Part 2: Building A Balanced Gaming PC

[pauldh]

[citation][nom]Sheol[/nom]So now comes the next point - why are nvidia's GPU-s consistently requiring a faster CPU to show what they can do, while Radeons perform very well even with a dual core?Best example of course is the GTX295 - are nvidia-s drivers really that lousy, or is there something else at play?[/citation]
Good question! Along with additional hardware, we'll allowing a driver refresh for part 3&4. Although with the CPU's OC'ed, we likely won't be able to judge their(driver) impact until 5&6 (SLI/Crossfire).

 

[pauldh]

[citation][nom]dark_lord69[/nom]I wish you did the 4870, cause the 4890 is expensive (Well, more than I want to pay). And I already knew the 4850 wasn't good enough for what I want to run/do.[/citation]
We staggered the cards because we couldn't use the complete GeForce/Radeon lineup. The 4870 was a good buy back when hardware was chosen, but I'm glad we didn't use it here. Have you noticed, like the 4870X2 and 4850, supply is drying up and prices increasing? Given recent pricing/availability, the GTX 260 and HD 4890 turned out to be two of the few cards that have not increased in price.

 

[grimjester]

It seems just absurd to me that the GTX285 needs an x3 720 to beat a 4890 & x2 550 and even that with a slim margin, for a price increase of 290->490. With the odd Nvidia-only CPU scaling, previous benchmarks of the single-GPU GTX series using high-end CPUs are horribly misleading and even back then the GTX:es seemed overpriced.

The bad performance of the 295 with low end processors is not really relevant, imo. No sane system builder would pair it with a dual or triple core processor anyway, so price/performance-wise it's likely a better buy than the 4870x2 despite what a quick glance at the charts would tell you.

Overclocking the CPU is likely going to make a GTX look less bad, but Nvidia really, really needs to drop prices.

 

[pauldh]

[citation][nom]jtt283[/nom]Very interesting article. VERY. I'll probably be accused of being an AMD fanboy now, but this article (and the previous Intel one) both appear to support the minimalist approach I prefer to take; CPU "I" may be a lot faster than CPU "A" but there is essentially no VISIBLE difference when gaming, and more money is better spent on a stronger GPU.I realize there's already a lot of data involved, but especially considering that older Intel CPUs were included, I was extremely disapointed to not see an older AMD CPU. Any of: 5400+, 6000, and/or 7750 would have been very useful to include at the low end, or even one of the newer Regor chips. Judging by these charts, any of the three tested AMD CPUs can game, but the article begins by saying (among other things) that it hopes to spare someone looking to upgrade an older system a potentially costly error. By excluding the older AMD CPUs, this article fails to provide that guidance for AMD upgraders. Furthermore, in the forums, I have seen a lot of budget builders advised to start with a Regor 250 (or 245), and this article does not show whether or not that's a good idea.I am also a little puzzled by the nVidia GPUs scaling with the CPU, especially in situations where the ATi cards do not.Finally, this puts the kibosh on any thoughts I had of getting a 1080p monitor, at least until I have upgraded my HD4850. Right now I don't have any newer games, but I'm going to pick up Dragon Age: Origins not later than early January and I want it to look great.[/citation]
I'm the type always looking for more...so would have enjoyed data for these other CPU's if possible. Judging by past reviews, we can see most of these AMD chips you mention fall below the E6300 at stock and won't OC as well. This one, for example shows the weakness in the Kuma. http://www.tomshardware.com/reviews/amd-cpu-overclock,2304.html

Athlon II's weren't an available option at the time, but I agree adding one to the series now would be useful.

Haven't played it myself, but my understanding is Dragon Age is well threaded and doesn't stress the GPU too hard. You'll need to do a little research, but likely a quad-core CPU and HD 4870/GTX 260 level graphics would go a long way (in resolution) in that title.

