Intel Dual Core Performance Preview Part II: A Deeper Look
by Anand Lal Shimpi on April 6, 2005 12:23 PM EST- Posted in
- CPUs
The Impact of NCQ on Multitasking Performance
Just under a year ago, we reviewed Maxtor's MaXLine III, a SATA hard drive that boasted two very important features: a 16MB buffer and support for Native Command Queuing (NCQ). The 16MB buffer was interesting as it was the first time that we had seen a desktop SATA drive with such a large buffer, but what truly intrigued us was the drive's support for NCQ. The explanation of NCQ below was from our MaXLine III review from June of 2004:
Hard drives are the slowest things in your PC and they are such mostly because they are the only component in your PC that still relies heavily on mechanics for its normal operation. That being said, there are definite ways of improving disk performance by optimizing the electronics that augment the mechanical functions of a hard drive.
Hard drives work like this: they receive read/write requests from the chipset's I/O controller (e.g. Intel's ICH6) that are then buffered by the disk's on-board memory and carried out by the disk's on-board controller, making the heads move to the correct platter and the right place on the platter to read or write the necessary data. The hard drive is, in fact, a very obedient device; it does exactly what it's told to do, which is a bit unfortunate. Here's why:
It is the hard drive, not the chipset's controller, not the CPU and not the OS that knows where all of the data is laid out across its various platters. So, when it receives requests for data, the requests are not always organized in the best manner for the hard disk to read them. They are organized in the order in which they are dispatched by the chipset's I/O controller.
Native Command Queuing is a technology that allows the hard drive to reorder dynamically its requests according to the location of the requests on a platter. It's like this - say you had to go to the grocery store and the drug store next to it, the mall and then back to the grocery store for something else. Doing it in that order would not make sense; you'd be wasting time and money. You would naturally re-order your errands to grocery store, grocery store, drug store and then the mall in order to improve efficiency. Native Command Queuing does just that for disk accesses.
For most desktop applications, NCQ isn't necessary. Desktop applications are mostly sequential in nature and exhibit a high degree of spatial locality. What this means is that most disk accesses for desktop systems occur around the same basic areas on a platter. Applications store all of their data around the same location on your disk as do games, so loading either one doesn't require many random accesses across the platter - reducing the need for NCQ. Instead, we see that most desktop applications benefit much more from higher platter densities (more data stored in the same physical area on a platter) and larger buffers to improve sequential read/write performance. This is the reason why Western Digital's 10,000 RPM Raptor can barely outperform the best 7200 RPM drives today.
Times are changing, however, and while a single desktop application may be sequential in nature, running two different desktop applications simultaneously changes the dynamics considerably. With Hyper Threading and multi-core processors being the things of the future, we can expect desktop hard disk access patterns to begin to slightly resemble those of servers - with more random accesses. It is with these true multitasking and multithreading environments that technologies such as NCQ can improve performance.
In the Maxtor MaXLine III review, we looked at NCQ as a feature that truly came to life when working in multitasking scenarios. Unfortunately, finding a benchmark to support this theory was difficult. In fact, only one benchmark (the first Multitasking Business Winstone 2004 test) actually showed a significant performance improvement due to NCQ.
After recovering from Part I and realizing that my nForce4 Intel Edition platform had died, I was hard at work on Part II of the dual core story. For the most part, when someone like AMD, Intel, ATI or NVIDIA launches a new part, they just send that particular product. In the event that the new product requires another one (such as a new motherboard/chipset) to work properly, they will sometimes send both and maybe even throw in some memory if that's also a more rare item. Every now and then, one of these companies will decide to actually build a complete system and ship that for review. For us, that usually means that we get a much larger box and we have to spend a little more time pulling the motherboard out of the case so we can test it out on one of our test benches instead - obviously, we never test a pre-configured system supplied by any manufacturer. This time around, both Intel and NVIDIA sent out fully configured systems for their separate reviews - two great huge boxes blocking our front door now.
When dissecting the Intel system, I noticed something - it used a SATA Seagate Barracuda 7200.7 with NCQ support. Our normal testbed hard drive is a 7200.7 Plus, basically the same drive without NCQ support. I decided to make Part I's system configuration as real world as possible and I used the 7200.7 with NCQ support. So, I used that one 7200.7 NCQ drive for all of the tests for Monday's review. Normally, only being able to run one system at a time would be a limitation. But given how much work I had to put into creating the tests, I wasn't going to be able to run multiple things at the same time while actually using each machine, so this wasn't a major issue. The results turned out as you saw in the first article and I went on with working on Part II.
For Part II, I was planning to create a couple more benchmarks, so I wasn't expecting to be able to compare things directly to Part I. I switched back to our normal testbed HDD, the 7200.7 Plus. Using our normal testbed HDD, I was able to set up more systems in parallel (since I had more HDDs) and thus, testing went a lot quicker. I finished all of the normal single threaded application benchmarks around 3AM (yes, including gaming tests) and I started installing all of the programs for my multitasking scenarios.
