Is the 5870 useless for us overclockers?

AMW1011

Distinguished
Ill start with the facts:
perfrel.gif

http://www.techpowerup.com/reviews/ASUS/EAH_5850_TOP_DirectCu/28.html


http://www.anandtech.com/show/2848

So the 5850 is a good 10-15% slower than the 5870. The 5850 also has 11% less SPs and TUs, so this makes sense. Also, the clocks are lower, 17% lower on the core, and 20% lower on the memory. With a bump of clockspeeds were the core is 11% and the memory is 7% slower on the 5850 TOP and the 5870 stock, we see the performance lead of the 5870 shrink to under 10%. This would tell us that at the same clock speeds, the performance difference between the 5850 and the 5870 is going to be well under 10%.

Take a card like the Asus 5850 TOP that can pretty easily hit 1GHz on the core and 1250-1300 MHz on the memory with voltage increasing. Compare this to a stock 5870, that can hit the same 1GHz on the core and 1300-1350 MHz on the memory with voltage increasing. Both top out at about the same level with the cores, and the 5870 seems to have an edge with the memory. This edge costs you $80 more for the reference 5870, which will not give much more performance so we are still looking at an under 10% difference here. Where is memory important? With extreme resolutions. However, we already know that any setup with 3 or less 1080p monitors doesn't show a benefit from even a full 2GB of VRAM, let alone a few clocks on the memory. This is backed up with the fact that the 5xxx series is core limited, not memory limited when it comes to clockspeeds.

So since the cores on the 5850 and the 5870 top off at about 1GHz with voltage adjustment, we are looking at the same cards with an 11% decrease in SPs and TUs on the 5850. The above benchmarks show that this 11% decrease does not equate to a linear performance decrease.

Here is my theory: With both the 5850 and the 5870 overclocked to 1GHz on the core, we are looking at an unnoticeable performance difference between the two, with one costing $100+ more. It seems to me that the 5870 is a useless waste of money for those of us who overclock. Am I missing something? I knew the 5850 was by far the better buy, but I never knew how little the difference actually was, I assumed a somewhat linear performance difference with the decrease in SPs and TUs.
 

jimishtar

Distinguished
May 15, 2009
622
1
19,010
u make a good point. the thing is, the majority of PC users don't know how to reinstall windows, not to mention overclocking CPU or VGA. that is why the typical user/gamer just goes in the shop and gets the most powerful card he can get.

yet there are overclockers that simply feel better just knowing that they have more raw power in their rig. my mugen 2 did his job just great, but I bought WC anyway.
I, personally, prefer the 5850, but if I had the money, I would get the 5870.
 

AMW1011

Distinguished
I'm in a similar category of enthusiast, but knowing that there is no way I can get a noticeable difference out of the 5870 over a 5850 for an extra $100 is just too much, I would not buy the 5870 even if I had the money.

Though WC is different. It allows you to cool everything you have very nicely, and is easily upgraded to cool new hardware. WC is more of an investment than most pieces of hardware such as GPUs.
 

Annisman

Distinguished
May 5, 2007
1,751
0
19,810
Well, the 5870 can be overclocked as well, if you have two of them, both overclocked in crossfire, the extra gpu power will add up.

Don't forget the 5850 is an amazing dollar per/performance card, the 5870 appeals to those with a bit more money to spend.
 

AMW1011

Distinguished


I compared them overclocked and showed that the extra performance is just not there, and crossfire will lower that difference because of scaling.

In the end, the 5870 should still clock a bit higher on water(100mhz in cases) and the extra SPs should still give it an advantage, youll end up with ~10-15% better perormance, which for some is a nice advantage for only $100.

Besides, there's people whos willing to pay an extra for a GTX480, which is barely better than a 5870, except in Tessellation and AA, and OCs lower.

Both the 5870 and 5850 top off at about 1GHz on air, on water who knows, we have no data on that. The extra SPs don't seem to do much at all, look above. With the 5850 clocked 11% on the core and 7% on the memory lower than the 5870 we are seeing a <10% difference, so where is this 10-15% coming from at the same clocks?

Also, the GTX 480 overclocks just as well as the 5870, 150-200 MHz on the core being very common. It also gains more per clock.
 

AMW1011

Distinguished
There is a stock 5850 and the Asus 5850 TOP. The TOP is clocked slightly higher which decreases the performance difference to under 10%, but it is still pretty far away from being the same clockspeed. The Asus 5850 TOP also costs $80+ less. Yeah an aftermarket 5870 would run cooler, but that is about it. I don't really see you point.
 

