5850 vs 5870 - My Results & Opinions

[555BUK]

After the earlier silly debates, I decided to post my own experience of the 5850 vs 5870. I have owned both cards and tested them on the same system so I am better placed than most to provide a fair comparison.

PLEASE - No bickering within this thread. Keep it clean and constructive.

My System
ASUS P6T-SE Mobo (flashed to P6T)
i7 920 C0 Processor @ 3.8GHz
6GB OCZ PC10666 DDR3 RAM
Corsair HX850W PSU
Antec Six Hundred Case
Win7 64 Home Premium
*ASUS 5850 (clocks to 1000core @ 1.25v / 1200mem)
*ASUS 5870 (clocks to 1000core @ 1.3v / 1300mem)

The Tests
Here are the results of 3DMark Vantage tests performed at varying speeds. MSI Afterburner was used for clock and voltage tweaks because it is easier to use than ASUS Smart Doctor.

Test 1 - Both Cards at stock settings

Test 2 - Both Cards at 5870 speeds (850core / 1200mem)

Test 3 - Max OC 5850 vs Stock 5870

Test 4 - Both Carda at Max OC

My Vantage Benchmark Links
ASUS 5850 Stock
ASUS 5850 OC to 5870 speeds
ASUS 5850 OC to 1000core / 1200mem
ASUS 5870 Stock
ASUS 5870 OC to 1000core / 1300mem


My Opinion (others are free to differ)
*For anyone that does not overclock there is a large difference between both cards performance. The 5870 is >20% faster.
*At the same clock speeds the 5870 is about 5% faster overall. In shader intensive tests this increases to ~10%, which is to be expected because the 5850 has 10% less shader cores.
*When both cards are overclcoked to the max there remains a ~5% advantage to the 5870.
*For anyone who overclocks, the 5850 offers much better value. 5% more performance currently costs 50% more money, and unless you are either a hardcore bencher, or have money to throw away it is really not worth it.
*Core clocks on both cards reached similar levels although the 5850 required slightlly less volts. This may be due to the 5850 having 10% less shaders.
*The memory on the 5870 overclocked to 1300 vs 1200 for the 5850. This may be pure luck, but it may also be that ATI supply lower voltage to memory IC's, regulation is poorer, or the IC's themselves are not as good.
*An overclocked 5850 can easily beat a stock 5870
*Some cards will overclock different than others so mileage may vary.


edit: Crysis Warhead benches added

edit 2: A few additional observations
*The 5870 cooler is better. At 1000core the 5870 ran 5 degrees cooler than the 5850, even though the latter needs 0.05v less volts and has 10% fewer shaders to power. From memory the 5870 also felt substabtially heavier.
*The extra inch of the 5870 is due to better phase control. This should allow for better average overclocks and possible better longevity.
*Both of my cards may overclock higher than 1000core. I have not tried above 1.3v on the 5870 and 1.25v on the 5850 due to temps. I had no intention of martyring either card.

edit 3: Unique Heaven Demo Results
These results compare my 5850 with oweneades 5870 (my old 5870).

Settings
1680x1050
DX11
Shaders - High
Tessellation - Enabled
Anisotrophy - 4
AA - Off
Full Screen - Yes

5850 @ Stock clocks 34.2fps
5870 @ Stock clocks 40.4624fps
difference 18.31%

5850 @ 5870 clocks 39.8fps
5870 @ Stock clocks 40.4624fps
difference 1.66%

5850 @ Max 1000/1200 clocks 44.7fps
5870 @ Stock clocks 40.4624fps
difference -10.2%

5850 @ 925/1200 clocks 42.3fps
5870 @ 925/1250 clocks 43.4fps
difference 2.6%
*attempted bench at 1250mem for my 5850 crashed during test 10. Mem clocks left at 1200.
*5850 benched on Win7 64, 5870 on Vista 64. This may contribute to the small differences.

edit 4: Only ASUS & MSI 58xx cards officially support voltage tweaking
If you want to overclock your 5850 to 5870 levels and above you will need a few things

i) A 5850 BIOS that allows clock speeds above normal CCC levels and voltage tweaks. Currently only ASUS and MSI cards include such BIOS, although all 5850's are physically the same. If you do not purchase one of these cards you will need to flash the BIOS and risk voiding your warranty.
ii). A utility that can alter GPU volts. MSI Afterburner is currently best IMHO.
iii). A card with GPU and memory that can handle the speed.

