Logo   Banner   TopRight
TopUnder
Transparent
DDR3 Round-Up: Core i7 Performance Analysis
Transparent
Date: Feb 23, 2009
Section:Misc
Author: Chris Connolly
Transparent
Overview

Since the great DRAM price plummet of 2007, memory prices continue to be extremely low. It's currently possible to pick up high capacity memory modules for very respectable prices. As the DRAM market is highly volatile, however, it's possible that we could see prices jump within the next month, week, or even by the day. However, it has been about a year since prices dropped, and things have yet to "recover", allowing prices to remain low. Regardless, continues to be a fairly good time to buy system memory.

The majority of the really good deals are for DDR2 memory, which has completely flooded the market and is available for ultra low prices (under $30 for 4 GB in some cases). With virtually no money to be made in DDR2 anymore, beyond high-end enthusiast modules, memory manufacturers are pinning their hopes on DDR3 to revive the market. Not that the situation is much better there, as 4 GB of solid DDR3-1333 memory can be had for about $70 on the low-end.

However, there is demand, and they have Intel to thank for it. As Intel has now integrated a triple-channel DDR3 memory controller onto their superb new Core i7 processors, potential buyers are now advised to buy three-or-six DDR3 memory modules per system, compared to two-or-four of previous generation Core 2 based systems. Instead of dual-module kits, triple-module kits are currently the flavor of the moment, which has allowed memory manufacturers to be a little more creative in terms of naming and packaging.

There's quite a lot of variance in DDR3 modules today, which can certainly be overwhelming for potential new buyers. While the Core i7 processor is officially rated to run at only DDR3-1066 speeds, modules that can run at DDR3-2000 or even higher are available for the Core i7, and many new motherboards support speeds even greater than these. There is also the introduction of huge 12 GB (6 x 2 GB modules) capacity kits to consider, and we have latencies in the CAS 7 to CAS 9 range, all of which affect pricing of these various modules kits dramatically.


Triple-Channel DDR3 Memory Kit from Kingston


Triple-Channel DDR3 Memory Kit from OCZ



Triple-Channel DDR3 Memory Kit from Corsair

Today, we're going to look at some of these new kits and see if we can break down what memory-related aspects you should look for when buying an X58 platform. We've got kits from big names like Corsair, Kingston, and OCZ here in the labs. Let's try to clear up some questions, shall we?


Transparent
Clock Speed

DDR3 is built to scale in terms of clock speeds. Initial speed grades of DDR3-800 and DDR3-1066 didn't show significant performance enhancements over DDR2 at the same clock speed. However, DDR2 has more or less been tapped out at a maximum clock speed of 1200 MHz. DDR3, on the other hand, has been able to quickly ramp up in clock speed, and now is available at speeds of 2000 MHz and higher with no end in sight.

We wanted to see how clock speed can affect system performance across the board. We configured a system with 6 GB of DDR3 and ran a set of sample benchmarks with the memory divider set to speeds of 800 MHz, 1066 MHz, 1333 MHz, 1600 MHz, and 1866 MHz, which appear to be the most common options on enthusiast class motherboards. As it's possible to set this divider without affecting any of the other system clocks, we can directly pinpoint how much performance is gained by additional memory clock speed alone. At each clock speed, memory timings were set to 9-9-9-24 for consistency across the board.



As you can see in the graphs above, there is a clear benefit to running higher-clocked memory. Performance increases as the memory clock increases, across the board.

Transparent
Clock Speed (Continued)

We continued our evaluation of DDR3 memory clock speed performance on the Core i7 with a few more applications... 




As you can see, memory clock speed can affect performance across the board, which isn't too surprising as all aspects of computing touch the memory subsystem. The biggest differences we see in our test suite are the raw memory bandwidth and latency numbers, which swing wildly from our lowest to highest clock speed settings.

In terms of actual application performance, we don't see a huge performance difference between the lowest and highest DDR3 clock speeds, although the difference is worth noting. Moving from DDR3-800 to DDR3-1866 bumped up Crysis by nearly 10 FPS, whereas our encoding, rendering, and image processing tests all were able to be completed a few seconds faster. Gaming has been one of the most memory intensive applications, so it's not surprising that it sees the biggest jump, and why enthusiast-class memory modules are targeted at gamers. If you game, faster memory does indeed help performance. If you're a workstation user, it's probably best to look at memory capacity rather than speed when looking to upgrade.

