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

[Rusty0611]

I have used 5 7950's and 2 7970's and the 50's don't come close to the vram clocks the 70 can achieve for the reasons outlined above-they have better chips on the vram in general, of course it's still down to the lottery but the 70's vram carries on when the 50's hits a brick wall and it can make the difference in high AA scenarios like BF3.

I don't believe the extra amount the 7970s clock to on the memory really account for much game performance.

Don't take my words and use them out of context to suit a different discussion Rusty, poor very poor indeed:

If you say so. But that IS what you said. Keep on twisting.

The only term implied talking about your results being outdated were in relation to a vram/bus discussion.

Since this discussion is entirely different-performance difference across the board, older titles are valid, hence the point about performance gains through newer drivers possibly being static in non important titles.

So don't mince my words for effect/leverage, call it what you like.

As I said, when it suits....

There's no nitpicking/face saving(obviously your looking along those lines though) but stating that driver improvements are on a game by game basis unless AMD release driver notes stating along the lines of-10-20% driver improvement across the whole board of titles supported in their drivers.

"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."

There was an issue somewhere with Matt's/my comparison that's why it was removed quickly, my CrossFire results don't have any x8 problems for comparison.

But in future please don't misquote me as it tends to put users down the wrong track as what happened in another thread that the quote was taken out of context again LEAVING FALSE ACCUSATIONS aimed at myself.

False accusations?!

I'm not going to tarnish the thread anymore because it's been turned south enough, yet again by the nitpicking, doubling back, twisting, turning..... so as the OP, I ask unless you've got anything to contribute to the topic at hand (clock vs clock 7950/7970 comparisons) please exit the thread. Nobody reading this has got any interest in it and to be honest, neither have I. What's been said has been said but anything more and I'll RTM or ask a mod to remove it to keep the thread on track. Thanks for your understanding.

You have to consider that the 7970's have a better pcb which aids overclocking. 7970's are able to deliver more power as most 7950's only have 2x6 pin power connections where as all 7970 have 8+6 pin. The 7970 also has extra shaders, compute units and the higher clocking memory. The 7970 will also have a higher TDP limit as well which will definitely be a factor when large overclocks are applied. You trigger the OCP limit, your card shuts off and is dead until you restart. It will take some effort to trigger the limt but it is possible ive done it myself and if my limit was 50-75 Watt less than it is now id trigger it a lot easier. How much of a difference this all makes is indeed a great debate and one neither of us can really answer. But i think these are things to be considered in light of saying get a 7950 as its only a few percent slower than a 7970. Then again people probably don't think or get anal about all those points like me but hey im thorough.

All well and good Matt, but how does that relate to what a card can really achieve at maximum in reality. I get the theory but silicon lottery still overrides all of those points to an extent.

Its probably also safe to assume that in percentage terms the 7970 will clock higher on the core as its stock clocks are higher.

Nope. I know what you're getting at but in reality I don't think it's true nor can I think of any technical reason as to why that would be the case.

 

[Rusty0611]

I'd rather buy a silicone lottery ticket from the 7970 pool than the 7950 pool. Wouldn't you?

It still comes down to a lottery but lets just say your odds go up a bit for the reasons mentioned. Otherwise we'd see more 7950's above 7970's in the benchmark threads/game benchmarks. As it stands there are very few.

Clock for clock the difference is not that big. But these other factors can add to that difference and that is what never gets mentioned or talked about.

The 7950 pool because I've seen no evidence to say that you're more likely to get a good 7970 over a good 7950 and we can see the clock for clock difference sits on or below 5% and they're cheaper. If you're comparing custom cooled vs custom cooled quite a difference as well in price. Not to say my way is the only way of looking at it but that is my POV.

My point is that all these factors add up to more performance which in my opinion is another point to be considered.

No they add up for the potential for more performance which is not the same as what you're saying. The things you've said don't automatically mean more performance.

I don't know any technical reason either. I've a good mind to ask Thracks.

Lets think about this sensibly though if 7950's had better cores than 7970's why are they in 7950's and not 7970's which are AMD's current flag ship cards?

I think they are probably identical but the reasons ive mentioned means that the 7970 core will on average go further than the 7950 core.

Impossible to answer without knowing what their internal benchmarking is for whether it becomes a 7950 or 7970. Still not seeing any evidence to suggest that the 7970s go further on a representative sample .

