Websites to objectively compare components?

[fuzzybabybunny]

So what websites do you guys use to objectively compare components when you're looking to build a system? I know benchmarks exist, but is there anywhere you go to look at benchmarks for *just* component XYZ?

ie.
nVidia GTX 123whatever gets a score of 1002
nVidia GTX 000whatever from 2 years ago gets a score of 332

CPU 123 gets a score of 3000
CPU 002 gets a score of 244

Then you combine CPU 002 with nVidia 123 and you get a combined score of 1246. Basically a completely objective way of measuring the performance of parts, all in one website.

 

[lehtv]

www.anandtech.com/bench

 

[fuzzybabybunny]

www.anandtech.com/bench

Thanks, but it's not entirely complete (missing every mobile part for example). And does AT have its own standards? ie is the score it gives standardized across all websites, or only for AT?

 

[A5]

There isn't really any way to compare mobile parts because you can have two things with the same name that have completely different performance based on the memory.

There isn't a site that does exactly what you want, but Tom's hierarchy chart is a good place to start: http://www.tomshardware.com/reviews/gaming-graphics-card-review,3107-7.html

From there, you can try to find a review that compares performance in games you actually play.

 

[Ken g6]

Then, there's FutureMark. They mix desktop and mobile parts, at least.

 

[dave_the_nerd]

Then you combine CPU 002 with nVidia 123 and you get a combined score of 1246. Basically a completely objective way of measuring the performance of parts, all in one website.

You can't really combine scores like that. If you're GPU limited, then upgrading your CPU won't help, even if it would yield a higher "score" in your model.

Mostly it's workload sensitive. You just have to know what you're using the computer for, how that works, and which components are most critical.

Like digital photography? CPU > All. Single threaded performance is still critical. Monitor color accuracy is way more important here than anywhere else.

Video Editing? Now HD Sequential I/O is important too. Maybe even more important than CPU performance, if you're using a GPU to encode/render. Multi-threaded performance is more useful than single-threaded. Fuck monitor color accuracy, you're just toting a projector to the family reuinion and throwing it up on a sheet after beer & brats.

Running a NAS box for your home? CPU is even less important. HD I/O is going to be network limited to single platter drive speeds, no matter how many SSDs you stripe together. Network design is more important. Aggregate links if possible. No discreet GPU needed. The monitor is a 17" CRT you found in the basement when you moved in.

Etc.

Good benchmark sites (like anandtech... yeah, I'm sucking up) will simulate a variety of workloads on the hardware they're reviewing so you have an idea of what to expect and if the component in question will help you or limit you. The Bitcoin miners don't care if it will play Crysis at 60fps, and gamers with big wallets and a need for speed will crossfire two 7970s together because it's fast and supports Eyefinity, not caring that they'd get more F@H PPD with a GTX 680.

 

[mfenn]

You can't really combine scores like that. If you're GPU limited, then upgrading your CPU won't help, even if it would yield a higher "score" in your model.

Mostly it's workload sensitive. You just have to know what you're using the computer for, how that works, and which components are most critical.

Like digital photography? CPU > All. Single threaded performance is still critical. Monitor color accuracy is way more important here than anywhere else.

Video Editing? Now HD Sequential I/O is important too. Maybe even more important than CPU performance, if you're using a GPU to encode/render. Multi-threaded performance is more useful than single-threaded. Fuck monitor color accuracy, you're just toting a projector to the family reuinion and throwing it up on a sheet after beer & brats.

Running a NAS box for your home? CPU is even less important. HD I/O is going to be network limited to single platter drive speeds, no matter how many SSDs you stripe together. Network design is more important. Aggregate links if possible. No discreet GPU needed. The monitor is a 17" CRT you found in the basement when you moved in.

Etc.

Good benchmark sites (like anandtech... yeah, I'm sucking up) will simulate a variety of workloads on the hardware they're reviewing so you have an idea of what to expect and if the component in question will help you or limit you. The Bitcoin miners don't care if it will play Crysis at 60fps, and gamers with big wallets and a need for speed will crossfire two 7970s together because it's fast and supports Eyefinity, not caring that they'd get more F@H PPD with a GTX 680.

:thumbsup::thumbsup: Exactly. A benchmark measures exactly one thing: how well a system runs that benchmark. There are a million different ways to use a computer, and unless you are looking to run benchmarks and only benchmarks all day long, none of those ways exactly match with a benchmark result. It takes some human intelligence to find benchmarks that most closely match your particular workload.

Furthermore, you cannot simply boil down an entire computer's performance to a single number. Products like PCMark certainly attempt to give you such a number, but they only really tell you how the systems perform in the workload that PCMark (or whatever) presents. Whether or not this is useful to you depends on how closely your intended workload matches what PCMark runs and whether or not you assign the same weights to the various sub-scores as it does.