7970 CF vs 7950 CF (A clock for clock comparison)

[Rusty0611]

Hi there,

Tonester and I have ran some benchmarks with 7970 CF vs 7950 CF at the same clock speed to try to quantify the difference in today's games while the two cards are at the same clock speed. As the 7970 GHz Edition and the 7950 have quite different clocks out the box it can distort the picture about the difference between the two cards in real terms.

Of course the 7970 is faster when both are clocked up but the common myth that a 7970 can clock further than a 7950 is quite false. The point of the comparison is to show you whether when you overclock your 7950/7970s is there any point in paying the extra for the 7970s. I won't offer any advice on this because to be honest, it's down to each person individually to have a look at the difference and determine what they want to do.

Game ran:

Hitman Absolution
Tomb Raider
Bioshock Infinite
Sleeping Dogs
DiRT Showdown
Metro 2033

Benchmarks ran:

3D Mark 11
Heaven Valley

My set up is in my signature as is Tonester's. My CPU was clocked to 4.4 GHz but the extra clock speed on Tonester's CPU is unlikely to affect the results that much other than in 3D Mark 11. All tests ran at 1920x1080 and 2560x1440 will follow.

You can take the results as either a simple clock for clock comparison or as a guide to how the cards perform in 'x' game. It's up to you .

7950 CF @ 1100/1500:

Spoiler

3D Mark 11:



Bioshock Infinite:



DiRT Showdown:



Heaven Valley:



Hitman Absolution:



Metro 2033:



Sleeping Dogs:



Tomb Raider:

7970 CF @ 1100/1500:

Spoiler

So to compare the clock for clock difference (average FPS) are as follows:

Bioshock Infinite: 3.8%
DiRT Showdown: 2.2%
Hitman Absolution: 2.7%
Metro 2033: -1.7% (potential testing anomaly as the benchmark is a bit sketchy)
Sleeping Dogs: 3.8%
Tomb Raider: 4%
3D Mark 11: 5.8%
Heaven Valley: 4.7%

 

[Rusty0611]

And here are the 2560x1440 results:

7950 CF @ 1100/1500:

Spoiler

Bioshock Infinite:



DiRT Showdown:



Hitman Absolution: (actual minimum FPS was 44 - benchmark bugged out)



Metro 2033:



Tomb Raider:

7970 CF @ 1100/1500:

Spoiler

Hitman Absolution: 1.9%
Tomb Raider: 5.6%
Bioshock Infinite: 3.8%
Sleeping Dogs: (Rusty unable to complete due to driver issue)
DiRT Showdown: 3.7%
Metro 2033: 9.4%

 

[Rusty0611]

Reserved just in case

 

[Rusty0611]

Have you tried doing them higher? 1200mhz or 1300mhz ?

It's not going to make any difference is it? I can do 1200 on both or 1250 on one and 1200 on the other but I wouldn't expect the gap to widen significantly, if at all, if they were both clocked up to 1200.

On 13.3 beta's and this version of MSI AB I'm having real problems keeping even 1100/1500 stable . Need to clean it all out but I only run 1075/1450 for 24/7 really.

 

[Rusty0611]

Ah, im not really a gfx card OC person it was really really easy to get my 7970's to 1150 on stock volts, and only a tiny tweak for 1250.

I see the difference is small, so im guessing the only real benefit to 7970 is a higher possible max OC?

A good 7950 will top out the same kind of points as a good 7970 will top out. The only difference will be the premium Lightning models really as they can go a bit further than normal.

 

[Rusty0611]

If your talking about gaming fair enough, benching is another ball game for the 7970 v 7950, although the 70 has higher vram clocks in it's tank with higher grade chips.

I don't think anybody serious at benching is really going to get a 7950 for that purpose though are they, Tommy .

The benchmark comparisons are there for reference.

 

[Rusty0611]

Nice one. Will update with my own results later on...

 

[Rusty0611]

2560x1440 results. Will update post #2 as well.

Had to skip Sleeping Dogs as on these drivers I can't get it to run at 2560x1440 when any overclock at all is applied so some issue there.

Spoiler

Bioshock Infinite:



DiRT Showdown:



Hitman Absolution: (actual minimum was 44 FPS but the benchmark bugged out upon completion)



Metro 2033:



Tomb Raider:

 

[Rusty0611]

Still not much at the end of the day though to notice any difference imho, perhaps the other games could show a bigger difference at high res multi monitor, but that's just a guess on my part.

To be honest I wouldn't draw any conclusions from the Metro benchmark. It's not very good . I could probably run it again and be faster than the 7970 results. It's that bad.

Can you try 1200/1800?

As you have said that the 7950 can clock just as much as the 7970.

Wrong debate for the thread. This is a clock for clock debate and unless you've got some theory for it I don't think there's any reason for there to be a non-linear relationship between performance and clock speed across the two cards.

