How to compare different gpu's?

[Gu____Vlad]

Can I compare gpu's by simply looking at memory bandwidth?
Higher memory bandwidth much better GPU and more fps?
Or we should look at something else?

 

[kyzarvs]

Use the Toms hierarchy chart for relative GPU performance?

 

[kewlbootz]

There's a LOT to take into account if you're just comparing GPUs by spec.

What I recommend is simply taking a look at the hieracrchy chart like kyzarvs says above, and looking up benchmarks. Assuming you're gaming, make sure you look at more than just one site's benches, though, because they can vary pretty substantially when the game doesn't have a built in benchmarking utility.

Do you have any GPUs in particular you're looking at? What will you be using them for? What resolution? What's your budget?

 

[rgd1101]

Need to look at everything. memory bandwidth is just one of many.

 

[Demonic Heart]

You should look at memory clock, memory config, sometimes cuda cores (if we talking nvidia) and above all else memory bandwidth...

i myself determine the power of graphics card by its memory bandwidth. more is better

 

[Kryptkeeper]

well no. the 770 970 and 980 all have the same memory bandwidth and bus rate but perform differently. the only and best way is to look at actual real world benchmarks

 

[kewlbootz]

To clarify, I definitely would NOT just look at memory bandwidth, GFLOPS, memory/core clocks, etc.

That information is near useless when comparing real-world performance. Of course specs affect performance, but they don't tell you anything by themselves. Benchmarks, and comparing different benchmarks, is the only real way to compare GPUs.

 

[Demonic Heart]

well no. the 770 970 and 980 all have the same memory bandwidth and bus rate but perform differently. the only and best way is to look at actual real world benchmarks

you are right but as i said before,the memory bandwidth alone does not determine the power of the graphics card..

Texture fill rate and memory config is the difference between the 970 and 980

 

[Gallarian]

Can I compare gpu's by simply looking at memory bandwidth?
Higher memory bandwidth much better GPU and more fps?
Or we should look at something else?

Definitely not. There are many, many other things at play other than memory bandwidth.

Simple way to rank a graphics card's performance at a glance (i.e. not having to having to look at a long chart) is its name.

With Nvidia, they currently have 'GTX' followed by three numbers. For example, a 'GTX 980'. The first number of the three is the generation, in this example '9'. The next two represent the power of the card within that generation, where '80' is usually the highest.

So a Nvidia GTX 980 is very powerful graphics card, with the 970 being just below it, and the 960 next after that. As a rules of thumb, a later generation card is usually as fast as the card that was above it last generation. So for example, a GTX 670 was roughly equal to a GTX 580 from the previous year.

Sometimes Nvidia ads 'TI' on the end of a cards number to denote that its faster, but not by a huge amount. For example, a 950ti is about mid way between a 950 and a 960.

Obviously this isn't a perfect way, but at a glance its roughly right.

As for AMD, they have a very similar system but its a little more complicated. For for example an 'R9 390' is better than an 'R9 380', with a 'R9 380x' being roughly half way between the two.

The hard bit is comparing between the two companies, as they overlap and don't match up at all, so you'll need a hierarchy chart for that.

 

[Gallarian]

Easiest way for you to get a clear and accurate comparison between two or more graphics cards is this; post what their names are in a comment on this thread and we'll rank them in order and tell you why

 

[blazorthon]

Comparing by specs alone requires a lot of research. It is easier to just look at benchmarks.
The problem with specs:

You can't directly compare core counts because different cores perform differently and even the same cores can perform differently under certain circumstances. For example, the individual CUDA cores in Kepler and Maxwell are almost identical, but Maxwell organized them differently and voila, now they seem to perform radically faster at the same clock frequency because they're fed data and things differently.

You can't compare memory bandwidth directly because of differing levels of compression and caching greatly affect how much the GPU gets out of what memory bandwidth it has. Maxwell has very large caches with very high memory compression ratios and that's part of why it can get away with less memory bandwidth than inferior-performing Kepler cards. These don't solve its lacking memory bandwidth completely, but greatly alleviate it.

