| Marvell SATA-6G SSD Performance vs Intel ICH10 |
| Articles - Featured Guides | |
| Written by Olin Coles | |
| Friday, 04 December 2009 | |
Marvell SATA 6G SSD Performance vs Intel ICH10When the Seagate Barracuda XT was revealed as the World's first-ever third generation SATA 6Gb/s hard drive, expectations for performance were raised to match the new bandwidth limit. While this drive offered tremendous storage capacity paired to good performance, the transfer speeds were nowhere near the second generation SATA-3 Gb/s limit. This is where Benchmark Reviews turns to our vast collection of Solid State Drive storage products, some of which eclipse the SATA-II range. However, our history of SSD performance tests on the Intel ICH10 Southbridge would prove to be very different than our results on the new Marvell 88SE9123 or 88SE9128 SATA-III chips. In this article, Benchmark Reviews compares Marvell SATA-6G SSD performance against tests on the Intel ICH10 chipset.
This project was created from by the perfect storm: Benchmark Reviews was already testing the Seagate Barracuda XT ST32000641AS on the ASUS P7P55D Premium (Intel P55/ICH10 + Marvell 88SE9123) and Gigabyte GA-P55A-UD6 (Intel P55/ICH10 + Marvell 88SE9128) motherboards. After receiving less than impressive differences between SATA-II and SATA-III benchmark performance with the new Seagate SATA 6G hard drive, a handful of SSDs were connected to help find potential differences. There were differences, to be sure, but not the kind that improve performance. Our results didn't fit-in well with the respective review projects, as they were off-topic, so this Marvell SATA 6G SSD Performance vs Intel ICH10 article was created, The basis of this article revolves around two notions: SATA 6G connections offer more bandwidth than SATA 3G controller did, and that SATA-II products that came close to the 3 Gb/s limit may see some improvement with the new SATA-III controller. Although it is still very early on in the SATA 6G lifecycle (Marvell announced the 88SE9128 SATA-III 6Gb/s controller on 22 September 2009), and only one major manufacturer offers a viable storage controller (the aforementioned Marvell Technology Group), the media hype has create a very high bar for expectations. So what's to be expected? About Intel CorporationIntel (NASDAQ: INTC), the world leader in silicon innovation, develops technologies, products and initiatives to continually advance how people work and live. Additional information about Intel is available at www.intel.com/ About Marvell Technology Group Ltd.Marvell (NASDAQ: MRVL) is a leader in the development of storage, communications, and consumer silicon solutions. The company's diverse product portfolio includes switching, transceiver, communications controller, wireless, and storage solutions that power the entire communications infrastructure including enterprise, metro, home, and storage networking. As used in this release, the terms "company" and "Marvell" refer to Marvell Technology Group Ltd. and its subsidiaries. For more information, visit https://www.marvell.com. Drive Testing MethodologyComparing a Solid State Disk to a standard Hard Disk Drives is always relative; even when you're comparing the fastest rotational spindle speeds. One is going to be much faster in response time (SSD's), while the other is usually going to have higher throughput bandwidth (HDD's). Additionally, there are certain factors which can affect the results of a test which we do our best to avoid. Solid State Drives have traveled a long winding course to finally get where they are today. Up to this point in technology, there have been several key differences separating Solid State Drives from magnetic rotational Hard Disk Drives. While the DRAM-based buffer size on desktop HDD's has recently reached 32 MB and is ever-increasing, there is still a hefty delay in the initial response time. This is one key area in which flash-based Solid State Drives continually dominates because they lack moving parts to "get up to speed". However the benefits inherent to SSD's have traditionally fallen off once the throughput begins, even though data reads or writes are executed at a high constant rate whereas the HDD tapers off in performance. This makes the average transaction speed of a SSD comparable to the data burst rate mentioned in HDD tests, albeit usually lower than the HDD's speed. SSD Testing DisclaimerEarly on in our SSD coverage, Benchmark Reviews published an article which detailed Solid State Drive Benchmark Performance Testing. The research and discussion that went into producing that article changed the way we now test SSD products. Our previous perceptions of this technology were lost on one particular difference: the wear leveling algorithm that makes data a moving target. Without conclusive linear bandwidth testing or some other method of total-capacity testing, our previous performance results were rough estimates at best. Our test results were obtained after each SSD had been prepared using DISKPART or Sanitary Erase tools. As a word of caution, applications such as these offer immediate but temporary restoration of original 'pristine' performance