Which graphics card for my system - 770 v 7970

[oweneades]

I'm surprised at how close the 7970 is to the 780 in an nVidia sponsored game too.

Yes, the 7970 really surpised me TBH. Overclocked It can hold onto a 780 stock in virtually all titles I tested. The 780 once overclocked can draw quite a lead in terms of playability (not specifically massively higher Avg FPS) in games like Crysis 3 (50FPS vs 39FPS average - doesn't sound like a lot but the 780 feels a lot smoother).

OVerall clock for clock the results are pretty much in line with a card that has around 20-25% more grunt (spoiler - it is 20-25% faster like for like (up to ~30% in the odd scenario)).

Good results and thanks for posting owen always interesting to see.

I can post my entire 780 Surround review in a seperate thread if you like? It is up on a couple other sites but I don't mind posting here. I will wanr you though, its a long read.

 

[LtMatt]

Yes. However, the issue at the moment is that the 770's memory bandwidth cannot reach 7970's memory bandwidth level, even if the memory was overclocked from 1753 (7000MHz) to 2000MHz (8000MHz).

GTX770 (memory at stock clock):
256/8 *1753 *4 = 224,384 (224GB/sec)

GTX770 (memory overclocked to 2000MHz):
256/8 * 2000 *4 = 256,000 (256GB/sec)

HD7970 (memory at stock clock)
384/8 * 1375 *4 = 264,000 (264GB/sec)

HD7970 GHz Edition (memory at stock clock)
384/8 * 1500 *4 = 288,000 (288GB/sec)

HD7970 (memory overclocked to 1700MHz)
384/8 * 1700 *4 = 326,400 (326GB/sec)


There's no going around the fact that the 256-bit is clearly holding back the memory bandwidth.

My 7970 (RIP ) could do 1823mhz on the memory without breaking sweat, giving it 350gb of bandwidth.

 

[spoffle]

Yes, the 7970 really surpised me TBH. Overclocked It can hold onto a 780 stock in virtually all titles I tested. The 780 once overclocked can draw quite a lead in terms of playability (not specifically massively higher Avg FPS) in games like Crysis 3 (50FPS vs 39FPS average - doesn't sound like a lot but the 780 feels a lot smoother).

OVerall clock for clock the results are pretty much in line with a card that has around 20-25% more grunt (spoiler - it is 20-25% faster like for like (up to ~30% in the odd scenario)).

Yeah I can imagine, the bit more GPU power will of course help!

These results are making me look forward to what AMD brings out next though, as it's going to be very interesting if they bring a 25-30% increase in GPU performance on the same process. It'll shake things up a fair amount.

I can post my entire 780 Surround review in a seperate thread if you like? It is up on a couple other sites but I don't mind posting here. I will wanr you though, its a long read.

That sounds fun!

 

[oweneades]

Yeah I can imagine, the bit more GPU power will of course help!

These results are making me look forward to what AMD brings out next though, as it's going to be very interesting if they bring a 25-30% increase in GPU performance on the same process. It'll shake things up a fair amount.



That sounds fun!

I have already got the go ahead from the Wife to put a HD9970 (or whatever it is called) into the test machine to see what it can do against a 780 in Surround/Eyefinity.

I will create a thread shortly with the review. It was fun to bench to begin with but due to some game problems I ended up redoing the same benches about 10 times in a row.... I can only watch the intro to Operation Swordbreaker so many times.

 

[Kaapstad]

Your example of the GTX580 shows you don't understand this topic.

So run this by me again

You are dead against people buying 2gb cards for use @1080p - 2 million pixels

But you use 3gb cards on your setup which has 11 million pixels

I think you should stop talking down to people as I really think you have lost the plot.

 

[spoffle]

So run this by me again

You are dead against people buying 2gb cards for use @1080p - 2 million pixels

But you use 3gb cards on your setup which has 11 million pixels

I think you should stop talking down to people as I really think you have lost the plot.

You are the issue here not me, because you are consistently misrepresenting what I am saying to make it look like I am talking rubbish.

How many times do I have to say that the importance is upon the memory bandwidth over RAM before it sinks in?

Additionally, as I've already pointed out, the increase in pixel count isn't a linear decrease in performance.

 

[dmferrett]

A GTX770 is ok at 1080p now but we dont know about the future!!!

Vram is important for 4kplus or multiscreen. The extra texture's eat up the Vram.

But having extra Vram alone if the memory bus is a bottleneck (256 or less) is pointless.

Likewise having a gay gpu processing information used for a 384bit's memory bus with 3gb or more vram is gonna ruin performance.

The Graphics card need the whole speel really 384bits, powerful gpu with loads of streams processes, 3gb plus of ram etc etc....

Its all been mentioned in one way or another previously......

 

[Suarez7]

A GTX770 is ok at 1080p now but we dont know about the future!!!

Vram is important for 4kplus or multiscreen. The extra texture's eat up the Vram.

But having extra Vram alone if the memory bus is a bottleneck (256 or less) is pointless.

Likewise having a gay gpu processing information used for a 384bit's memory bus with 3gb or more vram is gonna ruin performance.

The Graphics card need the whole speel really 384bits, powerful gpu with loads of streams processes, 3gb plus of ram etc etc....

Its all been mentioned in one way or another previously......

lol wut?

 

[dmferrett]

Yeah a bit of colourful, humourous, expressive english........

 

[Kaapstad]

lol wut?

Probably uses a pink bus.

 

[spoffle]

Bus wide or bus home

 

[Suarez7]

I'm still trying to figure out if he meant the card was happy or bats for the other team

 

[Marine-RX179]

You are the issue here not me, because you are consistently misrepresenting what I am saying to make it look like I am talking rubbish.

