How do they benchmark on any BOINC projects vs the other AMD/ATI cards?

Way ahead thats for sure - but that was always going to be the case being a 28nm card. What is only just beginning to dawn on all is just how much of a beast this Architecture is.The step change is on par with the leap forward the 5XXX range kicked off, arguably this one is greater.

You get a feel for it when looking at the very extensive multiple reviews at http://www.guru3d.com. Many individuals often quote the box level ratings on speed/capacity etc. However that only scratches the surface of whats under the hood. From within the Driver itself it will go from default 925/1175 to 1125/1525 without breaking sweat, and a a power slider adding 20% to facilitate that if needed iss within the driver. That overclock however is routine stuff. The heat level is astonishing - frankly its Cool coming out the case even at high clocks ...... my twin 5970 at times could cook breakfast if not careful.

3D Mark11 Comparitors by card: http://www.guru3d.com/article/radeon-hd-7970-overclock-guide/14

The serious overclock - as usual - comes from where the Vendor dont want you to go rofl - externally applied additional voltage. The routine overclock inside the driver boosted the card from stock 3DMark11 7924 to 9138, anyone can achieve that, total newbie as its all inside the driver with safeguards. Unprotected overclocking takes that to a stable 12000 - 13000 minimum, already reports of 14000+. The 3DMark11 scores for a quad setup are just insane, all for a single GPU card.

Yet to be fully factored in is driver optimisation - this is all with initial deployment drivers and all that entails, and no factored in effect of PCI-E3 or their new Architecture (GCN) - the latter is going to make a very large difference. Maybe 20-30% above the serious overclock performance ? Who knows wait and see, but ..... do the numbers.

Will NVidia also come up with the goods, yup I reckon this time they will, its not a repeat of the 5XXX deploymenr where NVidia was laid waste by ATI for two years until the 570/580 which are nice cards. NVidia have had every oportunity this time to make sure they dont repeat the firmi 3XX disaster and 4XX debacle. I reckon they will come up with something tasty this time, and that can only be good for competition .... and our favourite topic, push down prices :)

So yup, 7XXX series is going to be a very very nice start to the 28nm world - interesting times ahead this year.

Regards
Zy

Got one 7970 - but it not crunch

load 0%, no progress

could it be the infamous CAL interface deprecated? Maybe we will need an opencl app now?
Are you sure that the card works well in every other aspect?

How do they benchmark on any BOINC projects vs the other AMD/ATI cards?

Way ahead thats for sure - but that was always going to be the case being a 28nm card. What is only just beginning to dawn on all is just how much of a beast this Architecture is.The step change is on par with the leap forward the 5XXX range kicked off, arguably this one is greater.

The 7970 looks very promising but are there any projects that it works with now? I notice that MW, Collatz and Moo! are not working so far.

How do they benchmark on any BOINC projects vs the other AMD/ATI cards?

Way ahead thats for sure - but that was always going to be the case being a 28nm card. What is only just beginning to dawn on all is just how much of a beast this Architecture is.The step change is on par with the leap forward the 5XXX range kicked off, arguably this one is greater.

The 7970 looks very promising but are there any projects that it works with now? I notice that MW, Collatz and Moo! are not working so far.

It actually works if you use the Nvidia OpenCL one.

How do they benchmark on any BOINC projects vs the other AMD/ATI cards?

Way ahead thats for sure - but that was always going to be the case being a 28nm card. What is only just beginning to dawn on all is just how much of a beast this Architecture is.The step change is on par with the leap forward the 5XXX range kicked off, arguably this one is greater.

The 7970 looks very promising but are there any projects that it works with now? I notice that MW, Collatz and Moo! are not working so far.

It actually works if you use the Nvidia OpenCL one.

Ah, found it. Looks like device 0 is doing ~60 sec/WU & device 1 is at ~96 sec/WU:

Found 1 platform(s)
Platform 0 information:
Name: AMD Accelerated Parallel Processing
Version: OpenCL 1.1 AMD-APP (851.4)
Vendor: Advanced Micro Devices, Inc.
Extensions: cl_khr_icd cl_amd_event_callback cl_amd_offline_devices cl_khr_d3d10_sharing
Profile: FULL_PROFILE
Didn't find preferred platform
Using device 0 on platform 0
Found 3 CL device(s)
Trying to show unknown MWDoubleExts 6
Device Tahiti (Advanced Micro Devices, Inc.:0x1002) (CL_DEVICE_TYPE_GPU)
Driver version: CAL 1.4.1658 (VM)
Version: OpenCL 1.1 AMD-APP (851.4)
Compute capability: 0.0
Image support: CL_TRUE
Max compute units: 32
Clock frequency: 925 Mhz
Global mem size: 3221225472
Local mem size: 32768
Max const buf size: 65536
Double extension: Unhandled MWDoubleExts

I haven't checked if the IL kernels still work, but if they do it should be a bit faster still.