levels. In our tests, we discovered that the maximum performance results (charted) would decay as subsequent tests were performed. SSDs attached to TRIM enabled Operating Systems will benefit from continuously refreshed performance, whereas older O/S's will require a garbage collection (GC) tool to avoid 'dirty NAND' performance degradation. It's critically important to understand that no software for the Microsoft Windows platform can accurately measure SSD performance in a comparable fashion. Synthetic benchmark tools such as HD Tach and PCMark are helpful indicators, but should not be considered the ultimate determining factor. That factor should be measured in actual user experience of real-world applications. Benchmark Reviews includes both bandwidth benchmarks and application speed tests to present a conclusive measurement of product performance. Intel ICH10 Test System
Marvell 88SE9128 System
Marvell 88SE9123 Test System
System Notes:
Drive Hardware TestedThe following storage hardware has been used in our benchmark performance testing, and may be included in portions of this article:
Test Tools
Seagate Barracuda XT: SATA 3G vs 6GWhen Benchmark Reviews first discussed the upcoming SATA 6Gb/s hard drive in our Seagate Barracuda XT Preview, I can honestly say that my expectations were very optimistic. After all, it seemed that a 64MB cache buffer could really improve performance of this 7200 RPM disk, and designing it to be compliant with the third generation SATA specification was only going to raise the headroom for potential bandwidth speed. Being optimistic is very different from being convinced, and speculation doesn't offer the insight that benchmark performance tests can. In this section, Benchmark Reviews tests the Seagate Barracuda XT SATA 6Gb/s hard drive on three different platforms. To help compare and contrast performance differences, we've used three different test systems. All three have Intel's ICH10 Southbridge in common, a chip with a long driver development history that allows 3.0 Gb/s bandwidth throughput. The ASUS P7P55D Premium uses the Marvell 88SE9123 controller and the Gigabyte GA-P55A-UD6 motherboards utilizes a Marvell 88SE9128 SATA 6Gb/s controller. With the processor and memory matched as closely as possible between platforms, our test results should be as accurate as possible. On the ASUS and Gigabyte P55 motherboards our test performance on the ICH10 chipset was identical, so that specific test (labeled Intel P55/ICH10) is manufacturer non-specific.
Observing 4KB IOPS performance over four different SATA controllers, the Seagate Barracuda XT offered very little difference between them all. The ST32000641AS SATA 6Gb/s hard drive did perform well enough compared to other HDD storage devices, but it didn't seem to matter whether it was connected to SATA 3Gb/s or SATA 6Gb/s. So far, the difference between SATA-3G and SATA-6G is practically nil.
Switching to bandwidth speed tests didn't reveal anything new, and the difference between SATA controllers was almost irrelevant. Taking the marketing spin into consideration, the Seagate Barracuda XT was simply a SATA-II level product that received SATA-III compatibility. Other than one generation of future-proofing, there were no differences in performance between SATA-3G and SATA-6G results on the Barracuda XT, and it still remains to be seen what benefit the new designation has earned this high-capacity product. Since the Seagate Barracuda XT is the only SATA-6G storage product on the market (as of early December 2009), the remainder of our tests will utilize top-end SATA-3G products. Please continue on to see how our collection of SATA-II SSDs have fared on the SATA-6G controller. Drive Hardware TestedOCZ Agility SSD: SATA 3G vs 6GWhile Solid State Drive technology is still rather niche in its adoption rate, the OCZ Agility is one of the best values going in the SSD sector. Benchmark Reviews originally tested the OCZ Agility OCZSSD2-1AGT120G back in July 2009, and despite it's value-sensitive focus, the Agility is still among the fastest SSDs available because of its Indilinx 'Barefoot' controller chip. So the question remains, will there be any difference going from SATA-II to SATA-III? That's what Benchmark Reviews plans to answer, and we begin with Input/Output (IO) per second performance tests. Benchmark Reviews tests the performance results of the OCZ Agility Solid State Drive while connected to each of the SATA controllers, and our results are charted below.
Testing 4KB IOPS performance across four difference chipsets showed just how far Intel has come with their driver optimizations, and how far Marvell still needs to go. In the chart above, the OCZ Agility SSD performs at approximately 7,200 IOPS read, and 16,000 write on either the P55 or X58 platform and using Intel's ICH10 SATA controller. Switching to either the Marvell 88SE9123 or 88SE9123 SATA 6G controllers changed the IOPS performance dramatically, and IOPS dropped to approximately 5500 read and 8600 write. This equals a 28% improvement in IOPS read performance with Intel's ICH10 over Marvell's 9128/9123 chips, and 95% increase in write operations per second.
