i5 / P55 Lab Update - Now with more numbers
by Gary Key on September 15, 2009 12:05 AM EST- Posted in
- Motherboards
i5 / P55 Lab Update -
We welcomed Anand back into the office with open arms this past weekend. He immediately started working on an in-depth analysis of clock for clock comparisons for several processors as a follow up to our Lynnfield launch article (among many other things). This analysis along with a quick i7/860 performance review will be available in the near future.
In the meantime, I have additional performance results using the P55 motherboard test suite along with some unusual results from our gaming selections. I am not going to dwell on with commentary in this short update. We will let the numbers speak for themselves at this point. Let’s get right to the results today, but first, the test setup.
Test Setup-
For our test results we setup each board as closely as possible in regards to memory timings and sub-timings. The P55 and 790FX motherboards utilized 8GB of DDR3, while the X58 platform contained 6GB. The P55 and X58 DDR3 timings were set to 7-7-7-20 1T at DDR3-1600 for the i7/920, i7/870, and i7/860 processors at both stock and overclocked CPU settings.
We used DDR3-1333 6-6-6-18 1T timings for the i5/750 stock setup as DDR3-1600 is not natively supported in current BIOS releases for this processor at a stock Bclk setting of 133. We had early BIOS releases that offered the native 1600 setting but stability was a serious problem and support was pulled for the time being. Performance is essentially the same between the two settings. When we overclocked the i5/750 to 3.8GHz, we utilized the same DDR3-1600 7-7-7-20 1T timings as the i7 setups.
The AMD 790FX setup is slightly different as trying to run DDR3-1600 at CAS 7 timings on the 1:4 divider is extremely difficult. DDR3-1600 is not natively supported on the Phenom II series so this divider is provided with a caveat that you are overclocking the memory bus. The same holds true for the Lynnfield (i7/8xx, i5/7xx) processors as DDR3-1333 is officially the highest memory speed supported and it is DDR3-1066 for the Bloomfield (i7/9xx).
Without resorting to some serious overvolting and relaxing of sub-timings, we set our AMD board up at DDR3-1600 8-8-8-20 1T timings. The difference in performance between C7 and C8 DDR3-1600 is practically immeasurable in applications and games on this platform. You might pick up an additional few tenths of second in SuperPi or a couple of extra points in AquaMark or 3DMark 2001SE, but otherwise performance is about equal.
However, in order to satisfy some of our more enthusiastic AMD supporters, we also increased our Northbridge speed from 2000MHz to 2200MHz to equalize, if not improve, our memory performance on the AMD system. Yes, we know, further increasing the NB speed will certainly result in additional performance but the focus of this short article is to show clock for clock results at like settings. Personally, I would run DDR3-1333 C6 with 8GB as this platform favors tighter timings over pure bandwidth.
Last, but not least, I only ran the i5/750 without turbo enabled and the P45/C2Q setup is missing. I am still completing those numbers. Anand will be providing additional analysis on the other Lynnfield processors in his update. The image gallery below contains our Everest memory results with each processor overclocked at similar memory settings along with voltage/uncore/subtiming options. I will go into these in more detail once the motherboard roundups start. For the time being, the 860/P55 offers slightly better throughput and latency numbers than the 920/X58 when overclocked. At stock, the numbers favor the Lynnfield, but primarily due to the turbo mode.
Other than that we are in a holding pattern on the P55 roundups at this time trying to figure out some unusual game and 3D Render results with our GTX275 video cards. I will discuss this problem in the game results.
Facebook
Twitter
RSS
77 Comments
View All Comments
glenster - Wednesday, September 16, 2009 - link
Another consideration for comparison is that the Core i7 960 3.2GHz is due Q4 2009 for $562--see Wikipedia > Nehalem > 45 nm processor architecture. Another site says it's due Oct.18.grimpr - Tuesday, September 15, 2009 - link
Its always fun to watch cases of reversal...in this case, that the AMD/Nvidia combination is proven superior to the gamers and the oems market favorite Intel/Nvidia combo. We can sense the suspicions, fear and doubt on Garys comments and we wait more from him for our further enjoyment. Gary, keep on the Intel crusade...Gary Key - Tuesday, September 15, 2009 - link
There is no fear or doubt, just a quest to figure out why there is such a difference between the NV and AMD GPUs on each platform. If you are an Intel crusader I guess your suspicions would lead you to believe that NV is dogging performance on the Intel platforms.If you are an AMD crusader, you would think AMD is dogging performance on their own platform. Think about that for a moment. I seriously doubt either company is dogging it as they both depend heavily on GPU sales for their survival.
Please read this and just about any other article I have written on AMD and tell me if you really believe I am on an Intel crusade - http://www.anandtech.com/mb/showdoc.aspx?i=3506">http://www.anandtech.com/mb/showdoc.aspx?i=3506 . ;) I care about the best possible performance for the best possible price, with a dash of support thrown in. No hidden agendas around here and if you think so, you would be wrong.
yacoub - Tuesday, September 15, 2009 - link
The rest of us know it's not the case, Gary, but the trolls are just tossing out bait to get you to bite. Ignore 'em.lopri - Wednesday, September 16, 2009 - link
For real. If the test was done using dual-channel on X58 there'd be a huge outrage from certain users how X58 was arbitrarily disadvantage.I have no idea why Gary felt the need to defend himself. Any reasonable reader who followed his articles should know he is the most trust-worthy journalist, bar none. Plus many other writers who know alot about technology but don't know how to write, He is a rare case who actually can write. Look at the first sentense of this article: "We welcomed Anand back into the office with open arms this past weekend." This brought a smile in my face and you can't help but feel disarmed and charmed.