And we can't anyway. One of Tone's 7970s doesn't do above 1150 . And mine are unstable on beta 3's above 1100 but stable at 1200 on 13.1 drivers.

Very interesting. On shaders/cores alone the 7970 should be circa 15% ahead, but the 7950 having the same number of ROPs and memory bandwidth must close that gap, those games musn't be shader-bottlenecked like they used to mainly be.

any chance of a crysis 3 bench?

Does it have a benchmark tool? No reason why we couldn't if it has. If it doesn't there's too many inconsistencies between two people running a set part of the level in real time.

 

[Rusty0611]

None of the results are 100% conclusive, just a rough indication of what to expect between the 2 cards which I assumed the point of the thread.

There isn't a BM tool for C3 yet, but there is this part:

The point is Tommy, if nothing else changes other than the resolution, then if a card which is practically the same except for some CUs and shaders goes from '1.9%' slower to '9.4%' faster there's something wrong with the benchmark.

The other games are what you would expect to happen. I think you're looking at a 5% clock for clock difference at best not towards the upper echelons of 10%.

In benchmark terms, 5% is considerable but in gaming terms you have to look at what kind of FPS 5% is translating to. I'm keeping out of the "worth it?" debate principally because it's entirely subjective and I'm not about to force upon people what they should/shouldn't do .

(not aimed at you)

 

[Rusty0611]

Rusty,

Any chance you could compare mining bitcoins?

Nope, sorry. Not something I'm too interested in myself but you may have luck with others.

 

[Rusty0611]

Fantastic info imginy, same system testing showing that in a few instances, there is a larger/smaller difference with resolution and possible driver support for individual titles plays a part too.

Unfortunately:

[02/16/2012 02:30 AM]

 

[Rusty0611]

I understand that rusty, but it ties in with driver support too that I mentioned.

AMD I doubt will increase performance all across the board in every single title, some will have been left as they were as they were of little importance in the grand scheme of things and just concentrated on what they deemed of greater importance for improvement.

We'll never have a conclusive answer either way unless someone has the inclination to test all the titles in the article in the same system with both gpu's, so it's left to the individual to make their own mind up and not for any of us to preach the conclusive outcome.

Well you say you know that but some GPUs gained more than others from the recent driver improvements. 7870 and 7950 being prime examples. I haven't said anything yet about you consistently stating it was up to 10% but if that is what this is about then don't. The results are in and the real terms difference is there as you say "for all to see" . The chance for fluctuations and/or errors is greatly reduced with a fixed benchmark too.

I have to admit I was surprised that there wasn't at least a couple of 7-8% differences but that surprise was more from me just taking the spec differences in isolation.

I'd say Tonester's and my results are fairly conclusive. If you've got something that you'd like to see to increase the comprehensiveness of the comparison then be my guest and I'll see if we can incorporate. The point is the article linked is only valid for that point in time and unless anyone is using 12.3 drivers still (or whatever the drivers were at that time ) then things can change. I must admit, I didn't really read the review once I saw the date because Tonester's and my results supersede those.

 

[Rusty0611]

Tommy, this is not the thread for you to turn it into a multi page twist and turn rant. What I actually said is the FPS figures are for reference but benching for performance was not the purpose of the thread... it was to distinguish the difference hence the results themselves in spoiler tags. Our results aren't the be all and end all as in any differences outside of what we got are invalid. But that's not what I'm saying anyway. What I'm saying is a 14 month old comparison isn't much use in today's terms.

Well you have consistently stated that it can be up to 10% but we didn't get bear to that in any of the testing. If you believe that higher can be obtained then as I said let us know what game and can have a look into it.

Regarding the drivers there was a few summaries posted regarding driver gains on different cards since release. They showed the 7950/7870 gaining a larger amount. Which would be irrelevant were you not going on about a 14 month old review. They show the difference at that point in time, nothing more; nothing less. The same as Tonester's and mine show the differences at this point in time. It's not that difficult to understand is it?

 

[Rusty0611]

I have a feeling that we'll be benching every game known to man until we find one that does provide a 10% difference clock for clock!!

Only joking Tommy

 

[Rusty0611]

The difference in my results holds true to inconsistencies, again no mention of that data, no mention of per game driver improvements from yourself either, just your insistance that driver improvements on a whole deems an article void which it's far from fact due to reasons I pointed out above.

lol you said my results were invalid on something else just because they were from Autumn 2012 or whevever, now a 14 month comparison holds true today because Tommybhoy says so. Logic failure. I didn't take into account your results because what do you want me to do with them? You were 40% slower than Matt's 7970 at the same clock speed in Tomb Raider which can't purely be down to PCI 3.0 8x.

The multi page discussion is twofold, I can't help it if your in disagreement with my thoughts and respond to them, it takes two to tango buddy, no point laying the blame firmly at my door and present yourself as the innocent party.

