by Ilya Gavrichenkov
11/01/2010 | 10:35 AM
Choosing components for a home or office computer is a multifaceted problem. Someone wants to build as high-performance a platform as possible while another mostly cares about the noise factor. And others yet just want to have a cheap computer capable of handling their everyday tasks. We, at X-bit labs, are trying to test as much hardware as we can to give you well-grounded recommendations but sometimes we come across sophisticated issues calling for very unusual explorations. Today we are going to cover one such issue which is not generally touched upon in hardware reviews (at least we couldn't find much information about that topic on the Web).
Imagine a computer with a fast processor, perhaps even a flagship CPU model, and a top-performance graphics subsystem that easily copes with any load. Suppose the owner of such a dream PC has got some money to spend and wants to keep on upgrading his configuration. What should he pay attention to in the first place?
As a matter of fact, there are not so many options left unless he improves his mainboard, CPU or graphics subsystem. He can buy some peripherals, gadgets or accessories, for example. Or he can get somewhat deeper into the internals of his PC and try upgrading its disk subsystem or system memory. It is on these two opportunities that we want to dwell in this review.
We often upgrade our disk subsystem or system memory because we have to rather than because we intentionally want to boost its performance. We replace our hard disk when it becomes too small to store our data. We add more system memory when we find it irritating that the OS is constantly accessing its swap file. It is indeed but seldom that some user deliberately tries to upgrade these subsystems to boost his computer's performance. This is usually the case when the computer is used for a specific application that depends heavily on a specific subsystem.
For example, it is perfectly clear that the performance of a file server or any computer that has to process large amounts of data stored on its hard disk may be easily improved by replacing its single hard disk with a faster one or with a RAID array. Similarly, a computer running a dozen of heavy applications or virtual machines simultaneously must have a lot of system memory and may do better if equipped with more of it. Such scenarios are clear enough, though, and users of such computers won’t ask anybody what they need to upgrade because they already know everything without our advice.
But what if we have just a regular all-purpose computer that copes well enough with any task it is given and doesn’t betray any weak spots? What should you improve in the first place, its disk subsystem or memory? What will produce the biggest effect?
Installing a faster disk seems to be a very reasonable option, especially as you can now buy an affordable solid state drive delivering much higher performance than a traditional HDD. Our website and other resources offer lots of test data about HDDs but mostly from synthetic benchmarks. The most popular real-life test is to measure the time it takes to boot up an OS or some heavy applications. However, loading an OS or an application are but one-time events which are additionally accelerated by look-ahead data caching in modern OSes (e.g. Windows has SuperFetch technology for that purpose). A speedy SSD will surely help boot your OS up faster and will make the system more responsive overall, reducing the time you have to wait for a program to start or a file to open. But will it help improve performance in regular everyday applications?
We will try to answer this question by checking out a few work scenarios on our testbed using different storage devices. We will try to show you how the storage device affects the results of benchmarks we use to evaluate the overall performance of a computer, rather than that of its disk subsystem only.
Similarly, we will check out the influence of the amount of system memory on the overall computer performance. Thanks to SuperFetch, extra memory may help reduce the number of hard disk accesses by means of caching. So, it may turn out that installing more memory will have the same effect as replacing an HDD with an SSD!
We took three completely different hard disk drives for our test:
You can refer to the HDD section of our site for details about these products. For now, we will just remind you their speed level using a simple synthetic benchmark called CrystalDiskMark 3.0.
As you can see, these devices vary dramatically in their performance even though all of them are quite modern products. The WD Green series is designed to be energy-efficient and offer large storage capacities while the WD VelociRaptor has the highest rotation speed among all hard disk drives with SATA interface. Intel's SSD is just a very good and fast drive based on NAND flash memory. The SSD is expectedly ahead of its competitors in every application save for sequential writing of data. When processing random-address data blocks, it is many times as fast as the traditional HDDs. The HDDs vary between themselves, too, as they differ in such parameters as spindle rotation speed, recoding density and platter size.
