Core i5 / P55 Lab Update
by Gary Key on September 11, 2009 12:00 AM EST- Posted in
- Gary's First Looks
This has been an interesting week to say the least for those of us stuck in the labs and not at AMD's DX11 GPU press briefings. Based on feedback from the Lynnfield launch article we have spent the last couple of days running additional benchmarks to address overclocking and clock for clock requests. Yes, we do listen and respond to the comments no matter how outlandish (you know who you are) some may be at times.
I will interject a personal note here, the emails/private messages that outlined a strong case for additional research and testing certainly held a lot more weight than comments like "You are on Intel's payroll...", "Worst review ever...", and the moonshot , "Illegal benchmarking methods..". First off, if we were on Intel's payroll we would not be working here (a logical conclusion, right? ;) ) As for the other comments, everyone is entitled to their opinions. We do our best to keep an open forum and let the comments fall where they may, but offering constructive criticism and facts to back up those comments is what actually causes change, not endless shock posts or attention grabbing statements. I still have hope in people abiding by the rules of Internet Etiquette, but apparently we are still a long ways off from that happening. I will step off the soap box, well, until the next article....
Just to set this up now, our overclock comparisons will be at 3.8GHz for the Core i5/i7 and Phenom II x4 965BE processors. Why 3.8GHz, well it is an easy number for all of our processors to hit on fairly low voltages with retail or mid-range air coolers. It is also an ideal clock range for the "set it and forget crowd" interested in 24/7 overclocking. Certainly we could go higher on air or water cooling and actually ran most of our Core i5/i7 numbers at 4.2GHz for the motherboard roundups. Our Phenom II x4 965BE is the hold up for higher numbers in our clock for clock comparisons.
AMD continues to have serious problems with their Phenom II processor range clocking above 3.8~4GHz on air with a 64-bit operating system. Unfortunately, there is nothing AMD can do to correct this in the current stepping, but they are actively working on improvements with each processor release. In fact, the latest Athlon II x2 processors are the first products we have that allow for 24/7 stable operation at 4GHz under Windows 7 x64. The quad cores are still lagging although our latest retail 965BE is showing promise around 3.92GHz in early testing. I state this now so it does not come as surprise later.
I will post several benchmark results later today based on our motherboard test suite. Anand will provide a more in-depth analysis next week along with an updated look at the Core i7/860. He might even have a surprise announcement from AMD. In the meantime, I have just about completed this additional testing and will return my focus on completing the first (of many) P55 motherboard article(s) that will be up in a couple of days. Our first review will cover the Gigabyte GA-P55M-UD2 among others. We recently received several other micro-ATX P55 motherboards and will look at those shortly. For now, this board is a perfect match for the Core i5/750 for our mainstream audience looking to upgrade an older platform.
Our graph below is an example of the information we will provide late today. Hopefully, this type of information will be useful for your purchasing decision along with our commentary about the results. I know there is not a Core 2 product listed, that will be forthcoming in the near future.
9/11 Update - I am still working on the FarCry 2 and H.A.W.X. benchmarks so the short update will be delayed until tomorrow morning.
Facebook
Twitter
RSS
159 Comments
View All Comments
jasonbird - Friday, December 25, 2009 - link
http://www.socheapwholesale.com">http://www.socheapwholesale.comhttp://www.1stjerseys.com">http://www.1stjerseys.com
http://www.socheapwholesale.com/NFL-Jerseys.html">http://www.socheapwholesale.com/NFL-Jerseys.html
http://www.1stjerseys.com/NHL-Jerseys.html">http://www.1stjerseys.com/NHL-Jerseys.html
Eeqmcsq - Friday, September 11, 2009 - link
You guys should just ban snakeoil. All he does is make outlandish and idiotic claims just to get a rise out of other commenters. He does this too at Techreport. Now he's going to think that he's justified in all the noise he tries to stir up. It really ruins the more intellectual atmosphere I've come to expect out of Anandtech comments.Anyway. Personally, the test combo that I'm more interested in is testing with Turbo off as well as on. Since the performance boost provided by Turbo is variable based on whatever other apps the user or operating system may be multitasking in the background, testing with Turbo off provides the "guaranteed minimum", i.e. a baseline level of performance that a Turbo capable CPU will provide, and that any Turbo boosted extra is considered a free bonus.
Otherwise, people who skim through the graphs may think that the CPU will ALWAYS provide the Turbo boosted score, instead of understanding that the Turbo boosted score is an "ideal" score, when conditions of the CPU are at its best for Turbo.
Anyway guys, keep up the good and hard work.
Inkie - Sunday, September 13, 2009 - link
"You guys should just ban snakeoil. All he does is make outlandish and idiotic claims just to get a rise out of other commenters"Then they'd have to ban TA152H and others for the same reason. Just let them speak. I think it is an error for Anand or Gary to reply to them though. That just gives them status in their own eyes and encourages them to continue.
TA152H - Sunday, September 13, 2009 - link
Why are you bringing me into this?Actually, I back up my clains, unlike what you did. I don't just say this or that sucks, or was done wrong. I tell them why, and back it up.
Gary can complain about shock posts, but, I put more effort into my posts than they put into their P55 motherboard pictorals :-P. And I DO back them up.