How many times do I have to say that the importance is upon the memory bandwidth over RAM before it sinks in?

Additionally, as I've already pointed out, the increase in pixel count isn't a linear decrease in performance.

Yes...it's unbelievable. We are all here discussing memory bus-size and memory bandwidth for a while already, and Kaap is still running around in circle chasing his own tail trying to force the direction of the discussion back to memory capacity...

 

[Ian Evey]

Like i have said, a 670 is faster than a 7950 out of the box. If people dont want to overclock it they wont. The real question is, Why did amd release the card with such a low core clock in the first place. Sales lost there.

So if the 670 is faster out of the box it will sell well no matter how many times people tell them a 7950 is more future proof, which is all speculation anyway.

 

[Kaapstad]

Yes...it's unbelievable. We are all here discussing memory bus-size and memory bandwidth for a while already, and Kaap is still running around in circle chasing his own tail trying to force the direction of the discussion back to memory capacity...

I understand memory bandwidth, but I am not sure some of the people making comments do.

Just out of interest could someone (perhaps you) post how much bandwidth actually gets used on a number of popular games using a number of popular graphics cards at say 1080p.

I don't know the answer to this myself, what I do know is that all the games will not use all the available bandwidth on all the cards. The answers for each game and each card will be different. Also to quote simple vram usage figures does not answer the question.

 

[tommybhoy]

Memory Bandwidth: This is one of the single, most important aspects of graphics processors. Memory bandwidth determines your card's ability to utilize its onboard video RAM efficiently when under stress. Think of it like the lanes on a highway: if you have a highway with 3 lanes that is perpetually congested, then you add 3 more lanes to it over the weekend, you'll see a significant decrease in traffic (if not outright elimination of congestion). The same is true for GPUs: having tons of GDDR5+ memory won't do any good if the pipe is too small to use it in time.

Memory bandwidth is calculated by memory type (i.e., GDDR5, GDDR4, etc.), the memory clock, and the actual memory width. Calculate the maximum memory bandwidth by multiplying the memory clock by the memory width and the transfers-per-clock of the memory type.

Memory Clock: Quite simply, this is the speed of the video card's onboard memory. As above, the memory clock helps calculate memory bandwidth; a higher memory bandwidth equates better performance for anti-aliasing and other memory-intensive tasks.

Memory Interface: This is the memory's actual bus width, typically in the form of 128-bit, 256-bit, or 384-bit. The memory interface is used to calculate total bandwidth. A bigger interface means a bigger pipe. A smaller interface can be compensated for by faster memory clock speeds or different types of memory.

http://www.gamersnexus.net/guides/717-gpu-dictionary-understanding-gpu-video-card-specs



Hitman "Run For Your Life." level:

We also dropped the resolution to 1920x1080 to see how the TITAN performs on 1080p displays. We really wanted to push the video cards, so we opted to run the game at the highest 8X MSAA level at 1080p to see if it would be playable. The only two video cards that could even do this were the 6GB TITAN and 3GB 7970 GE, the 2GB GTX 680 was not capable of running this setting due to its small memory footprint.

With two TITAN's in SLI we were impressed to be able to enabled 4X MSAA at 5760x1200. In this game, this is a first to find 4X MSAA playable at this large resolution on 3-displays. No other video card combination has given us this ability, until now. As the resolution increases, the sensitivity to VRAM also increases in this game and demands very large framebuffers to run AA. The most we've been able to do in the past is 2X MSAA, thanks to the Radeon HD 7970 GHz Edition's 3GB of VRAM on board, and even that wasn't completely smooth.

With the 6GB of VRAM on the TITAN we had no problem running at 2X MSAA or 4X MSAA. The graph does show though, that even at 4X MSAA, it did drop more than a few times in the lower range of framerates, but the game still felt perfectly smooth and playable to us. With GTX 680 SLI we weren't able to use any MSAA at all, only FXAA, so TITAN has a huge advantage at 4X MSAA, or even 2X MSAA.

In the apples-to-apples test we have 2X MSAA enabled, and you can see that TITAN is literally over 100% faster. The reason is because GTX 680 SLI is severely bottlenecked at 2X MSAA due to its limited 2GB framebuffer, it isn't achieving the highest possible framerates it could, due to that limit. Therefore, TITAN is able to outshine it and achieve its full potential of performance in this game, which is amazing.

http://www.hardocp.com/article/2013/02/21/nvidia_geforce_gtx_titan_video_card_review/4#.UfFlo42QnDw

 

[bfar]

At 1080p I believe that future titles will top out 7970/770 gpu power before more than 2gb is really a necessity. At that res 2gb has 2 years of life yet.

 

[Krisali]

At 1080p I believe that future titles will top out 7970/770 gpu power before more than 2gb is really a necessity. At that res 2gb has 2 years of life yet.

I think so too.

 

[Ian Evey]

So from tommybhoy's link I came to the conclusion that if i use 5760x1200 with 2xmsaa i should not use my 670, which i wont be doing any time soon.

So we can all recommend AMD for people using 3 displays or a higher res than 1920x1200.

Yet at 1080 they are all still quite evenly matched even though the 670 and 680 and 760 and 770 are all using the so called limiting 256bit bus.

hmmm I might buy a Titan now. Wait. What am i thinking, A 780 would be a smarter buy.

 

[tommybhoy]

Again, reading only the bits you want and skipping the bits you don't like, what 2 cards out of 3 tested aren't limited@1080p with Max AA?