When I went to run the first multitasking scenario, I noticed something was very off - the DVD Shrink times were almost twice what they were in Monday's review. I spent more time working with the systems and uncovered that Firefox and iTunes weren't configured identically to the systems in Monday's review, so I fixed those problems and re-ran. Even after re-running, something still wasn't right - the performance was still a lot slower. It was fine in all other applications and tests, just not this one. I even ran the second multitasking scenario from Monday's review and the performance was dead on - something was definitely up. Then it hit me...NCQ.
I ghosted my non-NCQ drive to the NCQ drive and re-ran the test. Yep, same results as Monday. The difference was NCQ! Johan had been pushing me to use a Raptor in the tests to see how much of an impact disk performance had on them, and the Raptor sped things up a bit, but not nearly as much as using the 7200.7 did. How much of a performance difference? The following numbers use the same configuration from Monday's article, with the only variable being the HDD. I tested on the Athlon 64 FX-55 system:
Seagate Barracuda 7200.7 NCQ - 25.2 minutes
Seagate Barracuda 7200.7 no NCQ - 33.6 minutes
Western Digital Raptor WD740 - 30.9 minutes
The performance impact of NCQ is huge. But once again, just like the first NCQ article, this is the only test that I can get to be impacted by NCQ - the other Multitasking Scenarios remain unchanged. Even though these numbers were run on the AMD system, I managed to get similar results out of the Intel platform. Although, for whatever reason, the Intel benchmarks weren't nearly as consistent as the AMD benchmarks. Given that we're dealing with different drive controllers and vastly different platforms, there may be many explanations for that.
At first, I thought that this multitasking scenario was the only one where NCQ made an impact, but as you'll find out later on in this article, that's not exactly true.
106 Comments
View All Comments
segagenesis - Wednesday, April 6, 2005 - link
#45 - The Athlon 64 still lags behind on encoding cabability and its been shown over the past year. However some of the content tested was designed specifically for P4 optimization so its hard to get a reliable result. Who knows... until we see the new AMD64 core with SSE3 we cant really pass judgement.Quanticles - Wednesday, April 6, 2005 - link
AMD needs to send Anandtech one of their dual-cores with a DFI nForce4 SLI board.That'd shut a lot of people up.
If AMD owns single thread, then why wouldnt they own dual core. People would be throwing money away buying intel's dual core.
Spearhawk - Wednesday, April 6, 2005 - link
The first graph on the DVD Shrink/Game bench still seems a bit suspect. Why would the P4 outperform the PD and the A64 when under normal circumstances the A64 should be superior in singlethread and the PD in multithread?Anway, great article. I'm really looking forward to seeing what AMD has to offer since while I'd greatly like improved multitasking I'd also like a good gaming CPU.
Marlin1975 - Wednesday, April 6, 2005 - link
What about overclocking the Dual core chip???Regs - Wednesday, April 6, 2005 - link
Ah, I mis read it. You used the Intel 955X.I have to ask then, is it because of AMD's onboard memory controller that they don't have to manufacture another motherboard for the dual core CPUs? If so, you think AMD was thinking about this scenero (dual cores) well before the first clawhammer came out?
That would sound impressive. Compared to Intel's dual press hots.
segagenesis - Wednesday, April 6, 2005 - link
#37 - Well im hard pressed to really want one when my current setup is still sufficient. That and I have my heart set on AMD64 for gaming. Even at DVD backup I can do a movie in about an hour even with full quality under DVD Shrink. And really, I would use DVD Rebuilder which is very much single threaded and in my book Quality > Speed. Takes about 6-7 hours but better results than DVD Shrink could have wished for.Anand Lal Shimpi - Wednesday, April 6, 2005 - link
RegsThe nForce4 Intel Edition platform was unrelated to the Intel dual core platform, it was just something I was working on at the time.
The platform arrived DOA, I'm guessing it's an error on NV's part.
Take care,
Anand
Jeff7181 - Wednesday, April 6, 2005 - link
"AMD's dual core will be quite impressive, even more so than Intel's. Don't look at performance as the only vector to measure though... "You like to tease us, don't you? :)
Regs - Wednesday, April 6, 2005 - link
Hey Anand, I noted that you said in Part one that the Intel Dual core was one of the most stable config's you ever had. However in part two and quote, "After recovering from Part I and realizing that my nForce4 Intel Edition platform had died, I was hard at work on Part II of the dual core story. "Was this human error or was it a manufacturing error?
michael2k - Wednesday, April 6, 2005 - link
#26: Actually, encoding a DVD should be a multi-tasking event in of itself!Task1: DVD creation; menu's, transitions, etc
Task2: DVD encoding; background rendering of menu's and transitions
Task3: DVD encoding; background rendering of the actual menu
Task4: Burning of DVD; you should be able to start burning the DVD before the encoding in task 2 or 3 is complete, as long as the burner is properly buffered