As for the amount of system memory, typical desktop computers have as much as 4 gigabytes of it today. A 64-bit OS is needed to access that amount, but that's not a problem anymore. The most popular Microsoft Windows 7 OS feels the most comfortable, especially if accompanied with many applications, on 4 GB of system memory. By the way, this is why we have been equipping our testbeds with two 2GB modules for the last couple of years.
Meanwhile, with the 64-bit versions of Windows 7 being so popular and with inexpensive but high-capacity DDR3 SDRAM modules being widely available on the market, many users are quite ready to expand their system memory up to 8 gigabytes. Windows 7 is not an obstacle as each of its 64-bit versions will easily support that.
So, there are no technical problems. You just need a 64-bit OS and appropriate memory modules. Having more memory, your computer will access the HDD less frequently, improving the overall performance. In other words, increasing the amount of system memory may be considered as an indirect way of optimizing your disk subsystem. That’s why we are going to cover this issue in our review which is otherwise concerned with disk storage.
Today, every major maker of memory modules for enthusiasts offers 2x4GB memory kits. It doesn’t take much trouble to develop such modules since 2-gigabit DDR3 SDRAM chips have long been available. For our test we will use DDR3-1600 memory provided to us by Mushkin (the part number 996808).
This is a pair of dual-sided 4GB DDR3-1600 modules rated for 7-9-8-24 timings at a voltage of 1.65 volts. The memory kit supports XMP and contains two profiles: DDR3-1600 by default and DDR3-1333 with aggressive timings of 6-8-7-21.
The modules are equipped with the exclusive Ridgeback cooling system that consists of aluminum plates contacting with the memory chips and of a small heatsink above them. The whole arrangement is exactly 40 millimeters tall, which is considerably taller than ordinary memory modules.
Overclockers are often skeptical about increased-capacity modules but the overclockability of the Mushkin 996808 kit is no different from that of ordinary single-sided 2GB modules and installing the double amount of overclocker-friendly system memory looks quite a viable way to upgrade a top-performance computer.
Well, we need to check this out in practical tests. Let’s move on to them right now.
For our today’s test session we used the following testbed:
Complex benchmarks based on real-life applications are exactly what we need for our today’s test session. For example, SYSmark 2007 emulates a user solving practical tasks in a realistic environment: there are a few open applications simultaneously for the user to switch between. Data is transferred between the applications using the clipboard.
As we can see in the diagram, the influence of the disk subsystem and system memory on the overall performance should not be underestimated. The results vary by 25%, so a computer with a faster disk and a larger amount of memory is going to be more responsive at everyday work. This performance improvement is achieved by reducing the latency of disk operations in the first place as is clear from the individual SYSmark 2007 tests.
It is the Productivity scenario that shows the biggest dependence of the computer’s overall performance on its disk subsystem. This might have been expected as this scenario involves working with Microsoft Office documents and email. There is no wonder the fast SSD provides a 60% increase in performance. The disk subsystem is also influential when it comes to working with digital content in the Video Creation scenario, mostly due to Adobe Photoshop which makes a lot of disk accesses when processing large images. In the 3D scenario, which focuses on the final rendering in 3D modeling suites, the disk subsystem doesn’t play a big part while the CPU’s computing capabilities are far more important then.
The results of the configurations with different amounts of system memory are somewhat disappointing. Upgrading your disk affects the performance in SYSmark 2007 greatly but the larger amount of system memory has but a very small positive effect which we can only observe in one scenario, Video Creation (obviously, due to Adobe Photoshop again).
Most unexpectedly, increasing your system memory amount to 8 gigabytes may even lower your computer's performance as we can see in the tests of the SSD-based configuration. The performance hit is small but observable across a number of tests despite Windows 7 turning off its SuperFetch technology for SSDs.
The Intel High Definition Experience & Performance Ratings Test is another complex benchmark but, unlike SYSmark 2007, it focuses only on digital content creation, processing and playback. For this purpose it uses a set of applications including CyberLink PowerDVD 9, CyberLink PowerDirector 8, DivX Pro Codec 6.8.5, HDRsoft Photomatix 3, UnifiedColor HDR Photostudio v2.13.32, Adobe Flash Player v10.1, Apple iTunes 9.0.3, Adobe Photoshop Elements 8.0 and Windows Media Player 11. There are three scenarios in this benchmark: encoding HD video from a digital camera, processing digital photos, and preparing audio and video files for portable players.
