Performance of alpha versions has been pretty lacking compared to what's possible on nVidia hardware. Folding@Home is facing a similar problem. The current ATIs are hard to utilize effciently. GCN is changing this, it should be approximately equal to Fermi in terms of flexibility. If any ATI can be useful here it's GCN.

Rumored launch is in 4 days, or maybe the 9th january.

MrS

---

Scanning for our furry friends since Jan 2002

We simply have not resources to take on the debugging on all VLIW4 AMD cards. The present and the future is GCN. Then, if it happens a miracle and the stability of VLIW4 is good, then we can also release for that. But again, running gpugrid on VLIW4 cards is very inefficient.

gdf

As well as VLIW5, I take it?

MrS

---

Scanning for our furry friends since Jan 2002

GCN is now 1D Shader or like some say 16D? :confused

---

maybe opencl can open a door and could be easier to handle within a timescale ??

---

http://www.rechenkraft.net

if you like to "fold" some aminoacids/proteins with your GPU, you now can try POEM@Home and their new open CL app.
A the moment onlx Linuy is supported, Windows will follow. All ati-cards from the 4000, 5000 and 6000er serie are supported.

I'm still hoping to get some wu's for my AMD-Cards ...

Was ist neu im AMD APP SDK V2.6?

Key features supported in SDK 2.6 and the Catalyst 11.12 drivers include:

OpenCL runtime integration into Linux and Windows® Catalyst drivers.
Inclusion of the Khronos C++ wrapper API.
Multi-GPU support on Linux platforms.
PX5 support.
Preview: Support for AVX instructions on CPUs that support AVX.
Support for FMA4 instructions in OpenCL built-in function libraries on CPUs that support FMA4.
Kernel reflection, query kernel parameters, and enable use of OpenCL kernels in data-driven applications.
Support for atomic counters on APUs.
Redesign of OpenCL run-time on CPU, significantly improving performance.
Support for the cl_amd_media_ops2 extension, exposing hardware capabilities for accelerating image-related processing.
Async copies preview (set environment variable GPU_ASYNC_MEM_COPY=2 to enable).


The OpenCL 1.2 preview includes the following capabilities (requires 8.93.10 preview drivers):

Host access flags for memory objects enables more efficient buffer handling.
Pattern-based GPU buffer and image initialization eliminates need for certain buffer/image transfers.
Memory objects migration supports early transfer of buffers in preparation for when they are needed.
New generalized image creation API.
Enhanced image/buffer map operations.
OpenCL 1.2 CPU device partition, including partition of a CPU after addition to a context.
Generalized 1D and 2D images, image arrays, and image <--> buffer interop.
The 8.93.10 preview drivers also enable use of the static C++ kernel language.


gDEBugger version 6.1 is a major improvement in performance and robustness over version 6.0.
It can be downloaded for use with this SDK from http://developer.amd.com/gDEBugger.

Integrated with Microsoft® Visual Studio
Release Notes


APP KernelAnalyzer v 1.1:

Support for AMD Radeon™ HD7000 series GPUs (compilation only, no analysis).
Support for Catalyst revisions through 11.11.
Support for compiling kernels with the installed driver (select Installed Driver under the CAL version in the Options panel).
Format and Target Object Code are now separated.


APP Profiler v2.4 includes several key new features, including:

A kernel occupancy analyzer that estimates, for each kernel dispatch, the number of in-flight wavefronts on a compute unit as a percentage of the theoretical maximum number of wavefronts that the compute unit can support. In addition to reporting the occupancy percentage, the profiler can display a report that can help the developer achieve a higher occupancy percentage.
The ability to navigate from the API trace to the source code that called an OpenCL API.
Improved OpenCL API analysis that provides performance suggestions to the developer.
The ability to filter which OpenCL APIs are traced.
Several UI enhancements, including the ability to rename sessions from the Session Explorer Window, and the ability to automatically delete Profiler sessions when closing a Microsoft Visual Studio solution.
Preview: Support for profiling with AMD Radeon HD 7000 series GPUs (requires AMD APP SDK v2.6 and an AMD Catalyst version that supports this hardware).

Hope that helps!

if you like to "fold" some aminoacids/proteins with your GPU, you now can try POEM@Home and their new open CL app.
A the moment onlx Linuy is supported, Windows will follow. All ati-cards from the 4000, 5000 and 6000er serie are supported.

I'm still waiting for the windows version and the gpu version of hcc over at wcg.
The system requirements on this project are too high.
I'm out.