To keep things easy, I suggest buying MSI or ASUS. If you own a different card, or have no qualms flashing BIOS you can Google "5850 BIOS Flash" to find dozens of guides.

edit 5: 3DMark06 Tests Added

3DMark 06 Observations
*At equal clock speeds the 5870 has a ~2% performance advantage.
*Overclocking either card shows noticeable gains within the Canyon Flight Test only. Withn other tests there is little difference, even between a stock 5850 and an overclocked 5870. This leads me to believe that for 3DMark06 my system is either i) CPU limited - my 3.8GHz i7 is too weedy, ii) this benchmark has discovered an architectual limitation of 58xx series cards, or iii) Most tests within this benchmark are too outdated for relaible results. I think iii) is most likely.

Above 3DMark06 Results Links
5850 @ Stock
5850 @ 5870 (850/1200)
5850 @ 1000/1200
5870 @ Stock
5850 @ 1000/1200

 

[555BUK]

Unfortunately I no longer own the 5870, and most of the benchmarks I took at the time were not recorded. All I have are Vantage, 3DMark06, and Crysis. I may post 3DMark06 results tomorrow, if I have time to bench my 5850 (I only received the card today).

 

[555BUK]

Unique Heaven benches below. See original post for full details. Thanks to oweneades for the 5870 results.

5850 @ Stock clocks 34.2fps
5870 @ Stock clocks 40.4624fps
difference 18.31%

5850 @ 5870 clocks 39.8fps
5870 @ Stock clocks 40.4624fps
difference 1.02%

5850 @ 925/1200 clocks 42.3fps
5870 @ 925/1250 clocks 43.4fps
difference 1.03%

 

[555BUK]

The first two tests within 3DMark Vantage (Jane Nash & New Calico) seem to provide a good indication of true 5850 vs 5870 performance, when both cards are clocked at the same speeds.

The individual fill & texture tests show where the additional cores on the 5870 make noticeable differences. Texture Fill, Pixel Shader and Perlin Noise all scale exactly with the number of active shader cores. The remaining tests show no real advantage.

No game is solely dependent on Shader performance, and limiting factors may be found within one or more other elements of GPU and/or memory architectire. Crysis is probably the most shader intensive game currently available, and even there overall performance gains are only 5%. I doubt any game will ever show a clear 10% difference at the same clock speeds. Only benchmarks which isolate particular areas (such as the three Vantage tests) are able to show this.

I think this proves that both 58xx cards are "shader monsters", and that any weak points (bottlenecks) will likely appear within other areas.

 

[555BUK]

how do i clock a 5850 past 775 1125??

Currenty just two manufacturers (ASUS & MSI) produce 58xx cards with a BIOS that allows cards to clock past CCC defaults. If you do not own one of these cards you will need to flash the BIOS (to ASUS or MSI) in order to unleash the beast. This will likely void your cards warranty. There are plenty of guides available via Google.

 

[555BUK]

3DMark 06 tests added to original post.

 

[555BUK]

why didn't you do 925/1200 on the 5870?

I no longer own the 5870. oweneades ran the benchmark at 925/1250 on his (my old) 5870, but unfortunately my 5850 cannot match his memory clocks. Memory clocks seem to make little real difference within the Heaven benchmark, so 925/1200 vs 925/1250 will still be pretty accurate. If oweneades is still reading this thread perhaps he will run a bench 925/1200 for me.

 

[555BUK]

Theoretically, the absolute maximun advantage a 5870 will achieve over that of an equally clocked 5850 will be 10%. This 10% will only be possible if a game/benchmark requires shader resource, to an extent that the 5850's shader processing capacity becomes the sole bottleneck within the PC. For this to happen there must be no CPU, memory bandwidth, or other GPU limitations whatsoever within the system. It can also be argued that any game which utilises the shaders so highly, without placing loads on other parts of the system will in itself be "poorly coded" or "deliberately coded" to exploit such limitations.