Results in bandwidth speed tests for the Agility SSD were very similar to the IOPS benchmarks. The X58 and P55 platforms produced roughly 222 MBps reads and 165 MBps writes, all while connected to the Intel ICH10 Southbridge. In comparison, the Marvell SATA 6G controllers produced approximately 154 MBps read speed, and 106 MBps write. This results in a 42% advantage for the ICH10 SATA 3G controller over Marvell's 9123/9128 SATA 6G controller in read speed, and a 55% advantage in write-to performance. Why, you might ask? It's all in the driver. You see, Intel's ICH9 and ICH10 chips utilize a driver architecture that creates a virtual RAM-disk buffer for cached transactions. The amount of RAM-disk created depends on the chip and available system memory (which we detail in our spin-off article), but the combined driver refinement and memory cache capability lead to very high performance. This is where Marvell needs to grow, and do more than a simple reverse engineering Intel's work in creating their own driver. Just to make sure we aren't seeing an isolated issue, Benchmark Reviews went on to test several more SSDs on the ICH10/Marvell 9128 and 9123 controllers. Drive Hardware Tested
OCZ Agility EX SSD: SATA 3G vs 6GThe OCZ Agility-EX is a SLC Solid State Drive based on the same Indilinx Barefoot architecture as the MLC Agility SSD. Our thinking was that the OCZ Agility is a MLC product, and there could be a potential problem with its multi-layer cell architecture and the Marvell 88SE9128 or 88SE9123 SATA 6Gb/s chips. There was only one way to remove doubt, and that's with a few tests:
Once again we see that the Intel ICH10 controller offers significantly better performance, regardless of X58 or P55 chipsets. The write-to IOPS performance is substantially better (nearly double - 95%), while the read-from IO is improved approximately 28% on the ICH10 over the Marvell 9128/9123 chip. Next comes bandwidth transfer speeds:
Almost like clockwork, the Intel ICH10 mirrors performance differences that it gained over the Marvel 9128/9123 chip in the IOPS tests. While the OCZ Agility EX produced approximately 222 MBps reads and 204 MBps writes on the ICH10 controller, the performance dropped to only 156/106 with the SATA-III Marvell controllers. These tests illustrate that the performance deficiency is not limited to MLC SSDs, but also SLC Solid State Drive products. In the next section, Benchmark Review tests the OCZ Vertex Turbo to illustrate how well Marvell's SATA-6G controllers work with one of the fastest SSDs made... Drive Hardware Tested
OCZ Vertex Turbo SSD: SATA 3G vs 6GThe OCZ Vertex Turbo is a MLC Solid State Drive based on the Indilinx 'Barefoot' controller, and is one of the fastest high-performance SSDs currently on the market. We've already seen that the generation three SATA-6G Seagate Barracuda XT has demonstrated virtually no difference between 3.0Gb/s and 6.0Gb/s controllers, while SSDs have reacted very differently. The MLC OCZ Agility SSD performed nearly 70 MBps better read transfers and almost 100% better write operations, which the SLC Agility-EX proved was no fluke by matching these results. So now we see how a top-end premium SSD compares:
Measuring the difference between Intel's ICH10 SATA-3Gb/s Southbridge and Marvell's 9128/9123 SATA-6G controller on the OCZ Vertex Turbo produced some significant differences, but nothing quite as dramatic as the Agility and Agility-EX SSDs. IOPS read performance was approximately 6400 on the ICH10 controller, while write IOPS neared 13000. On the Marvell SATA-6G controllers the performance dropped to around 5600 IOPS read and 8400 write at their best. The difference here equals only a 14% advantage over the Marvell 9128/9123 chips during read-from IOPS, or 53% more during write-to IOPS.