I'm still waiting for the windows version and the gpu version of hcc over at wcg.
The system requirements on this project are too high.
I'm out.

I wait also windows app. :-)

I'm about to look up the HD 7970 and check if it fits my rather tight power limit requirements, or if I need to wait for a later board in that series that does.

POEM did it, and you ?

Expect to pay over $500 for the 7970, at least right at first anyway ...

GDF made the decision that GPUGrid will only try to use GCN based AMD GPU's due to the relatively poor performance of existing AMD models (up to and including the 6000 series), and presumably logistical reasons. Basically, this will limit any GPUGrid app to the high and probably mid range 7000 series AMD GPU's; Radeon HD 7970, 7950, 7990 and probably 7870 and 7850. I doubt if the intermediate 7770 and 7750 GPU's would be usable (depends on performance/GCN). Obviously testing can't happen until the GPU's have been released and the researchers have a GPU to test. STEVE pointed out the high costs, 'over $500 for the 7970'. Good point, there might not be a massive demand for an AMD app. I suspect cards might trickle out in pairs rather than all get released at once and the 7990 is not due until March.

As pointed out you can now use your old and present AMD GPU's at POEM. 4000 through to 6000 series.
Requirements are Boinc Beta version 7.x and I think Catalyst 11.x drivers. Works for Linux and Windows and is OpenCL.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

HI,
just to specify we still have not tried any GCN based card. We expect to be good, but don't have any number, nor we know if it works at all. We have amd to send us one, let's see.
The code is ready, so as soon as we get one in, we should be able to send out the numbers.

Looking at the numbers on compute published by some sites, I still expect kepler to be faster, but probably quite close.

gdf

GDF made the decision that GPUGrid will only try to use GCN based AMD GPU's due to the relatively poor performance of existing AMD models (up to and including the 6000 series), and presumably logistical reasons. Basically, this will limit any GPUGrid app to the high and probably mid range 7000 series AMD GPU's; Radeon HD 7970, 7950, 7990 and probably 7870 and 7850. I doubt if the intermediate 7770 and 7750 GPU's would be usable (depends on performance/GCN). Obviously testing can't happen until the GPU's have been released and the researchers have a GPU to test. STEVE pointed out the high costs, 'over $500 for the 7970'. Good point, there might not be a massive demand for an AMD app. I suspect cards might trickle out in pairs rather than all get released at once and the 7990 is not due until March.

As pointed out you can now use your old and present AMD GPU's at POEM. 4000 through to 6000 series.
Requirements are Boinc Beta version 7.x and I think Catalyst 11.x drivers. Works for Linux and Windows and is OpenCL.

You're wrong.
http://en.wikipedia.org/wiki/Comparison_of_ATI_graphics_processing_units#Southern_Islands_.28HD_7xxx.29_series
The only cards that will support GCN will be 79x0. Just like 69x0 were the only ones from the 6xxx series that supported VLIW4.

From my point of view, if poem did it, if folding did it, if wcg (most probably) did it, why couldn't gpugrid not do it?
I would have preferred to have a low gpu utilization just like over at folding, instead of not heaving any app. The guys from SETI faced the same problem, but they bypassed by running two applications simultaneously.
I actually would have wanted a low utilization since I would have let the app running non-stop while also doing other stuff on the computer.

All I can say is ...¡Feliz Navidad! and see you in 1 year+, when the new fusion series is release which, hopefully, will have GCN on it.

Probably doesn't mean definitely. So no GCN for 7800 series, fair enough. That changes the picture, for me. In Jan there would only be two very expensive GCN cards. Don't know how worth while it would be supporting a project with say ten 7900 GPU's attached?

As for that wiki page, I have my usual doubts about it's accuracy.

My take on things is that the initial 7000 release will use GDDR5, and not XDR2?
If so, and if future cards will use XDR2 then is it also possible later editions of the 7800 will be released with GCN?
Will the 7800's use VLIW5, just like the present (40nm) 7600M cards, and will it be any better than VLIW4 and VLIW3?

Was the AMD app ever tested on VLIW4 and VLIW5 GPU's?

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

We never tested on VLIW4. Yet, I would expect marginal improvements compared to VLIW5. It's not just a problem of VLIWX, there are many more problems in those cards for compute.

As I said as long that older cards are stable, we can support them. What we cannot support is too many generations because each one comes with their bugs. The more bugs there will be, the less we can support. A VLIW5 card would probably be 8 times slower than Kepler (consider the save in electricity).

gdf

Well, if this should or will indeed be the final verdict here, then this will be it for me! If that is all, you were able to conclude in around two years, then this is disappointing and poor!
I have two fairly newish AMD Radeon HD 6990/ATI Radeon HD 5970 cards here, that of course do well on other projects, but will supposedly not here...

