 |
|||||||||||
|
|
|
|||||
In the second installment of our Radeon HD 5870 Crossfire CPU Scaling Performance article we are including a total of 11 processors from both AMD and Intel. With over 2800 unique results, we have covered a number of processor series in a range of gaming titles...
When we started testing for this article back in January 2010, the Radeon HD 5870 was the most powerful single-GPU graphics card money could buy. Although at the time the dual-GPU Radeon HD 5970 was AMD’s flagship graphics card, a pair of Radeon HD 5870 graphics cards operating in Crossfire mode are actually quite a bit faster thanks to their 17% core clock speed advantage.
Since then Nvidia has stolen the performance crown from AMD in a rather unimpressive fashion, with the too hot to handle GeForce GTX 480 graphics card. Still, pushing aside all the short comings of the new GeForce GTX 480, placing two of these cards in SLI mode provides some pretty incredible numbers. Therefore we do hope to conduct similar testing with GeForce GTX 480 SLI cards in the future, in more modern games such as Battlefield Bad Company 2 which is known for being very CPU dependent. Still, throwing a pair of Radeon HD 5870 graphics cards into a gaming system is serious business for a few reasons. First and foremost this is one of the most powerful dual-card combination's possible, and secondly at over $800 US, it is also one of the most expensive. With all this GPU power, many readers often ask us what kind of CPU they would require to fully utilize such a setup. In the past we have done our best to address this question by testing a range of processors using the latest and greatest graphics cards. Today’s article is no different, as we set out to test a wide range of CPUs with a pair of Radeon HD 5870 graphics cards. However when compared to recent CPU scaling articles featuring the Radeon HD 5970, this version is a little different. The next page titled “Testing Methodology” better explains what I mean, so make sure you read this page before proceeding to the test results. Also, please be aware that the results from this article can in no way be compared to previous articles such as those featuring the Radeon HD 5970, as the test conditions are very different this time round. This article will feature eleven processors, and yes the Core 2 Duo E8xxx series is finally here, so you can all stop asking me about it. The complete lineup includes the Intel Core i7 9xx, Intel Core i5 7xx, Intel Core i3 5xx, Intel Pentium Gxxx, Intel Pentium E6xxx, Intel Core 2 Quad Q6xxx, Intel Core 2 Duo E8xxx, AMD Phenom II X4, AMD Phenom II X2, AMD Athlon II X4 and AMD Sempron 100. In total this article features over 2,800 unique results. However in total 16,800 test were conducted, focusing on the average and minimum frame rates, each test was carried out a total of six times allowing us to record the average result. This has been a massive undertaking and as I just mentioned, has been months in the making. Each series will be tested at six different frequencies ranging from 2.0GHz through to 4.0GHz in 400MHz steps. Please note that on the pages “Compiled Results” we are only comparing the 2.0GHz and 3.6GHz data. Including all the data in a single graph would have you scrolling down the page for the better part of a day. However the entire break down of how the CPU’s scaled at the various frequencies has been included as well on separate pages for those interested. For more information on how the testing has been conducted please read the next page... |
|||||
|
|
Kreeper |
Posted on: 04/06/2010 01:33 PM
Good stuff but where is my Phenom II X3? |
|
Dan |
Posted on: 04/19/2010 02:11 AM
Outstanding review. No effort can be all encompassing; however, this one is excellent and comes closer than most reviews I have read. Thank you for adding the E8600 CPU, and I look forward to Gulftown being added. Regarding Gulftown, excluding the number of cores, I am not sure how much the architectural change will make a difference in >2 core CPU gaming. On a side note, I would like to see 4 core [8 including virtual] Gulftown CPUs that are clocked at 4+ GHz from Intel that can be easily overclocked to >5GHz on air. I suspect that will make a great deal more difference to gamers than 6+ cores [12 including virtual], at least until 22nm arrives. Intel