Annisman

Distinguished
May 5, 2007
1,751
0
19,810
Maybe I don't get your point.... there are always more expensive GPUS out there with diminishing returns on performance. Remember the 8800Ultra ? The 'top' cards of any lineup demand a price higher than they should merit, this has always been the case. If you are trying to say the 5850 is a better bargain, well of course it is, that has never been disputed.

This is the same game that has been played for many years, so why take the time to point it out now ?
 

AMW1011

Distinguished
Well you have a fair point annisman, I am just surprised that the difference at the same clock speed is going to be actually unnoticeable, and both can overclock about the same. I am just surprised at how close they are, I never looked at the 5870 like an 8800 Ultra before.
 


I'm curious, how does crossfire scaling effect the difference between the 5850 and 5870?

Last I checked, a 10% increase remains a 10% increase, even when you add another ~50% by way of crossfire. (The 10% applies to the crossfired card too).

Crossfire will increase both cards performance by the same percentage, the 5870 will remain 10% faster. Percentages remain the same. You are thinking the performance is addative, it's multiplitive (is that even a word? :p)
 

AMW1011

Distinguished


What I meant was that a 10% clock increase on both cards is going to amount to a 10% performance increase with the setup, it won't double to 20% because you are comparing both in 2 card setups.
 

Annisman

Distinguished
May 5, 2007
1,751
0
19,810
Are you seriously pitting an overclocked 5850 vs a stock clocked 5870?

This is where I was confused, earlier at least.

Yeah, if you can get a monster 5850 that overclocks like a champ then I guess the 5870 looks less attractive, but if you get a 5870 that oc's like a champ does it make the 5970 looks less attractive too ? Of course not, the 5970 is made to blow both cards out of the water. I think what you are surprised about is that the 5850 and 70 are so close together, but you have to remember the price isn't THAT far apart from each other.

Again, I hope I've made the point, that performance between cards will not always be tied to their price on the market.
 


Again, I am still wondering. Why did you say crossfire lowers it's performance gain?

Even if crossfire increased performance by 300%, the difference between a 5870 and 5850 system would remain at 10%, if they are 10% different before you crossfire. But remember that 10% of 400% is more raw FPS gained, but still the same percentage difference.
 

AMW1011

Distinguished


That is because a non-reference 5870 won't get you much past 1GHz anyway, even the MSI Lightning, the king of non-reference 5870s.

Look here:
http://www.techpowerup.com/reviews/MSI/R5870_HD_5870_Lightning/32.html

He could only push 1030-1050 MHz on the core with safe voltage

http://www.guru3d.com/article/msi-r5870-lightning-review/24

Couldn't hit 1GHz even with voltage, which is rare.

http://hardwarezone.com/articles/view.php?id=3158&cid=3&pg=7

Hit 1GHz on the core with some comfortable voltage.

Now the TOP:
http://www.legitreviews.com/article/1273/10/

Easily hit 1GHz at a lowish voltage.

http://www.overclock3d.net/reviews/gpu_displays/asus_5850_top_review/4

1GHz at a similar voltage

http://www.techpowerup.com/reviews/ASUS/EAH_5850_TOP_DirectCu/31.html

Only could hit 970 MHz on max voltage

http://www.pureoverclock.com/review.php?id=954&page=15

1GHz at max voltage.

As you can see, even good non-reference 5850s and 5870s top out at about the same clocks. The 5870 might have a slight lead with some extreme watercooling because the voltages can be increase a little higher, but that is about it.
 

AMW1011

Distinguished


That is not what I meant, it was worded awkwardly. Refer below:



I meant that it will not, the extra performance from the overclock is constant not additive. I realize what I said sounds wrong, but what I meant was that you are comparing two cards in crossfire so the scaling of these two cards will equate to a direct increase from overclocking, you can't add the bonuses in this context.
 

Annisman

Distinguished
May 5, 2007
1,751
0
19,810
Ok, I think I get what you are saying: The 5850 seems to oc better than the 5870, so why bother with a 5870 at all.

I understand this perfectly and your point is excellent.

However not everyone is comfortable with volt mods (even software ones) and right now, I don't beleive there is a way to voltage tune cards in crossfire, I have tried to no success. Therefore two 5870's in CF still excell against two 5850's when overclocking (espescially without voltage adjustments) is out of the question.

But as your OP refers to us 'overclockers' I would say you are correct.
 


Well an aftermarket overclocked 5870 is not going to cost 100$ more than a 5850. Lets take this GB card as an example :

http://www.newegg.com/Product/Product.aspx?Item=N82E16814125323&cm_re=5870-_-14-125-323-_-Product

GB_5870.jpg


as you can see the above 5870 is priced at 499.99 and the clocks start at 950/1200. The DirectCu can do 1000/1275 easily just by raising the clocks/volts. I think this is what AMW is trying to explain here.

max_unique_1000_1275Mhz.png