Most things in life are not black or white, good or bad, but a shade of gray. People are simpletons, and like it simple, but it's just not how most things are. P55 isn't a perfect product, by any means, and it's not all bad. If you don't have people pointing out the flaws in the methodology, then you have holes in your information. Feedback is not only acceptable, it's necessary.
More to the point, when you idiots were all pointing out how good the performance was on the first P55 article, I was explaining the memory performance was abysmal for what the product was. No one else seemed to see it. Guess what? I was right. The numbers were improved for the released product (and I kept saying I thought it was from pre-release hardware), so the terrible performance didn't make sense after all. Did the writer see it? Nope. He was saying how it was the greatest thing since cheddar cheese? Did the masses see it? Nope. Like proper cattle that can't think for themselves, they just ate it up and did as they were told, so to speak.
You need critical people that are going to look at what you do, and take you to account when you do something wrong, or just don't do something. I need it when I do things to. I don't always like being shown it, but, at the end of the day, the end result is better.
That's really what they should be here for, right?
Inkie - Sunday, September 13, 2009 - link
Actually I've hardly made any posts on Anandtech in the past. I've just read your particular brand of posting recently, since I was interested in the 1156 platform. You have said many incorrect things about the 1156 platform, both before and after release, using ridiculous, hyperbolic language to do so and what's worse, you never acknowledge your mistakes and habitually attack others. You are no better than someone like snakeoil. Actually, snakeoil may be wrong, but he is a bit less free in insulting others. I think you are funny sometimes though, just because you can be so wrong and technically incompetent. Writing this post, I still have to chuckle eg. about what you said about pin-multiplexing on the DDR3 and PCIe interfaces. I don't want to start a flamewar, so I won't reply any further. No point anyway.TA152H - Monday, September 14, 2009 - link
I read stuff from little losers like you, and I'm wondering if you even have the common sense to realize the hypocrisy in what you're saying. Are you really that stupid? I guess you are.You insult me, and then whine that I insult people? Duh. What a moron.
You say I was wrong about things, but I don't see a single quote.
You have heard about multiplexing pins right? Processors have done it before, in fact the 8086 did it with the address and data bus. Intel shaved off 210 pins. I haven't seen the pinout yet, but it's probably out there. Oh, and I never said they multiplexed the pins, I said it's a possibility they did, and it would be nice if we got information like that instead of pictures of motherboards.
Keep it straight, OK loser?
Inkie - Monday, September 14, 2009 - link
There is no multiplexing of pins across the PCIe and DDR3 interfaces, I can tell you. I'm not going to waste time quoting you. Notice the only thing that could really be interpreted as an insult is that you are sometimes funny by virtue of being so wrong headed. Your mistakes, hyperbolic language, inabality to admit your mistakes, attacks on others or whatever are just facts that anyone can see.So, I did reply further. But, what do you care, I'm a loser and a moron, right?
Inkie - Monday, September 14, 2009 - link
Just a further expand the point: you realise that the DDR3 and PCIe interfaces are separate interfaces, routed to different places on the motherboard? I was really laughing that you would even make such a speculation.Inkie - Sunday, September 13, 2009 - link
"Anyway. Personally, the test combo that I'm more interested in is testing with Turbo off as well as on. Since the performance boost provided by Turbo is variable based on whatever other apps the user or operating system may be multitasking in the background, testing with Turbo off provides the "guaranteed minimum", i.e. a baseline level of performance that a Turbo capable CPU will provide, and that any Turbo boosted extra is considered a free bonus."Turbo off doesn't really give a "guaranteed minimum". If there is heavy mulitasking then performance on a given task can be much worse than the turbo-off score for that task measured in comparitive isolation.
Eeqmcsq - Sunday, September 13, 2009 - link
Yes, it's true that extremely heavy multitasking can overload a quad core CPU with too many threads, thus degrading performance for all tasks, but that's not the fault of the CPU. That's simply more tasks than the CPU can handle. Also there's the side issue of disk contention as multiple threads fight for access from the hard drive. That could also affect a benchmark, but that's also not the fault of the CPU.But I'm thinking of cases where the background multitasking pushes 1 or 2, maybe even 3 cores up to a constant and prolonged 100%, but the quad core still has enough "room" to theoretically run a single threaded benchmark without being affected by the background tasks.
For example, suppose an instance of Firefox and an instance of Openoffice get into a buggy state where each are running an infinite loop, thus is not causing disk contention. In this case, 2 of the 4 cores will reach a prolonged 100%, but there are still 2 cores available for a single threaded benchmark, such as Cinebench. On a Core 2 or Phenom II, there should NOT be a drop in Cinebench performance. But on a Lynnfield, Turbo either can't kick in, or will kick in with much less of a boost, so there SHOULD be a drop in Cinebench performance. Assuming all other conditions aren't otherwise affecting Cinebench, the performance drop should theoretically be no worse than a Turbo-off test. That's what I'm referring to about a "guaranteed minimum".
This is also why I suggested a multi-instance tests of single threaded benchmarks. That way, we can see how much of a drop off there is for 1x, 2x, 3x, 4x concurrent instances of single-threaded-Cinebench both in non-Turbo CPUs of Core 2 and Phenom 2, and also in the Core i5/i7/etc. Hopefully, Gary has read my suggestion and added such a test.