link
pseudoking what are u talking about? there is virtually NO effect on minimum frames on a dedicated GPU system. Ever since the memory controller moved to the CPU, the RAM timings have become ALOT a less important component in the system. The only way it shows a difference is when you go to all kinds of outlandish scenerios that involve isolating the GPU and CPU to situations that show some difference between RAM, but in a real world setting those situations are so rare that it becomes pointless to even entertain them.Ratman6161 - Thursday, October 25, 2012 - link
But add running virtual machines to your list of reasons why a lot of memory might be good. When working from home I've actually typically got the host machine where I'm doing most of my actual work plus at least two virtual machines running, each VPN'ed into a different remote network. So it isn't too uncommon for me to see about 90% of my 16 gb in use at any one time. And I do occasionally hit times when I have to shut down one VM in order to start another. So I wouldn't actually mind having 32 GB.On the other hand, while I need a large quantity of RAM, my 1600 MHz G-Skill works just fine performance wise so I don't need speed - I need quantity.
mmstick - Saturday, October 27, 2012 - link
Actually, yes you can tell the difference depending on the applications you run. In fact, the distributed OpenCL GPU computing projects I use over at BOINC, the biological research, requires extreme frequencies. The higher the processor frequency, the higher your GPU load and ability to install more GPUs for the project becomes. If I run my RAM at 1333Mhz with my 7950, I need to run 8 work units in order to get that memory to get 80% efficiency. After overclocking my RAM to 1800Mhz I was able to get the efficiency up to 95%. With faster memory, I could run less work units, and install a second graphics cards, although I would likely need 2133Mhz quad channel memory in order to saturate two 7950s in the HCC project. In POEM, I would actually need 4266Mhz quad channel DDR3 in order to saturate a single 7950 more than likely.Another scenario is AMD APUs, where FPS almost scales linearly based on RAM frequency because it uses system RAM as VRAM. GPUs are very memory intense, which is why GPUs have 256-384 bit memory interfaces. With an APU, the GPU on the die is restricted to the lame 64 bit memory interface we have with our system RAM.
RAM kits like these are sold to people who need them, if you don't need them, buy the lower frequency RAM.
mmstick - Saturday, October 27, 2012 - link
"The higher the processor frequency" I meant "The higher the memory frequency"mmstick - Saturday, October 27, 2012 - link
"to get that memory to 80%" should be "to get GPU utilization to 80%"Impulses - Thursday, October 25, 2012 - link
What's with the cost analysis of all these high speed $150 kits? Would you actually shell out that amount of money for a run of the mill gaming/enthusiast system when there's a ton of 1600 4x4GB or even 2x8GB kits on Newegg selling for $75-85?I've wishlisted like half a dozen G.Skill kits within that range AND with blue spreaders specifically... If I width the search there's tons more obviously, just hoping one of those goes on sale during Thanksgiving for like $50-60, even tho I'm in no dire need to upgrade from 8GB.
(and yeah, looking at blue purely for aesthetic reasons obviously!)
mmstick - Saturday, October 27, 2012 - link
I would because it would cut my research rate in half. I run OpenCL GPU projects on my systems, where if I had 2400Mhz memory I could output twice as many work units with my graphics cards per day in Help Conquer Cancer.n0x1ous - Thursday, October 25, 2012 - link
Green PCB? Really? YukBlack or Nothing
bigboxes - Thursday, October 25, 2012 - link
Do you ever look at your ram once you install it?JonnyDough - Friday, October 26, 2012 - link
That's how I feel. I used to care about the color of my computer case, etc but I realized that its more about longevity and ease of installation for me. It's nice to go with a color theme, but the extra cost each time I upgrade just isn't worth it. So now that I've matured a bit I've decided on mostly black boxes and hardware as its plentiful and price competitive.saturn85 - Thursday, October 25, 2012 - link
how about adding a folding on cpu benchmark?valnar - Friday, October 26, 2012 - link
Memory companies will continue to put out insanely spec'ed pieces so they get reviewed by Hardware sites, because it's the only way to get their name out there. Otherwise, memory as a whole is a pretty boring component. Why else do they also need to have interesting names and brightly colored heatsinks? Marketing at its best (or is it worst?)