HD5870 battle of the throne

We at OverClocking-TV are proud to present a new concept.

The concept is offering to use our matrix and professionally review the top candidates among the overclocking series of graphics cards from each and every vendor.

For our first edition we have focus on the HD5870 Series, Cypress XT.

This summer ASUS, Gigabyte, MSI, PowerColor, Sapphire and others have claimed king of the throne, but, there can only be one king? Or can there?

If you expect to see a lot of different game tests, well, you might as well just quit reading now, since this will only be a great waste of your time. We believe that if you are considering one of the overclocking models, you are already fully aware what to expect of the given series, since you are willing to pay the additional cash to get a “speciel Edition”

By now, you’d say either “mah” or “wuhuuu”, but, unfortunately we are not capable of getting all vendors in, so, hard luck for you, us and the vendors as well.

We got positive feedback from the three huge Taiwanese vendors of:

• Gigabyte
• ASUS
• MSI

I had in my possession an ES of the Gigabyte SOC, but, for this test, retail counts, not really fair when there is a chance of cherry picking, is it? So;

• Gigabyte kindly send me another sample of HD5870 SuperOverclocked, despite the fact that I already have an ES.
• ASUS provided a MATRIX card, however, this appeared being an ES, and did therefore not count for the test.
• MSI provided us their flagship LIGHTNING (We need a audio file from Cebit with Olivers “liiiiiightning”)

In addition to these cards, we have chosen to add in a reference design card, this being a XFX HD5870.

[nggallery id=265]

Starting off, lets see the performance as per default, but before doing so;lets represent

The test platform:

• Processor: Intel i7-980X @ 4455mhz (4125 uncore 3960 QPI)
• Motherboard: Gigabyte GA-X58A-UD3R (rev.2.0)
• Memory: Corsair Dominator GTX2 (2250 8-8-8-24) @ 996mhz 7-7-7-21
• PSU: Corsair TX950W
• SSD: Corsair Reactor 120GB
• OS: Microsoft Windows 7 Ultimate
• My own homebuild testbed.

Cooling method:

Watercooling on processor, for watercooling results on GPU, a Swiftech MCW60 was used, and temperature was kept within delta 2C°. For DICE (Dry Ice, compressed CO2), LN2 (Liquid-Nitrogen, Liquid N2), duniek pot. Assisting air for replicating max clocks and prevent condensation; Enermax 120mm (x2)

Accessories:

• Toolmate Voltmeter
• Voltcraft K204 Themometer

Software:

• Futuremark 3Dmark Vantage
• ATI Catalyst 10.6 Control Center
• MSI Afterburner
• Gigabyte OC Guru
• Capture
• GPU-Z 0.4.3

–       So summed down, enough with the talk, time for the walk!

First thought, “Not fair with the range”, oh yes it is, its predictable that it will be in about same region, so lets just make it easier to spot the difference in the results! Unfortunately, despite I would love to neglect adding more text, I feel a pressure to give a bit background on how the tests were performed, therefore a small chapter about testing methodology

[singlepic id=9601 w=320 h=240 float=center]

Testing methodology:

Running 3 test, separated by a reboot, running with exact same software running, even the software not used; Afterburner & SOC Guru. The average of the three runs is the listed run.

There is still some variables… Of course deltaT has increased over the test period, by 1-3C°, it is a very long test, and a 46m2 size room (3 meters tall, making it; 138m^3). Same driver used… 10.6, newest driver when project started. Appears getting liquid nitrogen isn’t that easy everywhere. Please do not compare performance, nothing has been done to improve or decrease so, its sourly based on comparing the individual cards with one another.

Next up, the question popped up, how about at same frequency, do they perform identical, or has the vendors been making tuning in the BIOS as well?

Easy to sum up: a big fat NO, these few points are clearly the error margin of the quantity of runs performed.

A bit more interesting for the mainstream consumer of fabric overclocked products, overclocking with the default cooler:

These results were a bit surprising, the reference card is almost taking the lead, however, one thing we have neglected in the test is the sound level, it was by far more painful for the ears, yet, the result is magnificent.

Moving on to water!

The reference design seems to hit the wall, while the SOC and lightning are pretty close.

Moving on to the more exotic cooling, the one you will not do daily, but only for your 3D tests!

Compressed CO2, dry ice, were not actually used, temperatures were held by -65C with LN2, this providing the cooling performance of dry ice.

Tests with LN2 were made as well, max cooling, but unfortunately, it seems that the cards simply did not scale further, a highly likely reason being voltage limitations in software/hardware restricted.

Summed up:

I made some more notes that you guys might find interesting, this not about the actual performance, but the frequencies – trust me, there was a lot of work to finding the max in each and every scenario:

Now if you are considering some of the card, and have value into perspective – forget about me writing prices on the products, however, the technical part I will deliver, so these calculations are from reference clocks 850/1200

So there you have it, some data to estimate your purchase from, hope it was of use! Finishing off I have a few more things.

Notes:

• With lightning, some artifacts appeared, especially in GT1
• Lightning overvolts; 1.3v = 1.33-1.34v during game tests
• Variables in the test; AMBIENT, please do have understanding, test takes more then 10 hours, for obvious reasons, ambient isn’t constant
• VGPU isn’t random, just no reason for writing (CRASH)
• GT1 crashes from high vGPU, while GT2 from too low… I did not change this between tests, so max is max stable for a full run.
• 2+3 in avg. With OC was made after SOC was tested, due instability on the Lightning… Therefore perhaps a slight impact on performance
• XFX BBA, measuring of vCore wasn’t performed
• SOC 1.225vgpu = 1.31v and 1.619 = 1.66v vdimm

[singlepic id=9602 w=320 h=240 float=center]

Conclusive thoughts:

By far if you want fastest stock card, go for the SuperOverclocked, while if you want to overclock to your max using the cooler the product is equipped with, we suggest that you go for a reference design as they simply feature a powerful cooling solution.

Once going above normal cooler, it is highly beneficial using SuperOverclocked or Lightning, they are simply stronger made products and will have much higher performances.

Also a bit individual guidance for the companies that choose to enter:

• MSI, Lightning simply doesn’t have the cooling that it deserves.
• Gigabyte; The 1.3v limit on the GPU is ashame, it has so much more power just waiting for being wheeled, but held back by the limitation set.

Feedback from our readers:

Please give us feedback if you appreciate this concept of testing, and please let us know if you wish for us to modify the products making them perform their outmost – we need your backup and confirmation in order to bring forward the word to the vendors, in case you wish for such tests.

We would like to thanks our Partners GIGABYTE, MSI and XFX without whom this battle wouldn’t have been possible.

Article by Marc Beier.