The disk subsystem affects the results but not as much as in the previous benchmark. Intel HDxPRT 2010 operates with high-resolution media files and their processing is limited by the CPU’s computing resources rather than by the storage device. Increasing the amount of system memory doesn’t produce any effect here. Clearly, typical processing of audio and video files doesn’t call for more than the standard 4 gigabytes. Photographs captured with modern digital cameras are well below that size, too.
If we take a closer look at the details of the tests, we can see that processing multimedia data is not the type of task that calls for a faster disk subsystem or larger amount of system RAM.
The main thing is to avoid too slow storage devices. The difference between the WD Caviar Green and the SSD in the video editing and encoding scenario amounts to 10%, which is quite noticeable. This difference cannot be observed in the other scenarios, though.
The positive effect of a fast disk subsystem is absolutely obvious if we measure the speed of loading applications and data from files. This is quite clear without any additional tests if you just take a look at synthetic benchmarks’ results. Therefore in this review we won’t do such primitive measurements. Instead, we will focus on investigating the effect of a fast disk subsystem on general computer performance.
That said, we are still going to start out with a standard test of measuring the boot-up time of Microsoft Windows 7 Ultimate x64. The diagram below shows the time it takes from pressing the Power button until the computer begins to react to user’s actions.
The difference between the slowest HDD and the SSD amounts to 30% whereas the amount of system memory doesn’t affect the OS boot-up time much.
It is also clear that the choice of a disk storage device can affect the performance of a task in which data is taken from the disk subsystem rather than from system memory. Archiving is a good example of this kind of task. As we know, the speed of data archiving in WinRAR depends on the speed of CPU and memory. But it turns out that the disk subsystem is an important factor as well.
Interestingly, we can observe an effect from the larger amount of system memory here. It shows up when the disk subsystem is not sufficiently fast. So, it looks like adding more gigabytes of memory to improve your disk subsystem performance only makes sense when you’ve got rather a slow HDD.
Encoding audio files into the MP3 format is similar to data archiving but many encoding tools can run in one or two execution threads only. In this case, even a very slow HDD is enough to deliver data at a required rate. But for our test we took the dBpoweramp converter which can effectively use all of CPU cores it can find, encoding multiple files in parallel: one file on one CPU core. The speed of the disk subsystem becomes quite important then.
Multithreaded writing of data is not a strong point of the WD Caviar Green and it fails in this test. The WD VelociRaptor and the Intel SSD are much faster but differ by 15% between each other. The larger amount of system memory has a positive effect in this test but we can’t say that 8 gigabytes of RAM can make a worthy alternative to a fast disk.
Now we’ve got another popular task that involves processing of data stored on the disk. It is video encoding. We took the x.264-based HandBrake tool to encode a 1080p MPEG-2 video clip into H.264 format.
Well, we can’t see any improvements from the faster disks or larger memory amount here. The Core i7-870 processor proves to be not fast enough to encode video files at the same rate as they are being read even from the slowest disk.
We have the same picture not only at ordinary transcoding but also when applying video effects, for example in Adobe After Effects.
For the disk subsystem to affect the speed of HD video processing CPUs will have to become faster.
Editing images in Adobe Photoshop is quite a different story. Of course, ordinary images from digital cameras are well below 4 gigabytes but Photoshop keeps track of user actions and may require more memory for that. Even at the default settings (20 recent states are recorded) this image-processing application begins to access the hard disk actively and its speed becomes dependent on the disk subsystem performance.
As we can see, the speed of our test scenario (it is an improved Retouch Artists Photoshop Speed Test that includes the processing of four 10-megapixel images captured with a digital camera) depends heavily on the disk. The difference between the WD Caviar Green and the Intel X25-M G2 is twofold! However, you can achieve even better results by increasing your memory amount. The numbers indicate that 8 gigabytes of RAM help do without disk caching, producing a fantastic performance boost.