Edit: to clarify my position, I'm not saying that what you're saying is necessarily incorrect, I'm just keeping an open mind as I haven't seen anything comprehensive enough to really believe it.

 

[Rusty0611]

This is a lottery though, price doesn't come into it. Clock for clock difference is 5% for arguments sake. The majority of 7950's will not match 7970 at maximum overclocks. If this was the case we'd see more of it here in the many game and benchmark threads.

You can't say majority without taking a statistical sample lol. You don't have the evidence to make those claims.

How does the evidence that the 7970 has faster memory not come into it your decision on which one to get?

I don't believe it makes that much difference in performance in games going from your average 7950 memory OC to your average 7970 memory OC.

How many 7970's reach 1300 core? Not loads but quite a few here. How many 7950's reach 1300 core? Much less. There's something in that. Assuming the cores are identical those other factors that you dismiss obviously do make a difference.

They are factors to be considered though and the faster memory will mean more performance as will the extra shaders and compute units. At the same clocks these seem to mean an extra 5%. With the extra clock speeds you will likely get from the core and memory i reckon you could be looking at another 5% maybe more.

Again statistics about more and less without evidence. Can't do that mate.

The rest is all on the assumption that they do clock faster on average. Still not seeing any evidence that they do. Mate you're not going to convince me based on a few samples from here where people have Accelero's and water blocks on them lol. The 7970 is the top end card and as such is likely to net the enthusiasts who benchmark more and push the cards more.

You're also mixing up maximum clocks a card can achieve and average clocks. I have no doubt what will top out higher of the two but I thought we were talking averages and now we're talking selective samples based on golden cards .

If we take a pretty decent overclock of 1250 on the core then what proportion of 7970s can do that speed vs what proportion of 7950s (voltage unlocked models)? I don't know but from my own mini sample I didn't see anything to change my mind that your chances of getting a card which can be pushed to this level is quite high with the 7950s. I don't know about the 7970s as I've not owned anyway but I don't see any reason for it be higher and I've seen no proof that is the case.

Also if we're talking games, then I struggle to believe that many people run their 7900 series cards at 1300+ 24/7 .

 

[Rusty0611]

I get that Matt but how sure are you that a) the cooling set up is the same and b) the competence of the person using the card is the same?

You might be right but you just can't make claims of majority, more and less without a large sample. Having a gander at a few benchies on the net doesn't count.

I paid £230 for my 7950. You've paid, what £300 + £50 for a cooler and yours while gaming performs (including the extra clocks) probably around 7-8% faster. Not worth an extra £120 to me personally but I can see why some would want to. If they were the same price then of course I'd prefer a 7970. But then if everything was the same price I'd want a Titan

 

[Rusty0611]

Cooling setup will vary but it won't on reference cards and that is where all the early official reviews confirm, the 7970 clocks higher on average.

In reviews they don't push cards to the utter limit.......

 

[Rusty0611]

But we were talking about averages as you reminded me a few posts back.

LOL yes the average of what they top out at just not the highest possible nor the average of what they do on stock volts or with a small adjustments or whatever sorcery they do in reviews these days .

 

[Rusty0611]

Fine by me, partner.

 

[Rusty0611]

Yeah I thought about using HWBOT but the amount of users pushing the card to the limit on the 7970 is by definition going to be far higher as users stick this under exotic cooling.

 

[Rusty0611]

Seperate tables for water, air, cascade and LN2

The sample size is 10x smaller for the 7950s, that shouldn't matter too much as its an average. The average does seem low though!

The highest clocking 7950 is 1330/1746...using LN2!
We've had someone clock to over 1300 on air on this forum lol (yours did 1300 didn't it?)

http://hwbot.org/submission/2342315_dblooi_3dmark11___performance_radeon_hd_7950_14386_marks

Mmm. I remember discussing the sample before on the forum. The sample is 10x smaller as you say so not really satisfactory to take as ultimate proof.

Yeah I had a 1300/1750 on air (max I achieved). That was with 1.25V pumped through but the dodgy cooler prevented me from trying more. I was nearing 95c with that .

I would have thought most people on hwbot were pretty serious about there overclocking thus pushing out what they can from there cards. From the start of the 7 series release it was always claimed by many that the 7970 would clock higher on average compared to the 7950 and from all that i have seen it does appear that you have a better chance to clock higher on a 7970. It may not be factually proven but i have seen more results to prove it than to disprove it.