CPUs have been bumping their head against 3.33GHz maximum factory clock for far too long. I noticed in your charts that there was often little or no difference between 3.6GHz and 4GHz. I suspect that may be due to further optimized RAM, as that would take a great deal of time, to make further changes for testing. If a 4 core Gulftown rated at 4GHz or higher from Intel were available I suspect that RAM timing would yield higher benchmark results because little tweaking in that area would be needed in order to keep a slightly overclocked CPU fed. Then again, I could be wrong, though it would be nice to know for certain. Again, thank you. Dan |
|
Java |
Posted on: 04/20/2010 11:13 AM
Great article, Ive been searching for a while for something like this and now that ive found it, its ended up giving me more questions lol. Ive looking for the cheapest upgrade from an athlon 6000, ASUS M3A78 pro, 4gb ram and HD4870 512mb to something that can handle eyefinity. Currently in the homework stages still but I was considering waiting until near the release of HD6xxx card when prices will be at their lowest (hopefully) but my problem is what else will I have to upgrade? my options are 1)Stick with current set up and just add the HD5870 2)Add a phenom II along with the HD5870 and at a later date either get an AM2+ crossfire board or possibly go AM3 if ram prices are lower and add the second HD5870 Tho looking above and using the Pentium E6xxx results @ 2.8GHz as a rough guide, the Athlon 6000 performance looks ok to be able to run a crossfire solution. Im now wondering if its worth upgrading the processor at all and if so whats the best upgrade path on my limited budget. If prices remain the way they are or increase more, Ill be waiting until HD6 series cards are out. /sigh such dilemmas |
|
RickkiR |
Posted on: 04/15/2010 08:52 PM
WOW run some tests LOL. Thanks for the article it was a very interesting read. |
|
Dan Posts: 2 Joined: 2010-04-18 |
Posted on: 04/19/2010 02:55 PM
Steve, If you could find the time please include the following. Because Gulftown is an unknown for gamers, when you get your hands on one [beg or borrow] I would be interested in seeing a repeat of the battery of tests performed in your "Performance Part 2" using just the Gulftown, with the same RAM timings used with the i9xx and i7xx to save time, and add it to the results. Also, because the multiplier is unlocked I would like to see how many GHz an "easy" overclock would yield with two cores turned off, then lower the RAM timings to the point of instability then increase the timings slightly for stability. With this set-up run the benchmarks again. And add the results to "Performance Part 2". I do not expect you to spend a lot of time working with the RAM; just take a few educated shots to find the fastest timing with 4 CPU cores running. We all know that if a CPU is "spinning its wheels" waiting on system memory data crunching will be restricted. Again, for the sake of time you could exclude “Compiled Results” and limit results to 3.6GHz and 4GHz for 6 core Gulftown and limit results to 3.6GHz and maximum "easy" overclock and fast RAM timings with the Gulftown. Enthusiast, especially those on a budget are very interested in maximum bang for the buck. Thank you for your time and effort. Again, you did an outstanding job with your findings. Dan Note: I am the same Dan as the unregistered Dan in the post below. |
|
Steven Walton Posts: 104 Joined: 2010-02-08 |
Posted on: 04/06/2010 09:41 PM
SShooter believe me you do not need to tell me the difference between the Kentsfield and Yorkfield architectures I have run countless tests over the years comparing them. Of course I still want to include the Yorkfield and other architectures for that matter but it’s all just a matter of time. Also Kreeper given time I will also add Phenom II X3 results and Chris it would be interesting to add the six-core processors as well... |
|
ProX |
Posted on: 04/06/2010 12:09 PM
Awesome article thank you so much and finally I get to see my Core 2 Duo results! Impressive work and it looks like my upgrade to the Core i5 750 was worth it. |
|
Chris |
Posted on: 04/06/2010 12:50 PM
This is nuts. I loved the first article but this one is worlds better. You might not want to hear this but can you please add results for the six core processors once the Phenom II X6 is out! |