From bechmarks posted, we see that the most likely differences will be in the region of 0% to 6%, dependent on game (with most games being closer to 0 than 6). The real advantages of the 5870 are:- better cooling, better power regulation, better stock clocks (for those that do not overclock) and better potential for overclocking.

 

[555BUK]

Just added my 4890 xfire to your figures. A very small OC of 900/1000.

That is very interesting. To me it seems 4890 Xfire and the 5870 are pretty even in performance terms. Some games will no doubt prefer a pair of 4890's and others will prefer the 5870 (or a highly clocked 5850). With the price of new and used 4890's so low, they still offer an awful lot of performance for your money.

 

[555BUK]

You can overclock the 5870 to still spank the pants off the 5850 anyway.
The 5870 is more future proof in both single and crossfire configs (the 5870cf being the fastest dual GPU soluition avaliable.

Please believe me that a 5870 will not spank a 5850. Even with both cards at stock speeds the 5870 is only a fifth faster. I agree that 5870's will generally overclock better than 5850's and that the extra 10% shaders will improve FPS by a few percent, but with both overclocked to the max the difference is likely to be less than 10% in real terms. Only benchmarks would make a 10% increase in FPS apparent.

I am not knocking the 5870. It is a great card, but for most people it is not worth 40-50% more money (or 30% if you are one of those people who used to borrow our Ashes).

 

[555BUK]

I guess anyone who games at 5760x1080 is pretty extreme and would understandably desire (and probably afford) the absolute best. At such resolutions the extra shaders may indeed be more noticeable than at 1920x1080, but it is also just as possible that lack of physical memory or memory bandwidth will be the bottleneck (High Res plus AA is memory hungry). For the >99% of people who game at 1920x1200 or lower, I still believe there is little "notceable" difference between either card when clocked at same or similar speeds. I also think it likely that gamers who use 5760x1080 will find a single 5850 or 5870 "insufficent" and would be better suited to a crossfired or 5970 setup.

IMO the only people who should consider the 5870 over the 5850 are as follows.
i). Hardcore gamers and must-have-the-best enthusiasts
ii). People who do not want to overclock, but who require great performance.
iii). People who have too much disposable cash to be concerned about value.

I purchased the 5870 because of i), but sold it on and kept the 5850 because it just made better sense for me. I was fortunate enough to be able to test both within the same system.

 

[555BUK]

You won't find a game more shader heavy than crysis at max settings so run the crysis bench at 1680x1050 @ very high @ 5870 clocks and I will guarantee you the 5870 will eat the 5850 alive

See first post for this exact benchmark. At the same clocks the 5870 is 5.43% faster. At max OC (both cards) the difference is 7.4%. An overall 5-8% difference within a very shader intensive game seems pretty "similar" to me, but it may qualify as "Eating" to others.

 

[555BUK]

Well i tried the 5850 and 5870 at stock in Crysis and warhead benchmarks at max 1680x1050 and the 5850 just wasnt smooth compared to the 5870 so i went with the 5870 - regardless of the percentage difference the real world playability difference was noticeable.

Out of interest, did you overclock the 5850 to similar levels as the 5870? I can understand a stock 5850 being noticeably slower but that would be caused mostly by slower clock speeds. If both were running at the same clocks I find it extremely unlikely differences would be noticeable. My benches showed a 1fps difference in min fps, and to be honnest both cards stuttered a little with settings on full and 4x/8xAA in certain scenes. Perhaps you had a sick 5850 which was throttling?

 

[555BUK]

At both 1680x1050 and 1920x1080, and at max settings the 5850 does a brilliant job.

 

[555BUK]

555BUK What cooler were you using on the 5850?

Stock/reference cooler only.

 

[555BUK]

how did you work out the % difference?

EDIT: it was the bottom 2 that confused me... uve rounded them up and put the decimal point in the wrong place
take the middle one.. its not 1.02% difference its ~102% (rounded up from 101.66432%) = 1.66% difference
bottom one... 102.60047% = 2.6% difference.
you got the top one correct though, well done

You are completely correct. I have now changed the figures within the original post. Thanks for correcting my maths.