[quote] Another thing that discourages me is that HA-Soft said the best configuration is a CPU with the AVX 2.0 instruction set extension. If I understand him correctly he is saying AVX 2.0 will complete an A@H task faster than a Titan. Well, I'll be ordering a Haswell with AVX 2.0 fairly soon and if it turns out faster than a GTX 670 on current A&H tasks then I see no reason to do a risky hack on an expensive video card.

No, it's not faster. GPU Titan is 3x faster than CPU Intel Haswell, but CPU is more efficient. CPU can run 4 task in parallel (4 cores) so 3 tasks GPU vs. 4 tasks CPU. Power consumption/cooling is better for CPU also and of course CPU is cheaper than GPU.

Thanks for clarifying. I guess now I should look into that report from BeemerBiker that someone benchmarked DP on a hacked 670 and found it was no faster than before the hack.

---

BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux

New updated gpu app has been released.

Cool :), what are the differences? Have you got a version history page explaining that?

Btw, I just had a thought, with the discussion going on here saying that A@H on GPUs uses DP, why has development started on NVidia cards rather than AMD??
More of the AMD/ATI cards have decent DP power.

---

Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7

Possibly because CUDA is much easier to work with they say. OpenCL is said to be cumbersome and generally hasn't been well received by programmers, at least that's the sense I get when reading about CUDA vs. OpenCL from many different sources around the web. CUDA and the NVIDIA architecture were designed and built for each other therefore you can expect there to be fewer compromises. OpenCL was designed to work with a number of existing architectures (otherwise it would be senseless to make it open) so there are necessarily compromises. Compromises negatively affect user friendliness and therefore adoption rates. There is also a perception that CUDA is the platform that gives the greater performance to most massively parallel algorithms with the least effort.

I don't program with either (CUDA or OpenCL) so bear that in mind. I'm just repeating what I read from sources I think know what's going on. H.A.-Soft's reasons for going with CUDA/NVIDIA first may not reflect my opinion.

Also, AMD has crippled DP on their latest cards. If they continue to do so in the future their DP advantage will disappear which decreases incentive to learn the more difficult OpenCL.

---

BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux

Cool :), what are the differences? Have you got a version history page explaining that?

New app is about 20% faster. No other changes.

Btw, I just had a thought, with the discussion going on here saying that A@H on GPUs uses DP, why has development started on NVidia cards rather than AMD??
More of the AMD/ATI cards have decent DP power.

Simply because I know CUDA so I was much faster with CUDA. And yes CUDA framework is much much easier to learn and use.

GTX 650 Ti GPUs processing new GPU tasks: good to see the GPUs working on this project.

John

OK - driver update fixed this: I had never updated drivers as my PC builder warned me against doing such....



AS far as I can see neither of my GTX 650 Ti GPUs is processing any GPU tasks. Do I need to do anything special to get such tasks?

Thanks

Your PC downloaded new tasks for GPU. So there are in queue.

Just processed two of these and I can see the power of CUDA at work: average time has dropped over 25% with CPU sec falling from 7,900 sec to 8 sec!!

OK - driver update fixed this: I had never updated drivers as my PC builder warned me against doing such....



AS far as I can see neither of my GTX 650 Ti GPUs is processing any GPU tasks. Do I need to do anything special to get such tasks?

Thanks

Your PC downloaded new tasks for GPU. So there are in queue.

Nice improvement! :)

HA Soft
Thx for reply, & fair enough! :)

Dagorath
Interesting, & sad to see that programmers are still put off from ATI/AMD cards :(.
And I see what you mean about recent AMD cards, bar the R9 280s (which apparently are virtually the same GPU as the 7900s) which has decent DP, the other cards DP are much weaker. Nothing new for the low end cards but it is something new for the top end! Although AMD are still much better than NVidia cards (bar the Titan which is stupid money).

Why on earth are AMD holding back DP power??
It's crazy that their 2nd from top current card (R9 290) is only 11% faster than their over 4yr old 5870 (2nd from top 5000 series card) in DP! Even the R9 290X is only a modest ~29% faster in DP than the 5870.