|
SShooter |
Posted on: 04/06/2010 05:31 PM
At last. I thought. I've been waiting for this test for a month. Everyday i checked in here to see how my new q9550 do compared to my old dualcore and the newer platforms. The test is fine but... Why, oh why did you use an older, qrippled Core 2 Quad in the test when the rest of the cpu's are more top of their lines? Why didn't you use a q9650 instead of the q6700? Faster fsb: 1300MHz vs 1066MHz New build: 45nm vs 65nm More cache: 12MB vs 8MB plus you would had reach a 4GHz overclock on the 45nm c2q too. And btw, you didn't even mention your c2q in the conclusion! According to this test, the c2q is a crap game processor with a couple of high end graphics card in crossfire. Please please, could you update the test with a new (better) Core 2 quad? |
|
Brutalis |
Posted on: 04/07/2010 12:51 AM
Steve, mate, I appreciate all the hard work, but there's something seriously wrong with your graphs. For example the first two graphs included - the ones in the Crysis Warhead Compiled Results page. The first graph displays a much wider gap between 52 and 51 FPS (1 FPS difference) than between 51 and 48 FPS (3 FPS difference). The second graph displays no less than five bars with 43 FPS on them, which are all different lengths. It's illogical and misleading, and I do believe there's probably some kind of involuntary mistake made here. Please reply or correct the graphs as soon as possible. Thanks in advance. |
|
Kreeper |
Posted on: 04/07/2010 01:44 AM
Brutalis you are doing my head in, can you even read. I suggest you read the first paragraph on that page... |
|
Steven Walton Posts: 104 Joined: 2010-02-08 |
Posted on: 04/07/2010 02:32 AM
Posted by Brutalis on 04/07/2010 01:51 AM
Brutalis it was explained in the review why the graphs are the way they are. If you have a better way of displaying all this data please show me I am open to suggestions. |
|
Brutalis |
Posted on: 04/07/2010 07:25 AM
Damn it, I clearly didn't notice THAT paragraph. Sorry :[ It was 4 AM when I posted, guess next time I should read reviews in the morning, after my coffee... Kreeper - no need to go all agro on my azz, dude  |
|
ProX |
Posted on: 04/07/2010 08:32 AM
Yeah they are the same graphs as the first article and they do make sense, but like you said not at 4am. Nothing does then  Kreeper did have a far go but at least you came back to admit your wrong. |
|
Steven Walton Posts: 104 Joined: 2010-02-08 |
Posted on: 04/07/2010 09:01 AM
Posted by Brutalis on 04/07/2010 08:25 AM
Not a problem and kudos for coming back Brutalis. |
|
SShooter |
Posted on: 04/07/2010 09:28 AM
That's great news. Keep up the good work LH. |
|
white |
Posted on: 04/07/2010 08:44 PM
Boo!! Where are the 45nm c2q quads?? L2 cache and die size matter for c2q. Hell, look how much even the E8400 smashes the Q6600 on dual-core optimized games. I was waiting months for this to come out to realize you went with the dead/old Q6600. Now I'll have to look elsewhere to find the results I need! Would have been my always used article for arguments/discussion/explaining but now I can't.  Gj though its good to see how the c2ds keep up. Still well done and thankyou Steve. |
|
JohnnyR |
Posted on: 04/08/2010 12:06 AM
Thank you for all your hard work! Glad to see my E8400@3.8GHz can still put out fps. |
|
ProX |
Posted on: 04/08/2010 12:16 AM
haha "white" I am sure Steve is hating not being at the center of nerd fights on forums now. If you are really that upset about the lack of 45nm Core 2 Quad's use your head a little more, the ingredients are there. Look at how the Pentium E6xxx and Core 2 Quad Q6xxx series compare and that is how the Core 2 Duo E8xxx will compare to the Core 2 Quad Q9xxx. |
|
banakon |
Posted on: 04/08/2010 09:42 AM
Nice article Steve....nice read... |
|
Ross Dent |
Posted on: 04/08/2010 10:13 PM
Amazing, simply amazing! Best GPU/GPU article ever. |
|
Jonny Intel |
Posted on: 04/10/2010 06:36 AM
Great work, not what i was expecting with Part 2 (was expecting more games benched), but again this is a true reflection for the individual CPU and not the GPU. That is what i would have done if im benching CPU and games! Now Steve, bring on the multi-threaded games like Dragon Age, Lost Planet 2, Resident Evil 5, Metro, Dirt 2! |
|
Rasmus |
Posted on: 04/10/2010 04:37 PM