While IOPS differences were huge between the Intel ICH10 and Marvell 9128/9123 chips, they weren't as jaw-dropping as we'd seen with the Agility or Agility-EX SSDs. Switching over to transfer rate performance, the OCZ Vertex Turbo picks up the trail and begins to reveal some familiar differences. The ICH10 reads were again 35% better than the Marvel SATA-6G controller, while the write-to performance was 93% better. So far these last three SSDs all share the Indilinx Barefoot controller, so there's yet another variable to negotiate. In our next section, we use the Kingston SSDNow V+ SSD to examine differences in the Samsung S3C29RBB01 controller... Drive Hardware Tested
Kingston SSDNow V+ SSD: SATA 3G vs 6GThe Kingston SSDNow V+ is a Solid State Drive product built on the Samsung S3C29RBB01 controller found in the PB22-J SSD. Benchmark Reviews has added the SSDNow V+ to determine if the problems we're seeing are isolated to the Marvel driver/controller as we suspect, or if it's an Indilinx-based issue. Beginning with IOPS performance, we discover the truth behind our problem:
Apparently, our differences are not only limited to the Indilinx Barefoot controller, as the Samsung S3C29RBB01 controller appears to experience the same phenomenon as all the others. The Kingston SSDNow V+ offered a 57% IOPS read improvement over the Marvell 9128/9123 chips, while the ICH10 increased IOPS write performance by 14%. The only point of interest is that for once the read performance gain was larger than the write performance. Next comes transfer rate tests...
Before looking of the chart and analyzing differences, it's important to note that the Kingston SSDNow V+ is a value-line product, and despite the premium paid for SSD products this is considered an entry-level SSD solution. In our transfer speed results, the Intel ICH10 controller produced approximately 170MBps reads, compared to 150MBps on the Marvell controllers. This amounts to a 13% advantage over the SATA-6GB interface for reads on this SSD. Write performance varied between platforms, but was altogether low despite the connection. Nevertheless, the Intel ICH10 Southbridge offered a substantially better transfer rate of the Marvell 9128/9123 chips. Drive Hardware TestedSATA-6G vs 3Gb/s Final ThoughtsAs of December 2009, the SATA 6Gb/s interface is still in its infancy. The SATA-II (SATA-3Gb/s) interface has been around since 2005, and after nearly five years of real-world production Intel has enjoyed the marketplace advantage. We've recently tested the first SATA 6Gb/s storage device available, and aside from being slightly more future-proof than its counterparts there were no advantages to the new interface. Conversely, Solid State Drive performance actually suffered at the hands of Marvell's 88SE9123 or 88SE9128 controllers. But it's not really fair to single-out the only existing SATA-6G controller available, is it? Since product marketing controls such a strong influence over consumer spending, it's difficult to forgive a disruptive technology. Although SATA 6Gb/s is merely a speed improvement over the previous generation, it's still enough of a change to fuel marketing hype; and that's exactly what we've seen so far. Making hard drive products that are SATA-6G compliant is about as critical as making a lawnmower more aerodynamic. At the end of the day, HDD technology is nowhere near the SATA-3G limit, and even further off from 6.0 Gb/s. This is why we concentrate on Solid State Drive technology.
Marvell vs ICH10 ConclusionConsumers don't like to trouble themselves with the technical minutiae that interests enthusiasts. But it's the little things that make all of the difference; most especially in regard to this Marvell SATA 6G SSD Performance vs Intel ICH10 article. For example, most consumers don't realize that the SATA controller they connect their storage device to actually rely on a portion of system memory to create a virtual RAM-drive buffer. In many cases, the Intel driver for the ICH9/10 Southbridge reserves a memory space for cache transactions. These memory allocations appear in the form of Cache Line Size and Prefetchable Memory Base, and enable faster transaction throughput. But Intel is not alone. Marvell's SE91XX driver actually follows along the same principals, but with a lack of refinement the differences are quite clear. Benchmark Reviews has exposed a terrible flaw in the design, and only time will tell if SSD performance on the Marvell SATA 6Gb/s controller chips will improve to meet the level Intel's ICH10 Southbridge can deliver. Until then it appears that the marketing hype is simply that, and hopefully Benchmark Reviews has helped you avoid a costly mistake. Questions? Comments? Benchmark Reviews really wants your feedback. We invite you to leave your remarks in our Discussion Forum.
Related Articles: |
|
Comments
If you check out the schematics of your motherboard (page 8):
#download.gigabyte.eu/FileList/Manual/mb_manual_ga-p55a-ud6_e.pdf
You will see that the Marvell chip gets it's PCI-Express lane from the P55 chip. The P55 chipset runs it's PCI-E 2.0 ports at half speed what results in the horrific performance you're getting.
I think it would be adviced to withdraw this article and save face.
What was that you were saying about saving face?
Nowhere do you mention why the Marvell controller is slower on those boards and what is needed to be done by motherboard manufacturers to make it fast (or which board(s) are implementing it good). Nor do you add a P55 motherboard that has it implemented correctly.