Transparent
Latency

DDR3 modules typically run at higher latencies compared to DDR2 modules at the same clock speed. Typical DDR3 modules run in the CAS 7-7-7 to CAS 8-8-8 range, whereas some modules can run as high as CAS 9-9-9. On the other end of the spectrum, it's possible to get memory modules which can run as low as 6-6-6 timings, which are likely some of the fastest you'll see on store shelves.



We wanted to see how latency affects memory performance on the Core i7. Here, we took 6 GB of DDR3-1333 memory and ran through our test suite again, with memory latency levels of 6-6-6-18, 7-7-7-20, 8-8-8-22, and 9-9-9-24. No other timing attribute was changed.  Results continue on the next page...

Transparent
Latency (Continued)



While it doesn't make as large of a difference as clock speed, latencies do indeed affect overall system performance and should be kept in mind when buying memory modules. However, if your memory modules can run at latencies under CAS 9, you should be mostly in the clear. Once you get under CAS-8, the performance gained is fairly small.

Transparent
Channels / Bandwidth

Just as the move from single channel DDR memory to dual channel DDR memory allowed for much improved memory throughput and performance, the move to triple-channel DDR memory with the Core i7 promises similar benefits. While it's unlikely that most will use a Core i7 without a triple-channel DDR3 subsystem (as most OEMs are building within these guidelines, and triple channel memory kits are relatively affordable), we were curious to know how much this new feature actually helps performance.

We put together some configurations using single channel (1x module), dual channel (2x modules) and triple channel (3x modules) to see how the additional memory bandwidth of a triple channel configuration helps synthetic and real-world benchmarks.



In our first few tests, with the exception of the SANDRA latency benchmark, moving from one, two, or three memory channels has a significant impact on performance with the Core i7.

Transparent
Channels / Bandwidth (Continued)



Using three or six memory modules to allow for a triple-channel DDR3 memory configuration is surely the best route to take with a new Core i7 system.  Our synthetic memory bandwidth numbers showed huge differences between the configurations, but in terms of real-world performance, the differences were small. Triple channel memory is not the big time real-world performance bullet which many claim it to be, although it does help deliver much more raw memory bandwidth, which in specific situations, can help greatly. For most applications, though, you won't be touching this level of memory bandwidth, so it's not worth stressing about it too much.

Transparent
Corsair Dominator and OCZ Platinum

Corsair Dominator XMS3 1866 MHz
Corsair’s Dominator XMS3 series are commonly known as some of the best designed, but most expensive modules on the market. Dominator modules typically run at higher clock speeds and lower-latencies compared to most other enthusiast class modules, but what really kicks up the price tag are their custom PCB designs and sleek, effective heatsink design.


The Dominator kit we have to look at today is one Corsair's latest.  It offers 6 GB of memory running at 1866 MHz clock speeds. It’s a triple-channel kit which is comprised of 3 x 2 GB memory modules. The modules run at fairly high 9-9-9-24 latencies, but do conform to Intel’s guidelines of a maximum voltage level of 1.65V.

Sold separately is Corsair’s Dominator Airflow kit, which is comprised of two low-noise 40mm fans which sit above the memory modules. It’s a fairly unique and well designed unit, as it’s easy to install, easy to remove, and will cool not only up to 6 memory modules, but will provide airflow to other components in the surrounding area. Of course, if you plan to overclock, the additional airflow will do nothing but help. In testing, we were able to get our Corsair Dominator 1866 MHz modules up to 2100 MHz with 1.9V memory levels, but we were not able to take it higher. Since the time of the 1866 MHz kit release, Corsair has gone on to certify kits at 2000 MHz and above, which means that the prices of this particular kit should be dropping fairly fast. If you can afford it, they are certainly some of the best memory modules on the market – there is a reason why Dell is equipping their high-end XPS gaming systems with Dominator modules.


OCZ Platinum XTC 1600 MHz

OCZ currently offers their Platinum lineup at speeds up to 2000 MHz, whereas the kit we're looking at today is their PC3-12800, which runs at 1600 MHz. While in the lower-end of the clock speed spectrum compared to the other kits we're looking at, this kit does boast the lowest stock latencies, running at CAS 7-7-7 while running at an Intel approved 1.65V core voltage.