Well when you're trying to nit pick and basically save face over the 5/10% debate (for reasons I won't pretend to understand) then yes, it is just you. These results are just... as they are for today's games on today's drivers. The difference may have been more/less back in February 2012 (lmao) but that is just irrelevant now.

The point is you're basing off an old article which doesn't represent today's performance due to driver changes. I'm not saying performance boosts are one way (you don't really think that do you?), I'm just saying it isn't representative of today's differences. You can't pick and choose old articles as and when you please and discount "old" analysis the next.

 

[Rusty0611]

We'll never have a truly accurate way of testing until it gets tested on the same pc regardless of what anyone else says. The difference in my opinion can be as little as 1% or as high as 8 or 9% depending on the title but for arguments sake and taking benchmarks into account you can say 1-10% depending on title, drivers and god knows what else. There will always be examples where the difference is lower or higher than 5%.

Oh yeah we can't be accurate to the dot (don't think anyone or me has actually said this) but what we can do is do all we can to minimise chances for differences. I don't think anybody is even saying, "5% end". What is being said is that the 5% figure looks to be more accurate than the 10% figure based on actual evidence.

What I will say is that I haven't seen an actual gap of anything close to 10% (discounting Metro as the benchmark is dodgy) but that I still have an open mind if anybody has a game to benchmark to test. Not to say it wasn't 10% back in the day (or even 15-20%) but that's not relevant in today's terms in the same way we don't use 14 month old comparisons of the 7950/7970 against the 670/680.

Tommy's argument fails because it assumes that for that review to still hold true today (which is what he's saying), driver improvements across the 7000 series of cards have produced universal increases in performance across the range which we know is not true.

As we're talking such small differences the problem with a non standard benchmark (e.g. BF3) is that it can open up a difference that isn't there or close up a difference which is actually there.

 

[Rusty0611]

that can be arranged lol. If Rusty wants to I'll trade for the sake of testing two 7970s for two 7950s. Hopefully different drivers will allow my second card to clock better in cf.

I would... if my motherboard didn't require a BIOS update and restoring of settings everytime I plug/unplug something from a PCI-E slot .

 

[Rusty0611]

Well I've yet to see one more than 1% higher than 5%!

A good 7950 goes the same kind of distance as a good 7970. The key word here is 'good' because a stinking 7970 is just as bad as a stinking 7950 (Tonester's second card). My good 7950 which I sent back due to the MSI problem did 1300/1750 but was limited by temperatures. If I took the cooler off and stuck an Accelero on it I'm sure I'd be around your clock speeds. My current Gigabyte one does 1275/1700 on 13.1 drivers.... At the end of the day, the silicon lottery overrides any debate as to which will overclock further and while connected, it is a touch tangential to a clock for clock comparison.

A 7950 PCB does indeed have worse memory chips, but we're talking the difference between ~1600 MHz and ~1800 MHz which isn't much in performance terms in games. Highlighted games because I've seen myself that a memory OC scales quite nicely in benchmarks.

thew 7970 will most of the time overclock much more.. Increasing the Gap

This is a clock for clock comparison thread but you haven't provided any proof that that is the case anyway.

Can give you the UD5H? Lol.
I'll be on X79 hopefully by the end of this week so 1155 spare till I sell (sold the CPU)

I'll still have to sit through the BIOS recovery screen, plug in the hard drive. update BIOS, restore settings as soon as I unplug a 7950 . I've done it so many times from when I was faffing around with the 680s and 7950s together that the pain of it is too much to bear.

 

[Rusty0611]

Yes you will get some 7950's that can match, the majority won't be able to for the reasons i've already mentioned. You will get more good 7970's than 7950's. A cooler running temp won't help memory speeds either, only core.

People say this but without any evidence to hand. I know there isn't anything. It's become common knowledge without ever being proven. A selective forum sample isn't evidence either.

I've had 4 7950s - 2 were temperature limited at 1250 & 1300/1750 by the dodgy MSI cooler, 1 does 1275/1700, the other does 1200/1650. My point is not that it means all 7950s will fall within this range, it's pointing out it's just impossible to make general statements like that without a large sample of evidence.

p.s. that was just the max clock speeds I ran.... zero memory artifacts but I didn't go further because I didn't have time to test for stability as they both went back to OcUK. I saw nothing to suggest that they wouldn't have gone another 50-100 MHz. Also let's be honest, your 7970 is far from the norm as well .

The point is all this talk of "majority" and "most" is without any kind of statistical analysis. I'm not expecting you to go away and do it p) I'm basically saying it's not a proven fact and there's no evidence to suggest that is the case. What I agree on is that generally the 7970 memory will do another 200 MHz or so compared to vanilla 7950 models. What I disagree on is that this makes enough difference in performance to realistically consider a 7970 over a 7950.