Overclock.net › Forums › Components › Memory › Help me through my bottleneck! ( 6-7-6-24-1T @ 1333mhz)
New Posts  All Forums:Forum Nav:

Help me through my bottleneck! ( 6-7-6-24-1T @ 1333mhz) - Page 3  

post #21 of 27
Quote:
Originally Posted by xd_1771 View Post

Well, one of the problems here is that you're linking to Intel platform reviews in a thread that happens to do with AMD platform memory performance. A lot of things are really different between the two, you know; i.e. bandwidth-scaling, and response to higher or lower timings. So it's not really representative of this situation.

If improving memory bandwidth really didn't make a tangible performance difference in real-world apps, then no one on here would be doing the so popular thing of OCing their CPU-NB/memory controller clock, which has been proven to make significant performance improvements in several apps.

The one AMD platform review that is linked to this thread is showing that improving memory bandwidth specifically by overclocking the CPU-NB does create noticeably large gains in performance.


It doesn't help to be discouraging on a site that is labelled to exist for the "pursuit of performance"....



If you actully read the links I provided they included links to testing of Phenom II systems and the AMD powered systems showed the exact same results as the Intel powered systems - NO tangible performance improvement in real applications. Benches show some gains but not real word apps.

The discussion in this thread is about quantifying the performance changes in real world applications vs. benchmarks for DDR3 RAM with frequencies of 1333 MHz. and higher.

No one is being discouraged at all. People are free to do whatever makes them happy. I don't tell people what to do, I just provide accurate information. The links substantiate that there is no tangible, i.e. "significant" performance gain in real apps. This is puposely stated in two of the three reviews.

If you can't see or feel a change in application performance by changing the latencies or frequency, then the point is moot.
Edited by AMD4ME - 2/10/12 at 8:16pm
post #22 of 27
Here, a 10% gain in an actual game that I play from going from DDR3-1200 to DDR3-1800 on my FX-4100 (all I did was lower my RAM divider two notches). Test is with crossfired HD 6850s at 1000MHZ core and 1200MHZ VRAM and CPU at 4500MHZ with maximum detail settings in the "action scene" benchmark built into the game.

315

309
Light Gamer #1
(12 items)
 
Light Gamer #2
(10 items)
 
 
CPUMotherboardGraphicsGraphics
i3-6100 ASRock Fatal1ty Z170 Gaming K4/D3 MSI r7850 TFIII OC MSI r7850 TFIII OC 
RAMHard DriveCoolingOS
G,Skill TridentX OCZ Vector 150 TPC-612 Windows 10 
KeyboardPowerCaseMouse
G105 Rosewill Capstone 550M HAF XB M570 
CPUMotherboardGraphicsRAM
Athlon 860K MSI A88X-G45 MSI GTX 460 G.Skill RipjawsZ 
Hard DriveCoolingOSKeyboard
PNY Optima Hyper T4 Windows 10 Devastator 2 
PowerCase
Rosewill Valens DIYPC Ranger-R4-R 
  hide details  
Light Gamer #1
(12 items)
 
Light Gamer #2
(10 items)
 
 
CPUMotherboardGraphicsGraphics
i3-6100 ASRock Fatal1ty Z170 Gaming K4/D3 MSI r7850 TFIII OC MSI r7850 TFIII OC 
RAMHard DriveCoolingOS
G,Skill TridentX OCZ Vector 150 TPC-612 Windows 10 
KeyboardPowerCaseMouse
G105 Rosewill Capstone 550M HAF XB M570 
CPUMotherboardGraphicsRAM
Athlon 860K MSI A88X-G45 MSI GTX 460 G.Skill RipjawsZ 
Hard DriveCoolingOSKeyboard
PNY Optima Hyper T4 Windows 10 Devastator 2 
PowerCase
Rosewill Valens DIYPC Ranger-R4-R 
  hide details  
post #23 of 27

Where are your benchmarks?


If I do so remember correctly you came into this thread with the specific promise that you would provide us with your benchmarks.  Instead, you provided us with benchmarks from three sites, and that could be perfectly reasonable - but 1. you failed to provide us with your so-promised benchmarks, and 2.n one of those reviews have to do with the AM3 platform - and only one links to/references a review that has anything to do with the AM3 platform (and I'm curious why you didn't link directly to that in support of providing the OP with some actually useful information).  That review uses a C2 revision processor with a relatively undeveloped IMC with memory speeds no higher than 1600Mhz.  No significant processor overclocks or CPU-NB overclocks were taken into account with those tests, where the performance differences associated with bandwidth increase from tweaking RAM can be amplified by the increase in throughput for both the CPU and the IMC.  Most AMD K10 based processors now cell with a C3 revision IMC which allows for higher memory speeds.