---

Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7

Nice improvement! :)

HA Soft
Thx for reply, & fair enough! :)

Dagorath
Interesting, & sad to see that programmers are still put off from ATI/AMD cards :(.

They're not put off from AMD, they're put off from OpenCL. Actually, it's probably more accurate to say they like CUDA better and since ATI/AMD doesn't support CUDA it kind of blows back on ATI/AMD. Difficult and clumsy (2 words often used to describe OpenCL) are perceptions in peoples' minds. Perceptions change; programming languages, libraries and platforms evolve. CUDA has been around longer which tends to make it more familiar and comfortable. Instructors are more familiar with it and that influences student perceptions. In time, when instructors grow more familiar and comfortable with OpenCL, their attitude will be passed on to their students.

And I see what you mean about recent AMD cards, bar the R9 280s (which apparently are virtually the same GPU as the 7900s) which has decent DP, the other cards DP are much weaker. Nothing new for the low end cards but it is something new for the top end! Although AMD are still much better than NVidia cards (bar the Titan which is stupid money).

If you believe AMD is better than NVIDIA then you likely believe Ford is better than Chevy too. BTW, did you know that here in Canada Ford trucks have a heated tailgate option? It's so your hands don't get cold when you push them ;-)

Why on earth are AMD holding back DP power??

For the same reason NVIDIA holds back on DP and for the same reason South Africa holds back on diamonds... to create artificial scarcity to buoy the price up. NVIDIA has been doing the same thing for even longer than AMD and the deBeers family (diamonds) and OPEC (oil) have been doing it even longer than NVIDIA. NVIDIA's Tesla cards have huge DP power but if you think Titan is stupid money wait until you check Tesla prices. IIRC the entry level card is ~$2,000 USD and the high end one is $5K?

It's crazy that their 2nd from top current card (R9 290) is only 11% faster than their over 4yr old 5870 (2nd from top 5000 series card) in DP! Even the R9 290X is only a modest ~29% faster in DP than the 5870.

It's crazy from your perspective but your perspective doesn't include a view of the money they pour into R&D to achieve performance gains. Is what they do fair? Is what they charge fair? There is no such thing as a fair price. There are 3 prices: the price the seller would like to get, the price the buyer would like to pay and the price the 2 parties finally agree on somewhere in the middle.

---

BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux

Nice improvement! :)

HA Soft
Thx for reply, & fair enough! :)

Dagorath
Interesting, & sad to see that programmers are still put off from ATI/AMD cards :(.

They're not put off from AMD, they're put off from OpenCL. Actually, it's probably more accurate to say they like CUDA better and since ATI/AMD doesn't support CUDA it kind of blows back on ATI/AMD. Difficult and clumsy (2 words often used to describe OpenCL) are perceptions in peoples' minds. Perceptions change; programming languages, libraries and platforms evolve. CUDA has been around longer which tends to make it more familiar and comfortable. Instructors are more familiar with it and that influences student perceptions. In time, when instructors grow more familiar and comfortable with OpenCL, their attitude will be passed on to their students.

And I see what you mean about recent AMD cards, bar the R9 280s (which apparently are virtually the same GPU as the 7900s) which has decent DP, the other cards DP are much weaker. Nothing new for the low end cards but it is something new for the top end! Although AMD are still much better than NVidia cards (bar the Titan which is stupid money).

If you believe AMD is better than NVIDIA then you likely believe Ford is better than Chevy too. BTW, did you know that here in Canada Ford trucks have a heated tailgate option? It's so your hands don't get cold when you push them ;-)

Why on earth are AMD holding back DP power??

For the same reason NVIDIA holds back on DP and for the same reason South Africa holds back on diamonds... to create artificial scarcity to buoy the price up. NVIDIA has been doing the same thing for even longer than AMD and the deBeers family (diamonds) and OPEC (oil) have been doing it even longer than NVIDIA. NVIDIA's Tesla cards have huge DP power but if you think Titan is stupid money wait until you check Tesla prices. IIRC the entry level card is ~$2,000 USD and the high end one is $5K?