The modules are low-profile, and the heat-spreaders which OCZ equips these modules with are thin, but appear to be quite effective. OCZ's Platinum Z3 XTC heatspreader uses an open-air mesh design which is based on aluminum alloy, which allows for airflow directly on the chips and PCB and for direct heat conduction. These 1600 MHz stock speed modules were able to clock up to 2050 MHz in our testbed at 1.9V, slightly below the Corsair 1866 MHz modules, but mighty impressive considering these modules are roughly half the price of the Corsair memory modules.

Transparent
Kingston HyperX 1600 / 1800

Kingston HyperX DDR3 1600 MHz / Kingston HyperX DDR3 1800 MHz
Kingston’s HyperX modules have never had the enthusiast appeal as Corsair’s XMS series, but they have been lock-step in line with Corsair for every major evolution of the high-end enthusiast memory market. The latest generation HyperX DDR3 modules sport an updated heatsink design which we feel is one of the most attractive on the market, matching a dodger blue aluminum alloy heat-spreader with a brushed aluminum piece on top. The modules are also significantly shorter compared to Corsair’s Dominator, which is a good thing, assuming the heatsinks are powerful enough to handle the heat created by these modules.




We were able to test two of Kingston’s new HyperX DDR3 series. Our HyperX DDR3 1600 MHz modules (KHX12800D3LLK3) ran with 8-8-8-24 latencies by default, whereas the second kit (KHX14400D3K3) runs at 1800 MHz with 9-9-9-24 latencies. Both modules have the exact same external design and are impossible to tell apart on appearances alone. Interestingly enough, both modules were able to overclock to roughly the same level, peaking out at 2050 MHz at 1.9V voltage levels. It looks to us that both these memory kits are using a vast amount of identical components between them, and might even be using the same memory chips underneath, just re-purposed for different markets.

They cost about the same, too, so it’s really up to you which you would prefer. Higher clock speeds with higher stock latencies or slightly lower clock speed with slightly lower latencies. From a performance perspective, as you’ll see later, the 1800 MHz with higher latencies typically performs a tick better, so this would get our nod. All in all, really nice modules from Kingston, and a nice evolution of the HyperX line. Kingston tends to play it a bit more conservative than Corsair or OCZ, but this is not necessarily a bad thing. In our time testing these modules, the number of incompatibilities with motherboard with Kingston modules has been close to nil, which we can’t say for some of Kingston’s competitors.

Transparent
Testbed and Synthetics
Test System Details
Specifications and Revisions

  • Intel Core i7 920 (2.66 GHz Quad-Core, 45nm)
  • Asus P6T Intel X58 Motherboard
  • Nvidia GeForce 9600 512 MB
  • Western Digital VelociRaptor 150 GB SATA-II 10K RPM Hard Drive
  • Plextor PX-755SA DVD+/-RW Drive
  • Corsair HX620W 620W Power Supply
  • Microsoft Windows Vista Ultimate Edition (64-bit, SP1)

  • Patriot Baseline Set (1066 MHz, CAS 9-9-9-24)
  • Corsair XMS3 Dominator (1866 MHz, CAS 9-9-9-24)
  • OCZ Platinum XT (1600 MHz, CAS 7-7-7-24)
  • Kingston HyperX 1600 (1600 MHz, CAS 8-8-8-24)
  • Kingston HyperX 1800 (1800 MHZ, CAS 9-9-9-27)
Synthetic Memory Benchmarks
SiSoft Sandra 2008 SP1
 


With their stock settings enabled, the OCZ Platinum and Corsair Dominator modules give the best performance out of box. Both deliver about 26.5 GB/s of memory bandwidth with sub-60ns latencies. Our Kingston HyperX modules delivered about 24 GB/s with 60ns+ latencies. The OCZ modules tested surprisingly well here, considering they are running at a significantly lower clock speed compared to the Corsair. However, our Corsair modules have fairly high latencies by default, whereas the OCZ modules are a tight 7-7-7 which helps make up for the clock speed deficiencies.

Transparent
Crysis and Windows Media Encoder
Crysis - CPU Benchmark
1024 x 768 Resolution


Microsoft Windows Media Encoder 9 x64
Timed Video Encoding Benchmark


While we saw heavy variance in our synthetic tests, our first round of real-world test shows small performance differences between our kits. Our memory intensive Crysis test shows less than 1 FPS difference between all of the kits, despite all the clock speed and timing differences. Of course, in comparison to Intel's default Core i7 spec of DDR3-1066, all of these high-end kits do perform substantially better.