There are also many other factors such as the ROPs, tessellation, and more. Some of them have theoretical performance specs that can be very different from real-world performance too, so you need certain benchmarks anyway.

Basically, most of the specs are only relevant if you want to learn about this stuff. Comparing them for what to buy is much easier to do with simpler gaming benchmarks.

 

[blazorthon]

Can I compare gpu's by simply looking at memory bandwidth?
Higher memory bandwidth much better GPU and more fps?
Or we should look at something else?

Definitely not. There are many, many other things at play other than memory bandwidth.

Simple way to rank a graphics card's performance at a glance (i.e. not having to having to look at a long chart) is its name.

With Nvidia, they currently have 'GTX' followed by three numbers. For example, a 'GTX 980'. The first number of the three is the generation, in this example '9'. The next two represent the power of the card within that generation, where '80' is usually the highest.

So a Nvidia GTX 980 is very powerful graphics card, with the 970 being just below it, and the 960 next after that. As a rules of thumb, a later generation card is usually as fast as the card that was above it last generation. So for example, a GTX 670 was roughly equal to a GTX 580 from the previous year.

Sometimes Nvidia ads 'TI' on the end of a cards number to denote that its faster, but not by a huge amount. For example, a 950ti is about mid way between a 950 and a 960.

Obviously this isn't a perfect way, but at a glance its roughly right.

As for AMD, they have a very similar system but its a little more complicated. For for example an 'R9 390' is better than an 'R9 380', with a 'R9 380x' being roughly half way between the two.

The hard bit is comparing between the two companies, as they overlap and don't match up at all, so you'll need a hierarchy chart for that.

The naming systems are both useless for comparing performance with anything more than this card is anywhere from slightly faster (but not enough to even see a difference) than the previous card to almost twice as fast. For one, the GTX 960 isn't just a little bit slower than the GTX 970. The 970 is MUCH faster. On the other hand, the difference between the 980 and the 970 is about half of the difference between the 960 and the 970. The 950 is almost exactly as fast as the 960 in most situations.

Also, that rule of thumb for a card being faster than the one above it in the previous generation- definitely not reliable. The 670 was not simply almost equal to the 580, the 670 is almost equal to the 680 and both are considerably faster than the 580. The 570 was equal to the 480, but that was partially because they have almost the same exact GPU whereas the 670 has a very different GPU from the 580. The same is true for many of the newer models.

Ti can be a huge amount, a small amount, or not much at all faster than its similarly named counterpart. For example, the 980 Ti is much faster than the 980 whereas the 780 and the 780 Ti are almost equal. Of course, that's only because the 980 Ti uses a much larger chip whereas the 780 Ti and 780 have the same basic chip, but you wouldn't know that just by looking at the model numbers.

AMD's system is no more accurate and has the same flaws. For example, the 380X is slightly faster than the 380, but nowhere near the 390. It certainly isn't anywhere close to being halfway between the two. Also, AMD sometimes uses a 5 as the last digit instead of adding an X. No, you don't know even know which is faster without benchmarks.

I realize you added the disclaimer that this method isn't perfect, but it isn't even roughly right. It's pretty much useless. Benchmarks are the only accurate way of discerning performance and even they are a matter of debate.

 

[Gallarian]

I realize you added the disclaimer that this method isn't perfect, but it isn't even roughly right. It's pretty much useless. Benchmarks are the only accurate way of discerning performance and even they are a matter of debate.

So youre saying it wouldn't be accurate to say a GTX 980 is better than a GTX 970? or a 570 is better than a 560? You and I both know you dont need benchmarks for that.

I get your point, but as the OP is simply looking for an easy to way to tell if a card is more powerful than another one (not by how much), so your first go to should be the designation. If you want to know by HOW MUCH, then jump on over to some benchmarks.

The naming systems are useless if you want to compare between brands, and it becomes more difficult if you try and compare between generations, but it is definitely not useless if you want to know which card is more powerful than another in the same generation at at first glance.

 

[rgd1101]

Thread closed. We answer the OP question