The way you present it, one might conclude that it's a controller deficiency.
There are many ways to get the Marvell up to speed. At least mention one but instead you run the same test five times. To top it of, you conclude (wrongly) that the deficiency is caused by driver optimizations.
Your claim that I "wrongly" concluded that driver was not yet optimized is also made in ignorance. Since Marvell's chipset operates on the same bus as Intel ICH10 controller, the performance should be similar for lower-speed devices. Look at the transfer rate tests, and you'll see that speeds didn't reach 250 MB/s (half-speed) and they were still faster on the Intel ICH.
Furthermore, since bandwidth has little impact on IOPS performance, please explain how 2.5 lanes would impact very low input/output rates.
About the driver:
Obviously you came to that wrong conclusion because you think the Marvell controller is on the same bus as the Intel controller. Tests have shown that the Intel AHCI controller has around 650MB/s effective bandwidth at it's disposal.
I haven't seen P55's PCI-E ports go over 200MB/s (showing that theoretical bandwidth doesn't mean effective bandwidth).
Another thing that bothers me is that if you have so much doubt about the Marvell driver, why didn't you bother trying the MS AHCI driver? Because that would have show you straight away that nothing is wrong with the Marvell driver (even though it's a really early one).
IOPS are in direct relation with bandwidth.
Did you set the 'GSATA Ctrl Mode' to AHCI?
If 'Turbo SATA3' is enabled, the Marvell controller will bypass the P55 chipset and connect straight to the CPU (effectively taking one controller and thus putting the GPU to 8x mode).
Or you're missing something or your board/bios has 'Turbo SATA3' non-functioning.
Even if the P55 was the only chipset to have 6Gb/s ports, the Intel system has a much faster CPU and more RAM. SATA port performance is affected by your system, so how is this a fair "technology comparison" in any way...
I think it's obvious you didn't spend enough time on this article or with this hardware and you should really consider retracting this article.
Hey Milli, you sound very defensive as if you wrote these drivers or designed the chipset on the Marvell side. Could you please look into having some updated drivers released that can fix the BSOD's?
If you read around the ASUS forums, there have been quite a few RMAs related to this chip. I'm just staying away from mobos with Marvell chips in the future since board manufacturers can't figure out how to implement them correctly...
The title mainly says "Marvell SATA-6G SSD Performance vs Intel ICH10". Not that we are saying anything wrong here since we are comparing these exact 2 controllers. The issue is that someone will see the "Sata-6g" included in the title and immediately assume right off the bat that this is going to be a review based off the 'new' SATA 6Gb/s compared to the old SATA 3Gb/s.
The author does a great job at exposing the infancy of the Marvel driver performance compared to the mature Intel drivers performance however us readers are assuming we are going to be looking at a review showing off the full bandwidth of the Marvel controller even though it's only limited to 5GT/s.
Perhaps the title could have been named something like "Marvell 912x driver performance vs Intel ICH10"
That's just my 2 cents :)
Can you imagine? Its wrong. A marketing trick from PC manufactures. Oh sure...it´s 6gbps speed...but it isn´t a native controller. It´s an addon. So don´t blame us in the future. THIS ARTICLE IS REALLY IMPORTANT! BECAUSE INTEL IS THE BEST AMONG THESE RIDICULOUS MARVELL MANUFACTURES THAT TRICKED ME AND MANY MANY PEOPLE WITH THIS STUPID CONTROLLER. Motherfuc....
Marvell, get jobs at walmart, Please, where you can make a real difference. Your updated firmware that you just made for gig and forget about the other boards is sad, must be to busy handling all the complaints.
Testing raptors in RAID on Marvell, Raptors in RAID on the Ich10r, the Marvell is GARBAGE.
I would love to hear from the IT professionals above how to make the Marvell get higher read write speeds then the Ich10r IN THE REAL WORLD, and not in their fantasy land.
Pathetic speeds from Marvell, have no right to call it 6g, whit its limited using pci-e.
I don't even think if Marvell gets off their butts and puts out a new Firmware for the 88se9128 on Asus boards they can achieve a 600MB transfer.
Thanks again Olin Coles, the only thing decent about having the Marvell, is being able to run an OS on the Marvell and use Ich10r for a RAID set storage, to keep size off the SSD's and speed up by having app's and all small writes to the RAID array,EX- browser. Until the Marvell is on a pci-e 3.0, it should be called Sata 2.5
Otherwise you can´t get them to work as they should.