No that's your own take on the matter. It's always been the case on here that they roughly top out at the same point just recently now the clock for clock difference has been quantified there's been some purchase justification wheeled out when really the enthusiast doesn't need to justify one's purchase... they got the best card.

Well isn't that surprise? No not really.

Thanks for the links though Ton.

As above, with one sample so small compared to one so large you can't draw any meaningful conclusion. If you're going to draw conclusions, you need sample sizes the same otherwise the conclusions aren't acceptable.

 

[Rusty0611]

Its not my take on it at all. I have never used either card so i have no take on it. I do read a lot though and there are many who say this. It was banded about a lot especially at release and soon after.

LOL well of course it is your take on it because as you say there is no proof either way...

The problem with using HWBOT as a representative average is:

a) sample sizes
b) the average overclock on air for the 7950 is quite low and then shoots up on water when in fact at 1142 MHz you're nowhere near being temperature limited anyway so why would moving to water suddenly increase your average overclock

Ah, I hear you say, people put the better chips under water and push... I agree but again, without the number of samples increased to smooth out the averages, we're guessing by using these results.

Anyway, waffling slightly the basic answer is the sample sizes are vastly different so you can't use this as definitive.

 

[Rusty0611]

Strangely my experience differs in that out of 6x 7970s and 3x 7950s I have tested the best overclocker was a 7950 that did 1230/1700. On average 7970s and 7950s have all overclocked to around 1170/1600. I have never had a 7970 that would go beyond 1200 core stable for gaming and that includes a Matrix Platinum.

My own testing shows no perference for the 7970 to have better OC potential than a 7950.

You can find a lot of similar people having experienced the same. And you can also find a lot of people having experienced the opposite.

Personally I think it just isn't right to say definitively that the 7970 clocks further than the 7950 based on anecdotal evidence and therefore should be factored into the decision of purchasing.

It might be a smaller sample size on the 7950 but its still a good amount of 7950's and with a bigger sample size the average oc could go down not up. Is the sample size big enough, i can't answer that but it supports whats already been said since release. From all the reading i have done there are many more people reaching over 1300mhz on a 7970 than there is on 7950's. Has there been more 7970's sold than 7950's? which i doubt is the case. I have seen some people with 7950's over 1300mhz but its been very rare compared to the amount of 7970's even on this forum.

It doesn't matter if you've got a "good" amount of 7950s, the size of the sample is vastly different therefore....

What's been said since release is just anecdotal evidence and if we're talking in the realms of chance/probability of overclocking to x clock speed it doesn't hold any weight. As is the rest of what you're saying.

 

[Rusty0611]

There is more evidence in support of the 7970 clocking better compared to the 7950. We can never prove this 100% but even you have to admit theres more evidence in support than there is against. I myself love facts but sometimes you just have to look at the evidence in hand and make an informed decision.

If by evidence you mean stories of what people have achieved then, yes, there's more. But what about all the people who haven't posted what clocks they've got, either way... what about people like me who had good 7950s (3 out of 4) and didn't post them on HWBOT?

My point is not that the people who say that the 7970 overclocks further than the 7950 are necessarily wrong... it's more that the evidence people are using to back up this claim is purely anecdotal and any technical reasonings put forward does not satisfy the hypothesis .

You can spend more on a 7970, get ~5% more performance clock for clock, and hope that you get a decent clocker. Or you can get a 7950, get ~5% less performance clock for clock and hope you get a decent clocker. Hoping for a decent clocker is common throughout and then if you move from that and start saying: "well your chances of getting a good clocker with a 7970 is higher".... you'll get people like me saying, OK, where's your statistical analysis?

Where I think you do have a higher chance of overclocking further would be with the Lightnings/Matrix Plats...

I think you just replied without reading.

My post basically states that I feel based on my own experience, it is a myth that a 7970 core will overclock higher than a 7950 core.

No I think that is you friend. I was agreeing with you.

 

[Rusty0611]

It's been said vram clocking doesn't make much difference, but BF3 absolutely loves high vram clocks to name but one title of benefit from vram clocking.

I'm going to run a test at 1500/1300 memory clock speeds to see what difference it makes in the same games as benched above later on. Not convinced it makes a massive difference and I've not seen much evidence of games scaling all that brilliantly with a memory overclock.