We see absolutely zero difference in our Windows Media Encoding test, as all of our sets deliver the same result. Not surprising, given this is a CPU intensive benchmark. This does highlight an important point, in that if you are bottlenecked by another system component, faster memory just won't help.

Transparent
Photoshop CS3 and Cinebench
Adobe Photoshop CS3
Batch Converstion of RAW Photos to JPG


Cinebench R10 X64 Edition
Multi-Threaded CPU Rendering Benchmark


Memory-intensive Photoshop CS3 shows the best performance with the OCZ Platinum modules, besting out the other high-end kits, even the Corsair Dominator modules at a higher clock speed. Interestingly enough, the OCZ kit falls a smidge behind in our Cinebench test, whereas kits from Kingston and Corsair have the performance edge.

Transparent
Our Conclusion

There are some blatantly obvious things which we’ve proven with this round of analysis today. Higher clock speed memory modules perform better than lower clocked modules with a Core i7, and likewise, lower-latency modules perform better than higher-latency modules. Triple-channel memory configurations will give you more memory bandwidth compared to single-channel or dual-channel configurations. But anyone who knows a smidge about memory knows that these are basic rules that typically remain true with any platform. Our question is, how much performance will you gain by going out of the way to purchase enthusiast-class high-clock / low-latency modules?



Our tests show that Intel’s Core i7 platform, when equipped with triple-channel memory of any sort, has boatloads of bandwidth to utilize. While the platform is only officially rated to handle 1066 MHz DDR3 memory modules, we found in testing that the numbers really start to come alive once you hit the 1333 MHz to 1600 MHz range. Lower than this, and you’ll be taking a slight performance hit. Above 1600 MHz and you start to see performance gains taper off.

In terms of latency, obviously lower is better. If your modules are in the CAS 8-8-8 to 7-7-7 range, you’ll be getting nearly all of the available performance from a Core i7 system. If you manage to go down to CAS 6-6-6, you’ll see a very slight boost, but not enough to really push for. We wouldn’t necessarily recommend modules in the CAS-9 range, as they tend to not be as effective per clock in comparison to lower-latency modules. Case in point, our lower-clocked CAS 7 OCZ modules typically performed on par with our CAS-9 Corsair modules at a 266 MHz higher clock speed.

Interestingly enough, while the Core i7 platform has tons of bandwidth when utilizing its triple channel memory interface, when we broke this by using 1/2/4/5 module combinations, we didn’t see a large drop in performance. Sure, our synthetic memory bandwidth benchmarks went down significantly, but real-world performance was more or less un-affected. For the best Core i7 experience, obviously we would recommend going for a triple channel kit (as they are readily available now, and priced pretty competitively) but if you have DDR3 modules from a previous system, there really is no harm in equipping a Core i7 with modules in a dual-channel configuration, so long as they'll run reliably at the lower voltages of a standard Core i7 memory interface.

Of the kits we’ve tested today, given the price, default clock speeds/latencies, heatsink designs, and overclockability, our modules of choice would likely be the OCZ Platinum XT. They are priced at roughly half that of the Corsair Dominator modules, but delivered almost identical performance and only a smidge less overclockability. In terms of price/performance, these are the ones to grab. Corsair's modules were impressive indeed, but we feel need major price drops to be competitive. Of course, if money is no option, Corsair is likely your best bet. Kingston's HyperX modules both tested out quite well, but were middle of the road in terms of performance. Price wise, they are very competitive, however, and we have no major qualms against them. If it was our money, though, in these times, the more cost-effective OCZ modules simply make the most sense.

 
Corsair Dominator XMS3 1866 MHz

  
  • Custom Heatsink Design
  • Dominator Cooling Fan
  • Excellent Stock Performance

 

  • High Latency
  • Excessive Cost

 


OCZ Platinum XTC 1600 MHz

  
  • Low Latency Modules
  • Excellent Stock Performance
  • Cost Effective

 

  • None


 

Kingston HyperX DDR3 1600 MHz
Kingston HyperX DDR3 1800 MHz


  
  • Simple, Effecicient Heatsink Design
  • Good Price/Performance

 

  • High Stock Latencies
  • Middle of the Road Performance

 




Content Property of HotHardware.com