 

[pauldh]

[citation][nom]limwsv[/nom]I am starting a petition to have the article part on the ATI 58xx move up to be part 3. It'll be good to see how the CPU/GPU stack up with the newer graphic card as I think that will be what most people will be buying now, rather the soon-to-be obsolete 48xx.[/citation]
No petition needed, we aim to add the Radeon 5 series to parts 3-6.

[citation][nom]airgreek[/nom]When can we expect part 3 and part 4 Paul?[/citation]
Not for a while. I’m currently working on the next SBM while gathering hardware for parts 3&4. Got a Core i5 coming for part 3 but we are still waiting on graphics.

 

[dirtdiver]

^ Thanks, I was wondering this as well!

 

[Guest]

I need to see a slower processor like the Q6600, too! Thank you.

 

[Onus]

Yeah, I remember the Kuma being weak. I didn't consider it a worthwhile upgrade from a 4850e; your since-released CPU chart also shows it's not.

 

[nerrawg]

Great article! I love the graphs and they give a clear indication of whether performance is being CPU or GPU limited. I'm looking forward to the overclocking version of this article which I hope is also coming out soon. While on this topic - any possibility that we could get an article looking at cpu limitation of high end gfx card solutions like crossfire/SLI comparing the new Athlon II X3 and X4's to the Phenom II's when they are both overclocked? (at the highest reasonable 24/7 settings)

Reason is: If lack of L3 cache and slightly lower overclocks does not effect high-end multicard setups at high resolutions where they are neeeded, then the Athlon II X3 and X4 series are a real bargain. I would then see no reason for a value oriented gamer to consider the phenom II.

 

[Jawbreaker]

how come the HD 4870 x2 is so bad at playing GTA4?

 

[falchard]

I think another thing that is clear from these benchmarks is that in future hardware tests, there needs to be multiple benchmarking platforms. There is no standard that will make a particular piece of hardware shine. It should at the very least include an AMD and Intel benchmarking platform.

 

[Onus]

Falchard, this is why they use top of the line components for whatever is NOT under test, so any bottleneck will be in the tested item. HOWEVER (it looks like especially for nVidia cards), it may be useful to test with a lesser platform to see if some kind of "artificial bottleneck" will be imposed. It doesn't need a chart series (unless the results are really awful) so much as a comment that "You will see disproportionately lower results if you have a lesser platform for application/game 'X'."

 

[Guest]

Well, I liked the article, but I wish there was some data on truly mainstream graphics cards, like the GeForce 9600 series or the Radeon 4670~4770 series. I know they're somewhat old, but there are lots of people who can't afford to have Radeon 4850s and beyond. And they're still popular, as there are also people who can't afford to buy a new card every six months.
Otherwise, great article as always, keep up the good work!

 

[slmarker]

I like the article but how does someone without the money to spend on mutiple gpus and cpus tell if the cpu or gpu is slowing them down. The old question what to upgrade next the cpu or gpu.

 

[masterjaw]

Nice article as always. Kudos to THG. More interesting than the first one (because I use AMD). I was surprised that my X2 550 would be able to handle cards up to 4890 without that much bottleneck.

 

[dark41]

Another nice article. Can't wait to see how the overclocked E8400 compares with the pricier CPUs.

 

[BartG]

I know i`m pushing it, but how AWESOME would it be to have 1 site, in which you put all your comp specs and it will show the suggested performance figures. You can then change those parts online and see the difference... nice! These charts are nice...!

I`m realistic enough to know that would make for way to much benchmarking, but i like the thought...

 

[dirtdiver]

Another nice article. Can't wait to see how the overclocked E8400 compares with the pricier CPUs.

Likewise! I'm running my E8400 at 4.05ghz, I hope the review goes at least to 4ghz or even 4.4ghz!

 

[pauldh]

Likewise! I'm running my E8400 at 4.05ghz, I hope the review goes at least to 4ghz or even 4.4ghz!