The 7970 looks very promising but are there any projects that it works with now? I notice that MW, Collatz and Moo! are not working so far.

Its working over at POEM, thats the only one so far.

However, all the others are falling over quickly (less Collatz) aka MW & Moo around a few secs then boom. Prime Grid's takes about 30-40 seconds then falls over. POEM runs - but POEM is very low utilisation circa 18-22% on a 7970. At Collatz, the WU goes through ok, but doesnt validate against a Wingman - whether thats because the Validator doesnt play with a 7970 or in fact the output is- for now - bad, dont yet know.

The card itself is incredible, they have done a really good number on it - and well engineered .... it feels chunky and solid, not a plastic monstroscity

Look forward to trying it here when its tweeked to accept a 7970.

Regards
Zy

It actually works if you use the Nvidia OpenCL one.

Ok ... a fingers and thumbs 101 question - how would I pursuade it to download the NVidia one and crunch it?

app_info ?? If so what are the basic file name settings et al?

Regards
Zy

Some fun with 7970

best clock without voltage change


best clock


I think OC over 1200 Mhz is worthless. You can get 1200 Mhz without voltage change/with small change and get extra 3 sec. For 1300 Mhz I had to set almost 1,3V to speed up only 3 more seconds :(.

It actually works if you use the Nvidia OpenCL one.

Ok ... a fingers and thumbs 101 question - how would I pursuade it to download the NVidia one and crunch it?

app_info ?? If so what are the basic file name settings et al?

Regards
Zy

http://www.arkayn.us/forum/index.php?action=downloads;sa=view;down=33

I would make a guess that the plan class would be openCL and the type would be AMD.

---

I think I might pickup a pair of 7970s for running this project. The best I can get out of my 580 GTX is 128 seconds for a single task which seems to be decent but hopefully I can get even better performance out of the 7970.

How do they benchmark on any BOINC projects vs the other AMD/ATI cards?

Way ahead thats for sure - but that was always going to be the case being a 28nm card. What is only just beginning to dawn on all is just how much of a beast this Architecture is.The step change is on par with the leap forward the 5XXX range kicked off, arguably this one is greater.

The 7970 looks very promising but are there any projects that it works with now? I notice that MW, Collatz and Moo! are not working so far.

It actually works if you use the Nvidia OpenCL one.

So, how fast does it process WUs?

---

At the moment its running NVidia OpenCL app at around 52-53 seconds settings @1115/300. Which is a good time for a WU written to (principly) run on NVidia hardware.

Matt is getting a 7970, so we'll see performance against a AMD optimised app when he gets chance to sort it.

Good start though, right off the bat with no tweeking and running an NVidia app ...... would be interesting to see how a 6970 copes with the NVidia OpenCL app, that would give an early level playing field comparison.

I would expect significant speed increaae once the app is written as AMD Optimised. Further improvements will occur omce the GNC Architecture is brought into play on the apps.

Regards
Zy

Isn't those times faster than any of the NVidia cards?

---

Significantly faster, but that was always expected, its a 28nm card with better DP scaling than NVidia. Once someone runs a 6970 on that app, or Matt gets the app suited for AMD, a better comparison can be made. Its going to be significantly faster all round, no doubts there. But of course, some real world figures are where its at, and we will not get those comparitors until someone runs that NVidia app on their 6970, or Matt has the time to sort an AMD variant.

Regards
Zy

At the moment its running NVidia OpenCL app at around 52-53 seconds settings @1115/300. Which is a good time for a WU written to (principly) run on NVidia hardware.

Try to run 2 wus in parallel, you should get few extra seconds :). Now I have about 43 sec/wu @1200/685

Voltage Tweeks ..... hmmmm ... Back Satan .... back I say .....

Oh all right then :)

Beit on your Voodo doll pins head (got some sharp ones rofl)

Will report Carnage or otherwise shortly :)

Regards
Zy

Nice session .... :)

Got it down to 46 with two per GPU, running at 1200/685 1.2v, but didnt feel comfortable staying there 7x24. Ended up at default volts, and pulled back to 1125/685 with two per gpu, that felt more comfortable, and could sleep on that :)

Interesting session though. Found similar to you re going over 1200 gpu, did get to 1225 gpu at one point on 1.2v but it felt flaky with two per gpu. Really didnt want to press the new toy further, so backed off.

Settled down to two per gpu coiming out at 48-50 secs, that'll do for now :)

Comparing that with yours does verify how good the scaling in Crossfire is though, certainly happy with that.

Regards
Zy

Isn't those times faster than any of the NVidia cards?

Forget nVidia for MW, it's a waste of compute resources.

MrS

---

Scanning for our furry friends since Jan 2002

Isn't those times faster than any of the NVidia cards?

Forget nVidia for MW, it's a waste of compute resources.

MrS

But people are using CPUs for MW, and they much faster than CPUs.

---