It's crazy that their 2nd from top current card (R9 290) is only 11% faster than their over 4yr old 5870 (2nd from top 5000 series card) in DP! Even the R9 290X is only a modest ~29% faster in DP than the 5870.

It's crazy from your perspective but your perspective doesn't include a view of the money they pour into R&D to achieve performance gains. Is what they do fair? Is what they charge fair? There is no such thing as a fair price. There are 3 prices: the price the seller would like to get, the price the buyer would like to pay and the price the 2 parties finally agree on somewhere in the middle.

You mis-understood my post about my comment of AMD vs NVidia, it was in the context of the paragraph so I was only referring to DP power, as in AMDs DP power is still better than NVidias (bar the Titan).
I'm well aware their is rarely if ever a time where either AMD or NVidia is in total, better than the other :).

Lol on the Ford crack, you know Ford stands for Fix Or Repair Daily? ;), actually Fords are ok for the most part, well at least Euro Fords anyway (I'm a mechanic btw).

Not just me who would think the small DP power increase is crazy, but fellow MW crunchers & other DP users, don't you? ;). Alas we are an iddy biddy tiny segment of the market.

Anyway, better not bring the thread further OT

---

Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7

You told me you are a mechanic and I remembered and knew you'd like the Ford crack. They're good trucks.

Is the small increase in DP crazy? Depends what you mean by crazy.

Saying it's crazy is expression of a personal opinion. I don't have and never will have a personal relationship with AMD or NVIDIA. We have a business relationship and that's as far as it goes. They can do whatever they want with their products. I think between Intel, AMD and NVIDIA there is enough competition in the market to guarantee that if one starts to gouge consumers the other 2 will take advantage of it and the gouging will stop.

---

BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux

Not just me who would think the small DP power increase is crazy ...

Do you understand that DP power is artificially crippled on 'consumer' (gaming) chips?
In fact this artificial cripple is enabled not on the chip itself but on the PCB of the video card using resistors.
(we discussed this with Dagorath in PMs a few months ago)

Read how to unleash the DP power by hard-mods:
http://www.eevblog.com/forum/chat/hacking-nvidia-cards-into-their-professional-counterparts/

http://www.guztech.nl/wordpress/index.php/2013/11/researching-nvidia-gpus-geforce-gtx780-and-gtx-titan-to-tesla-k20-and-tesla-k20x/


I wonder if some Hardware company (e.g. in China) will not see the opportunity to do these hard-mods in a factory for a small fee (e.g. $20-50)
(you send card, they mod and send back) and/or resell pre-modded cards.
(such company may already exist, I didn't search)

---

- ALF - "Find out what you don't do well ..... then don't do it!" :)

Not just me who would think the small DP power increase is crazy ...

Do you understand that DP power is artificially crippled on 'consumer' (gaming) chips?
In fact this artificial cripple is enabled not on the chip itself but on the PCB of the video card using resistors.
(we discussed this with Dagorath in PMs a few months ago)

Read how to unleash the DP power by hard-mods:
http://www.eevblog.com/forum/chat/hacking-nvidia-cards-into-their-professional-counterparts/

http://www.guztech.nl/wordpress/index.php/2013/11/researching-nvidia-gpus-geforce-gtx780-and-gtx-titan-to-tesla-k20-and-tesla-k20x/


I wonder if some Hardware company (e.g. in China) will not see the opportunity to do these hard-mods in a factory for a small fee (e.g. $20-50)
(you send card, they mod and send back) and/or resell pre-modded cards.
(such company may already exist, I didn't search)

As bitcoins goes more mainstream and the mining does too, maybe DP will be used there and then those mods will be VERY cost effective.

Mining do not use floating point calculations.

Mining do not use floating point calculations.

Ah, thanks I did not know that.

New version of GPU app has been released.

Mining do not use floating point calculations.

Yes it is all INT (integers). No FP32 (single precision) or FP64 (double precision).

New version of GPU app has been released.

- Added support for nVidia Maxwell cards (compute capability 5.0)
- Changed grid size to decrease screen lag on slower cards (GT640...)