 

Some of the benchmarks are also ridiculous; I'm a video editor and I can tell you straight up that video editing is a process and is not something you can measure by how much time it takes to run a specific list of tasks.  Where I may be editing several large and small high definition video files that have to constantly be accessed, moved around, and copied in memory, I find that the extra bandwidth and lowered latency I get from my own higher-performance memory setup (which was purchased during a time when performance memory came at a lot less of a premium) is really helpful.

 

Now, where have you been? damric provided his benchmarks as promised, that show a helpful gain in minimal as well as average framerates. Again, if I do so remember correctly you came into this thread with the specific promise that you would provide us with your benchmarks..... /etc/

 

I find it also to be very hypocritical to be calling us out on using opinions in this argument when you are using several opinions of your own.  There is obviously a difference between what you consider to be "tangible" and what we consider to be "tangible", as there is no general definition of "tangible".  I know some people who revel at the idea of a 3 FPS increase in a game's minimum FPS.  There is also a difference between what you might consider a "real world" app and what we do as "real world" tasks with "real world" apps.  There's not really any way to tell, and I wouldn't be making that generalization and would instead be asking the OP those questions.

 

This doesn't seem to be about showing us the proper examples so much as I thought, nor helping out the OP who obviously seems to have indeed seen a performance increase through the increase in bandwidth provided by further timings and CPU-NB tweaks.  You'll just have to take his word for it; i.e. what I said above concerning video editing: it is a process that you just cannot properly measure with benchmarks of any sort, including with real-world apps.  A problem with benchmarks and apps is that they only make up a partial replication of the computing experience, and while they can provide more accurate results they are really limited to being just bases to follow.

 

By the way, if I haven't already pointed this out, no one seems to be have disagreed with you.  You came in with a non-misguided statement concerning system performance to memory bandwidth and we respect that, and if you look around none of us had disagreed with you concerning the focus is on "sytstem" (do you mean, non-specific apps?) performance.  I found my own recommendations to performance memory (beyond the regularly found performance standards today such as DDR3-1600 CL9) have been disappearing as the gap in pricing between these generic and mainstream-performance kits and high-end performance kits (i.e. beyond 1600) widened.  damric had done nothing wrong by providing his reasonable opinion on a related but not specific-to-yours matter and you had no reason to argue with him and contribute to taking a thread in a direction that leads to the creation of content that doesn't help out the user questioning.  I support initiatives and choices that help out the user questioning, and so should you.


Edited by xd_1771 - 2/11/12 at 2:38am
post #24 of 27
^^^^ I previously explained why I wasn't posting my benches because certain people would just find fault with them. Instead I posted links to THREE reputable PC industry websites who had also conducted extensive DDR3 RAM testing who reached the exact same conclusion I and many other's have from testing real world applications. These links also included Phenom II DDR3 RAM testing which showed exactly the same results - NO significant increase in real world apps. Did you even bother to read the references?

Actually Damric has made all sorts of disparaging accusations based on his beliefs. Making it personal is not going to change reality. If you want to argue with the industry sources who conducted the extensive benches and concluded there in no significant change in real world apps, that's up to you. I provided the PROOF that was demanded and now some folks have an attitude over it because it differs with their opinion. My comments in this thread are specifically about quantifying tangible gains in real world applications with DDR3 latencies and frequencies above the Phenom II default 1333 MHz. that the OP has

What you chose to believe is entirely up to you.

BTW - xd_1771, in other memory threads you have also given people poor advice on mixing DDR3 RAM that can cost people more money than if they had bought one, proper matched RAM kit the first time instead of having to buy it after mixed RAM failed to work for them... You should be clear up front that mixing of DDR3 RAM often does not work and that just because you are willing to take the financial risk of buying and mixing, that may not be the best option for other folks.
Edited by AMD4ME - 2/13/12 at 7:00am
post #25 of 27

If you believed your benchmarks would have given no merit then why did you suggest them in the first place?  It seems you like to be in this for the joy of proving people wrong (or appearing to prove people wrong).  What's there to prove wrong?  damric made a reasonable, unarguable case: every 1-2 FPS adds up when you tune each component and run the most demanding application settings.  This is Overclock.net - you can expect that a lot of us users here actually run reasonably demanding applications, thus we overclock to get the best performance while running these demanding applications.

 

Once more, only one of those reviews links to/references a review that has anything to do with the AM3 platform (and I'm curious why you didn't link directly to that in support of providing the OP with some actually useful information).  That review uses a C2 revision processor with a relatively undeveloped IMC with memory speeds no higher than 1600Mhz.  No significant processor overclocks or CPU-NB overclocks were taken into account with those tests, where the performance differences associated with bandwidth increase from tweaking RAM can be amplified further by the increase in data throughput to both the CPU and the integrated memory controller (IMC).  Also, most AMD K10 based processors now sell with a C3 revision IMC which allows for even higher memory speeds thus creating even higher, more tangible (or not, depending on user) benefits.