Well, talking about image processing at large, we must admit that Photoshop is not a typical application. Its memory appetite is quite a peculiar trait. When we take another program, for example when we process a batch of similar images in Adobe Photoshop Lightroom, the influence of system memory is very low.
The choice of a disk storage device can only improve the resulting performance by 10% at best.
We are also interested in checking out our disks and memory in modern 3D visualization suites. We took Autodesk Maya as it has a SPEC benchmark operating with rather complex models.
Well, Maya seems to be perfectly satisfied with 4 gigabytes of system memory and does not depend on the speed of the disk subsystem.
The next test is about compiling large software projects.
There are but negligible performance benefits from the faster disk or larger amount of system memory even though the computer was rattling with the HDD’s heads quite actively during this task.
When it comes to gaming, the owner of a faster disk subsystem will surely enjoy faster loading of game applications, levels and saves. That's obvious, so let’s instead check out the influence of the disk and memory subsystems on the gameplay proper, i.e. on the frame rate.
As a matter of fact, we tried a number of games for this review and never met any game that wouldn't be satisfied with 4 gigabytes of RAM. Thus, modern gaming applications are unlikely to use the swap file on a typical PC configuration. Here are the numbers:
As you can see, there is no dependence of the frame rate on the speed of the disk or amount of system memory. Note that both the average and bottom frame rate is the same, indicating that all data fits comfortably into system memory. The games do not have to load data from the disk as long as you stay within the same game level.
Well, you can feel the benefits of a fast disk or a large amount of memory in games if you change your usage model. If you use your computer as a gaming console, there is no need for an SSD and 8 gigabytes of RAM. But if you are playing a game while running some programs in the background, the situation may be different. It is different even if the game and the background application both fit into system memory.
To illustrate this, we measured the frame rate of Far Cry 2 while running HandBrake in the background to encode HD video. These programs were not larger than 4 gigabytes, yet the faster disk and additional memory increased the bottom frame rate considerably.
Although the average frame rate is unaffected, the background activities may provoke slowdowns in the game you are playing. You need a faster disk or more memory to avoid that.
Our tests in real-life applications and complex benchmarks help us chart out the preferable ways to upgrade. As we’ve found out, it is preferable to use a fast SSD in your system than to increase the amount of system memory above 4 gigabytes.
The standard amount of 4 GB is enough for most situations and adding more memory doesn’t bring any noticeable benefits. You can only get a positive effect from such an upgrade if you've got a very slow HDD in your system. The SuperFetch technology which uses system memory for look-ahead data caching may make your computer more responsive then. However, it would be wiser to spend the money to improve your disk subsystem and get much better results.
Windows 7, especially in its 64-bit version, has got an advanced memory manager that can put the available system memory to very good and efficient use. This newer OS needs less memory than Windows Vista. Even if you see in Task Manager that nearly all of your physical memory is used up, this doesn't mean your computer lacks it. This is just the result of effective data caching.
As opposed to additional RAM, a fast disk subsystem based on a solid state drive is going to be advantageous always. Besides improving the speed of disk-accessing applications, an SSD increases the overall responsiveness of the platform. The OS and applications take less time to start up and files open up faster. Although such things cannot improve your productivity much, they do make the computer more enjoyable to use.
You must realize that upgrading your disk subsystem is not a universal solution. There are some specific applications for which 8 gigabytes of system memory may prove preferable to a faster disk. For example, Adobe Photoshop can use up even 8 GB of RAM. You also need a large amount of memory to run virtual machines (like VMware or Oracle VM Virtual Box). In this case, you won’t have to stop one VM to launch another as you will be able to simply switch between them. Such usage scenarios are professional, though. And professionals do not ask questions about upgrading. They already know what they need to get higher performance.
Summing everything up, here is our recommendation for ordinary users. If you choose between different ways to upgrade your high-performance computer further, we suggest that you consider improving your disk subsystem by using SSDs. It is only if you really feel a lack of system memory and the OS’s incessant swapping makes you irritated that you should think about adding more RAM.