A project that thinks its volunteers can afford high-priced high-end cards every year, is severely mistaken, has lost it and obviously is living in a dream world! Why not simply making the prerequisite of a working quantum computer, if you want to sort out only wealthy gals and guys? I do not intend to spend countless €s on new state-of-the-art GPU hardware every year just to be able to run your project!

And btw regarding that there has allegedly not been enough time to take care of this matter earlier: was the new website REALLY that important for the project or the science? The old one did it for me personally quite well IMHO.

Sorry, this becomes frustrating and unnerving! Every new post in here makes one think: "Oooh, now there is good news!" and the news just go from bad to worse!

Have a nice Christmas though and I hope Santa is putting a thousand brand new, top model AMD Radeon HD 7990 under your Christmas trees even before release date, or however the enthusiast card will be named then! ;)

Ban me, if you must, for I am beginning to not care anymore!

---

I'm far from disappointed or frustrated; I was holding onto an HD5850 in the hope I could use it here, but now that POEM has launched an AMD GPU app, I am delighted I can use it there on a Bio-science project, somewhat similar to this project. Even if GPUGrid launched the AMD app, I don't think I would want to use it here - well not if it would only contribute 1/8th the performance of a comparable NVidia GPU, and we are talking about a 2088GFlops GPU performing like a GT240, and not if the performance at POEM is 4 or 5 times better. Performance is very much down to the research requirements, and GPUGrid does some of the most complex GPU research anywhere.

You have a GTX 480 running here, and doing good research. Be happy with that, and use your AMD GPU's at POEM.

GDF and the other researchers/developers have to make decisions like this. They have limited numbers and resources, so if they feel now is not a good time to launch an AMD app (due to poor performance, numerous bugs over several generations, and other projects) then I can live with that. The time and effort required to facilitate and support a poor performing app is probably not worth it. Kepler is also coming out soon, and the team has to prepare to facilitate those cards, and they will probably be twice as fast at crunching.

A lot of effort went into testing both a new server system and site. Other ongoing considerations include new apps, and better recognition systems. The site definitely needed an overhaul and in my opinion has resolved many issues and made many significant improvements. The server upgrade simply had to be done.

I will not be rushing out to buy a 7950, and I would not suggest anyone else does - they will be too expensive, for me, and who knows what the performance will be like. Six months or a year from now, when we know what the performances are and cost relative to comparable NVidia cards, then perhaps I will get one, or perhaps I will get one for another project, POEM, or possibly WCG.

PS. Nobody is getting banned for speaking their mind. If you say it like you see it, keep it clean, and don't attack anyone, I have no issues. Throw a wobbler and crazy posts just get deleted, not your account. Have a good Christmas.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Hi,
let me clarify because I see that there are a lot of people that would like to use their cards and fear that they cannot. I am not saying anything new and we are not working against you, but with you.

1) Bad, well known news. The VLIW5/4 AMD cards are very powerful cards and usually cheaper than nvidia, also the top cards beat nvidia in terms of raw performance. However, only very numerical intense algorithms can benefit. For gpugrid, a top VLIW5 card consumes almost like a gtx580 but has a performance of a gt8800, a card of several years ago which is not recommended. It's not a decision, there is nothing we can do about it and we tried a lot. Other algorithms are better with this and electricity costs 3 times the card.

2) Good news, to be confirmed. In the future, new Amd cards based on GCN should level up with Nvidia fermi/kepler. We could not check it but all the indications are good. So take this with a level of uncertainty. Expect to have an equal representation of nvidia and amd at gpugrid in the future. This will be great because half (or more) of the world cannot crunch for us now. In the past few months we have spent plenty of hours to prepare for AMD.

Of course, don't go and buy a new card because of this! The HD7950 is expensive because now it's the fastest card worldwide. It will cost 100USD or less as soon as nvidia wakes up with kepler. There will be other reduced GCN cards also cheaper and less power hungry. According the news websites, AMD will put out all the new cards within 6 months, so there should be plenty.

We only have a single VLIW5 card now. That's all. Maybe soon a GCN card. We will test on those. If it works, meaning it is really stable, the application will go out. If you want to use it on VLIW5 do it, but your are not using it for its capabilities here and there is nothing, absolutely nothing we or amd can do about it.

Enjoy holidays and merry Christmas.

gdf