 

The financial loss attributed by taking a small step forward in ensuring memory compatibility is something I have found to be rather useless.  I have been an active contributor on this area of the forum for more than 2 years and I have seen few, if any, situations where the mixing of memory kits that do not pose an obvious difference in the used ICs (i.e. different speed/timings/voltage ratings) have caused any problems.  Why you're even pointing this out when it again has no relation to what the OP wants astounds me - I see this as a personal attack (also, whatever happened to the mentality you promote that you "making it personal does not change the reality?").  If you're interested in reality, I have found that the majority of the times when my advice concerning memory-mixing has been taken for (sometimes significant) cost-saving purposes there have been no problems involved.  Thus the financial risk of mixing is very, very low.  If mixing RAM in any way is supposed to be completely unacceptable (to you), then why are we seeing several computers from prebuilt manufacturers and several custom-computer stores/builders like NCIX being sold with different DIMM sizes, that may even be using different ICs?  RAM is RAM - it's not as inflexible as you seem to think it is.  RAM is not like trying to pair a specific CPU with a specific socket.  Think of why AMD allows CrossfireX between cards that service totally different tiers or have other slight differences (i.e. 5750 to 5770) - they allow it because it can be done.  If different RAM kits were as compatible as you so claim them to be then the manufacturers would not be giving us the opportunities to mix and we would be seeing tons of different DIMM sizes available on the market right now: 768MB DIMMs, 2.5GB DIMMs, 3GB DIMMs.... and yet we don't.  Why: it's simply more efficient, less confusing, less complicated (in terms of DIMM design) and less costly for IC manufacturers when DIMMs are sold in these size denominations that offer constantly-scaling size-multiplies, using (the modern standard on most DIMMs is) 8 ICs per side on one or two sides, and allow builders to mix DIMMs to the extent where they can get the amount they want or need.

 

Nevertheless, I don't set a low bar for this either; I personally draw the line here: when mixing memory kits, one must ideally for best compatibility and running use kits that meet ALL of the three following specifications: same rated speed, timings and voltage (assuming intended operation is at that level and not at any settings that may offer increased performance) - even if they do not meet these specifications: brand, ICs, heatspreaders.  This is a bar that I consider to be pretty high when the issue of mixing memory kits comes across.  Of course, advice is advice - and that is something that I like to differentiate with the provision of facts when personal experiences/observations come into play.  Who wouldn't unless there is an absolute need to?  I'm not working for any RAM-related or computer-related company - I don't need to pick a side when it comes to advising people here as to what to do.

post #26 of 27
^^^ Too many words... A 10,000 word responses don't change anything. rolleyes.gif

It's obvious from Damric's inappropriate and antagonistic comments, when people disagree with him, it becomes personal. Posting my test results would have added 50 more inappropriate comments to the thread and changed nothing.

I highly doubt that the THREE PC industry sources provided, are conspiring to dupe all enthusiasts since they are PC enthusiast websites. If anything they'd be inclined to promote OC'ing the RAM if it showed any tangible benefit in real world apps. vs. theoretical gains in benches above 1333 MHz. or with tighter latencies.

Real world apps show no tangible gains. One review specifically states "No significant gain.. in real world apps". The other reviews states: "insignificant gains in real world apps..."

As I stated, people are free to believe whatever makes them happy.

Enjoy.
Edited by AMD4ME - 2/15/12 at 7:10pm
post #27 of 27
I've cleaned this thread before and that's it. We're done here.

The pointless bickering is appealing guys, please remain civil and read the posts you're discussing.

Thanks.
Bye Bye 775
(13 items)
 
  
CPUMotherboardGraphicsRAM
i5 2500k @ 4.5GHz ASUS P8P67 PRO MSI GTX 660 Ti Power Edition  2x4gb Ripjaws@1866MHz 
Hard DriveOptical DriveOSMonitor
840Pro 128 | 3xWD Vraptor | 1.5TB Green LG 6X Blu-ray /HD-DVD combo Win7 Ultimate 64bit 2 x U2311H 1920x1080 
KeyboardPowerCaseMouse
Logitech G710+ Corsair HX620 Corsair 600t Logitech G500 
  hide details  
Bye Bye 775
(13 items)
 
  
CPUMotherboardGraphicsRAM
i5 2500k @ 4.5GHz ASUS P8P67 PRO MSI GTX 660 Ti Power Edition  2x4gb Ripjaws@1866MHz 
Hard DriveOptical DriveOSMonitor
840Pro 128 | 3xWD Vraptor | 1.5TB Green LG 6X Blu-ray /HD-DVD combo Win7 Ultimate 64bit 2 x U2311H 1920x1080 
KeyboardPowerCaseMouse
Logitech G710+ Corsair HX620 Corsair 600t Logitech G500 
  hide details  
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Memory
This thread is locked  
Overclock.net › Forums › Components › Memory › Help me through my bottleneck! ( 6-7-6-24-1T @ 1333mhz)