Clocks for 3 & 4 testing haven't been determined yet. We'll push them, but won't shoot for a max air-cooled OC. I'm thinking just a bit conservative for the performance numbers in this series, although it will be hard to resist at least checking what each sample can do.

For testing the E8400, it will likely be whatever our (E0) sample does at 1.4V or less. Phenom II's likely 1.45V as beyond that we often see diminishing returns. Even those probably exceed the limits most readers would use for daily gaming. Tough call; No single voltage we use will please everyone, and testing multiple core speeds for each CPU is certainly not a possibility.

 

[dirtdiver]

For reference, my E0 chip runs 4.05ghz at 1.3v on air. 24/7 stable, tested with Everest and prime95 12+ hours.

I have no doubt at all that you know what you're doing, just wanted to give some advice from the gaming/overclocking community. My OC is not extreme, and I think would be a fairly good range to test with. Of course, I am biased because I dont have the means to benchmark my CPU as much as you guys, so I would like you to run all the tests so I can see!

 

[pauldh]

That sounds about right in line for my expectations with this chip, but we'll have to see. Just wanted to give a heads up (and accept feedback) that we won't be going too extreme on voltages as in some of the OC guides we've done.

The last E0 E8400 I had in the lab only ran 3.6 GHz at 1.25V VID, which was a big disappointment for that stepping. In contrast, had an E8500 stable at 3.95 GHz at 1.25V. (used here http://www.tomshardware.com/reviews/overclock-undervolt-efficiency,2134-5.html

 

[john boy]

Another great informative article which by no doubt, was very time consuming to compile. Anxiously looking foward to parts three & four series for the over clocking series and results for the 775 series for the E8400 and Q9550 Quad and the 775 boards and graphics cards chosen for these tests. I have a C2D and C2Quad and motherboards in two systems, so I look foward to any info an results and maybe some upgrades to these to run and stretch my systems to use a while longer, while saving up for a I-7 system along with raising a family which comes first.

I realize it takes time to compile components and to run these tests accordingly, for which I'm most appreciative, for I'm sure it has opened eyes for many of us, but you have us hooked now, waiting on the next series, like a child looking through the candy case wanting that next piece of candy. No biggie, just a little humor throwed in there, and patience is of virtue, so I look foward to the next article. Merry Christmas, and thanks for all you do!

 

[cnox]

Thank you for the article. I've been really studying the numbers here to make a determination about how to proceed with my next system build, and I guess my only disappointment is that we're not seeing enough of a diversity in the AMD offerings. In the Intel article, we saw Bloomfield, Yorkfield and Wolfdale, each sporting different architectures and capabilities. In the AMD article, we have a Phenom II, a Phenom II and a ... Phenom II. What this article is demonstrating is how MHZ and cores impact the balance of the GPU/CPU. That's good to know, however compared to all the things that were coverd in the Intel article, I feel that it would have been fair to show an Athalon II along side the Phenom IIs (who's main difference is no L3 cache). It would have been nice to see how L3 cache impacts balance, and also how multiple cores + L3 cache impacts the balance.

So, great article, I'm glad to finally see it, but I wish it just went a bit further and incorporated Athalon II data so we can see the data side-by-side. I know that the answer to this statement will be "Wait for parts 5 and 6", but that's a long wait, and also hard to compare graphs from separate articles.

On the note of the graphs: do you think you could keep the same graph dimensions for each game benchmark (at the specific resolution) so that if you have 2 tabs open side by side and the graph positioned at exactly the same spot on both tabs, you can flip between tabs and sorta see how the data compares. Some of the article 1 and article 2 graphs are identically scaled, but others are not and it's hard to compare.

Again, thanks a lot, and have a good holiday.

-C

 

[cdcsparky]

Is the Game Resolution, Details chart correct for Part 2? I only see 3 AMD CPUs being tested, shouldn't the chart reflect 3 CPUs, not 4?

 

[cnox]

Sparky, there's only 3 AMD cpus being tested. And they're all Phenom IIs. What is the 4th you were expecting?