Dispite it being short, the amount of information in this article is simply overwhelming. What's more, it is apparent that the testings were pursued from actual users' perspectives. Arrogant, nasty, and sensational analysis might help raising controversies (and page views), but in the end what benefits us enthusiasts most is these painstaking and seemingly mundane tests that the author undertakes on behalf of the readers.
Still, as is often the case, Gary choose to be modest than aggrandizing ("Quick Thoughts" instead of "harder, deeper, baby!" / "choice for me" instead of "Dark knight to rescue the civilization"), and I am very satisfied to find myself identifying with the author as a potential user with regard to the testing methodology. Plus, seemingly impossible multitasking tests on top of myriads of the configurations tested substantiate the author's subjective opinion significantly, and I for one am very grateful that someone else took all that beating!
Now if I could make a request that could possibly make the author's life even more miserable - any chance to add virtualization to the multitasking mix? With the introduction of XP mode in Windows 7, I think the time has finally come that virtualization has some relevance to desktop users.
TA152H - Tuesday, September 15, 2009 - link
Gary,There are a couple of missing data points that would be helpful, and you probably should include with every article regarding benchmarks.
What were the uncore rates on these processors? Were the i7 920 and 860 running with the same uncore speed?
Also, were you using the D0 i7 920, which offers slightly higher performance, lower power use, and better clock speed potential, or the older C0?
Lastly, do you think it was fair for the Lynnfields to have more memory? I'm guessing this might have hurt their results a little in most applications since typically one has to relax timings. Did you test for this at all to see if there was additional latency when adding the extra memory? Probably it made no difference, but it could. Also, why not test the Bloomfield in dual channel memory mode? Certainly someone will use it this way, although not too many people (say someone has good memory from a C2 system, in pairs, he or she might be inclined to use it 'for now', with an eye towards faster memory in the future).
Gary Key - Tuesday, September 15, 2009 - link
The processor steppings are in the setup table. Yes, we did use the D0 stepping on the 920.The uncore and NB speeds are in the CPUZ screenshots in the gallery. The boards were all run at their stock ratios (except the 965BE as explained) at both base speeds and when overclocking.
Overclocking -
920 - 3407MHz
750 - 3210MHz
860 - 3612MHz
965BE - 2200MHz
Memory Timings were kept as equal as possible between all the systems,except for the 790FX/965BE as explained. The only difference in memory timings between the Bloomfield and Lynnfield systems was tRFC, 88 for Bloomfield, 89 for Lynnfield.
Running 8GB of memory did not hurt the Lynnfield systems, in fact, a few of the numbers were actually better, mainly WinRAR and AutoCad that will be shown in the full review. I have a lot of motherboards to review and cannot spend any further time comparing additional items like dual channel on 920 vs 860. I will see if we can do that in a memory specific article.
Right now, we ran the numbers based on stock ratios, adjusted the base timings to ensure equality, and did our best to properly tune each system to ensure the settings were as close as possible. Even the Bclk speeds were changed to get as close as possible between the Intel systems, although we were still off 4MHz.
We can play the "what if" game all week long, what if the Bloomfield Uncore was increased to 3800, what if CAS6 timings were used on each system, what if the 965BE NB speed was at 2800MHz, what if we allowed the P55 setup to run memory at 2400MHz instead of curtailing it to 1600 to match the 920/965BE, and the list goes on. Guess what, none those fundamentally change the results between each platform.
Anand will have additional results in the near future, in the meantime, I have a lot of retesting to complete with different video cards for the P55 roundup. :)
TA152H - Wednesday, September 16, 2009 - link
Thanks for the information, I guess I should look closer at the pictures.So, basically, what you're saying is, the i7 920 generally outperformed the 860, despite using a considerably slower uncore, which lowers performance for the 920 not only in memory access, but also lowers the level three cache.
Hmmmmm, and since it's actually easier to run the 920 at higher uncore speeds, based on the flexibility of it, it's a little strange you didn't think this was worth mentioning except to bury it in a picture.
Really strange, indeed.
It just verifies my conviction that you guys are doing everything you can to make the P55 platform look good, although you've traded some of your reputation doing so. I'm not suggesting you're getting any money or any stupid stuff like that, just that for some reason you guys fell in love with the P55, and now give 'information' to further that claim.
Not mentioning the uncore is much faster, on a comparison that was supposed to be apple to apple, is a SERIOUS oversight.
Clearly, the 920 would outperform the 860 at the overlcocks at the same uncore. But, you wouldn't want to show that, right?
The i7 920 is a real tough act to follow. If Intel kills it, it would greatly expand the actual usefulness of the Lynnfield. That's why I fear they might. Then again, they might just leave it for the technical savvy since the number of these users is relatively small and the Lynnfield should sell well to the hoi polloi.
pervisanathema - Tuesday, September 15, 2009 - link
There are some really good trolls around here. I'm impressed. Thanks for the laughs.gstrickler - Tuesday, September 15, 2009 - link
After reading the article and the updates (great job Gary), here's the conclusions I draw:1. You can't go wrong with any of these CPUs or motherboards. They're all very fast, your budget and computing needs are more important than the mostly small performance differences between these systems.
2. If you want to OC or want lower power/heat, go with an Intel Core i5/i7 CPU.
3. If you're a hardcore gamer, use an AMD/ATI Radeon card on Core i5/i7 based systems or an Nvidia card on AMD based systems.
4. You don't have to spend more than $300 for a quad core CPU that will deliver top notch performance.
5. The Core i7/860 is comparable to the Core i7/920 in nearly every respect.