Very nice piece of work. I would really really like to see how much medium settings Anti aliasing and Anisotropic filtering changes things though. I don't think any one who buys a that expensive graphics card wouldn't use those settings. I know testing without gives more room for the cpu's “to stretch their legs,” but it does not give a realistic picture of the difference most gamers would be experiencing. I know it would be a HUGE work to do all the testing again, but could you possible just test one cpu from each series at 3,6GHZ in just a couple of games. Best regards |
|
ProX |
Posted on: 04/10/2010 10:50 PM
Rasmus did you read the original? http://www.legionhardware.com/articles_pages/cpu_scaling_with_the_radeon_hd_5970,1.html |
|
Rasmus |
Posted on: 04/11/2010 11:09 AM
ProX. Yes, i did read that one to. Really great read, but not really what I were looking for. What I was hoping to see ,was what kinda “real life gaming performance” normal gamers with maybe a old core2duo would get if they only upgraded the GPU. How much performance do you loose if you slap a brand new enthusiast GPU (HD5850, HD5870 or similar, as I do not see many of us gamers buying a HD5970) into your old system, instead of building a brand new system. Testing in the most realistic way a gamer would play the games. I am not claiming to know the best way to test, not at all! But I hope we can agree that most gamers spending money on a HD5850 would at least play with some AA/AF. Best regards |
|
ProX |
Posted on: 04/11/2010 12:32 PM
Right well then the first article answered your question did it not? If you enabled AA/AF there is next to no difference in performance between the various processors as most of the load is not being placed on the GPU and not the CPU. That first article is giving you exactly what you want. There is no point testing with the Radeon HD 5850. The whole point of testing with the latest and greatest graphics cards it to show what if any difference the processor will make. If there is little impact with the Radeon HD 5970 with AA/AF enabled then there is sure as hell going to be no difference with a single Radeon HD 5850. |
|
Rasmus |
Posted on: 04/12/2010 12:14 PM
What you are saying is right. And that article COULD have been exactly what I was looking for. But since it only includes core i7 and phenom II x4 it is not. Had there been some older CPU's in that test it would have been perfect. Best regards. |
|
bob smith |
Posted on: 04/28/2010 10:53 PM
awesome article, thank you for all the tests, is there anyway you post some of the screen shots of what those min or avg frame rates look like when playing the game? i would love to see how many explosions or units moving it takes to slow down the super fast systems |
|
RedSky0 |
Posted on: 05/14/2010 03:48 PM
Hey, in case anyone does happen to come across this again, which CPU in the benchmark would approximate an E7200 the best? The G6950/E8600 performance or that of the E6600? Cheers! |
|
ProX |
Posted on: 05/14/2010 09:03 PM
Hey, it would be the E6600 for sure. It only has 1MB less cache but by default works 500MHz faster. So clock for clock it is going to deliver roughly the same performance. |
|
Michael |
Posted on: 05/23/2010 02:04 PM
Obviously much work has gone into these two articles. Upon reading both articles, i surmise there will be a third which will actually be article "1.5" (easy as cake, piece of pie to see), lol |
|
flibble |
Posted on: 07/14/2010 09:25 AM
I just came across this as it was linked from a thread on TH. I'm a little perplexed. You say you test the AMD Phenom II. You don't mention which one! There are 3 different chips on the AM3 platform and then the AM2 ones. The 940 (AM2/AM3 stock 3Ghz), 945 (AM3 stock 3Ghz), 955 (AM3 stock 3.2Ghz), and the 965 (AM3 stock 3.4Ghz) and why bother with the Sempron but not the Celeron which is its counter. |
|
ProX |
Posted on: 07/14/2010 09:50 AM
@ flibble - well then you can imagine how perplexed we were when you failed to read the second page of the article. Also there are no new modern Celeron processors, the Celeron G1101 is OEM only. So why bother, they included enough old crappy LGA775 processors. |
|
Laci |
Posted on: 03/19/2011 11:16 PM
Please make a CPU scaling performance test with the new radeon HD6990 with Phenom II X4 and Core I7 ! !PLEASE! |