The price doesn't even matter if the chip can't perform as good as, if not better than, intel's offering. Another question is the longevity of the AM3+ socket. Nobody is going to try saving money going AMD if the socket was to be discontinued in 6 months while 1155 would carry forward to IB.

Whats the point of using old software on your new 6 cores CPU?
you can skip the wall of text from below if you are not interested to read

Old software are ought to be abandoned... if they dont support new cpu tech.

And again reply to the bolded part, If the software already support multicore, theres prolly no need to rewrite since it will detect how many cores are available and will utilize as many as it can.

Just my noob opinion

Of course price matters especially when price/performance matters. I would definitely not pay $1000 for a BD knowing how would it perform. It is no where near in the range of the LGA2011 so I would say price here definitely is an important factor no matter what you pay for. The price that I have stated there take it with a grain of salt. That is why I labeled it 'rumor' because it has OBR all over it and my assumption here is that it should price similar or more expensive than Core i7 2600K.

It will definitely perform as good or better than Intel that I can be certain of but my assumption is only limited to heavily threaded applications and not single threaded applications. AMD would have to change it sooner or later and the very old socket is becoming of a bottleneck. It is either they make a switch now or never but I do agree that it would be a waste to have such a short lifespan. As for IB, nobody is 100% sure whether the Cougar Point chipsets would limit users or LGA1155 will still be used but Panther Point is a requirement to use IB.

The current programming model has poor multicore support and rewrite or recompile is necessary. That is why there are programming languages like OpenCL to simplify and make better use of multicore without diminishing returns.

Not all programs right? i have no knowledge of programing, but not all program benefits from using CUDA/AMD APP and OpengCL,
For example, the x264 encoder, though i stopped follow thier development very long time ago, untill now their devs never added any new stuffed eg the OpenCL, even CUDA.
Only CPU via POSIX thread support.

If the code is really well written there is probably no need to rewrite the codes..
There have been so many attempt to add OpenCL and CUDA support to x264 by other guys but i think no one success...
related: http://sites.google.com/site/x264cuda/comm...-with-x264-devs
and the quote

Many GPU accelerated h264 encoder can BEAT x264, in term of speed, many time faster than it but quality is far from acceptable.

Sorry if this is OT

It depends but more and more programs are getting on the bandwagon to utilize more cores as they possibly could despite the difficulty posed by making a program to be multithreaded. Definitely not all programs will benefit from OpenCL or could benefit from OpenCL but developers are getting there soon. OpenCL is quite a recent development and not as common as the C language but it is quite similar to C in terms of writing so I think it wouldn't be too much trouble for developers to learn it.

Llano and the whole Fusion concept will only become more popular if AMD manages to work with developers to push the development of GPU accelerated programs. They are trying and some software developers are making GPGPU accelerated software but if they fail then the Fusion platform is a flop.

If the quality is not on par, then they haven't beaten x264. It is possible to write a H.264 AVC encoder that can encode a 1080p movie in ten seconds, but all you see is a black screen.

Lol, correction for my own post, there arent many GPGPU h264 encoder around i think, mostly using the same tech, eg CUDA.
Tmpgenc, Badaboom ect, are using CUDA and intel.
Never saw AMD APP support yet, but AMD do have thier own Avivo encoder, same crappy quality i think.

---

Added on July 28, 2011, 1:51 amSadly the OpenCL Havok disappeared. currently i still dont see any program that using OpenCL.
Obviously game is the only software that can fully utilize APU.

other than that, this will probably make use of the entire APU:
Decoding video from MFT decoder(Media foundation, similar to Dxva, already exist but no there arent any simple way to use this decoder) -> Video encoding (CPU, x264) + Avisynth fft3dgpu GPU accelerated denoiser(example), -> output.

Method above will fully utilize the APU but its not beefy enough to do this, if the video is 1080p in and 1080p out.
the GPU inside APU only have 80 shader processor, consider my 4890 can hardly perform fft3dgpu 1080p denoising at real time.
80sp GPU will not cut it.

Gonna need bulldozer + 6870 inside the APU to perform that smoothly

This post has been edited by 1024kbps: Jul 28 2011, 02:01 AM

Last time GPUs are meant to push pixels but as time goes by we get GPUs that are becoming more and more multipurpose. That is what CPUs are today, they are multipurpose and no fixed function logic compared to last time and that is what makes modern CPU what it is today. We all know that AMD is pushing their GPUs to be morecompute intensive and more like a general purpose processor compared to pushing pixels, becoming much more like CUDA cores. GPUs these days are plenty powerful no thanks to console counterparts slowing down development so GPU compute is another market to tap.

CUDA was good but since it is proprietary to Nvidia, development is slow as with any closed source software. Make it open source like OpenCL then we can see development becomes faster. There are already AMD APP accelerated software, I forgot where is the link already but you can search through this thread, I have posted something about Cyberlink before. As with any programming language, it takes time before it becomes popular(if). We'll have to see the market's adoption rate towards GPU accelerated programs and web browser GPU acceleration is a good start.

wow, interesting if this is true.
not sure when komodo apu will come out.
hopefully am3+ socket can last longer.

hehe nice pricing indeed..
hope they hit intel performance benchmark push the pricing slightly backward.

This post has been edited by yinchet: Jul 28 2011, 03:42 AM

Follow the oBrbvsky blog test result
Performce is between i5 2500k and x6 1090t......
Abit sad on the result...

Do you know that many of us still use old software to run on the fastest CPU we can got? if you check this article u'll know that since Athlon64/Core 2 duo our single threaded performance has not increase much. That means that if u running on an Old software back in 2006 that only use 2 core, but it lag heavily because its high CPU req, you will still wont see huge improvement now even with Sandy bridge 2600K.Unless u overclock, which to some non-technical people they do not know how.

As for the software to take advantages of new cores, dma0991 has explained it well enough. It gets more difficult as the core count increase, unless they start using API like OpenCL. AMD/Nvidia has been pushing parallel processing, but CUDA/GPGPU has been around for many years, it is only now we start seeing common consumer product start benefit from it. It order to get most software support this it is gonna take even longer.

APU = CPU + GPU.
Game is the only software that fully utilizes APU (aka CPU+GPU)?
I don't think it's a right statement.

APU vs discrete CPU + GPU?
It depends on the price.
It is the price that determines the value.
If it has close price range, I'd pick discrete CPU+GPU anytime.
No shared memory. No shared cooling. Better headroom for overclocking & upgradability.

---

Added on July 28, 2011, 12:18 pmTake it with a pinch of salt.
We need to look at the real performance from retail product.
We need to look at the price to determine the value of this product.

---

Added on July 28, 2011, 12:22 pmIMO, it is not only about the utilization of cores.
It is about the efficiency of cores too.

More cores but inefficient implementation would give you bad performance nevertheless.
Highly efficient 4-cores might run faster than inefficient 8-cores.
That's my point.

By the way, APU = CPU + GPU.
Fusion optimized?
You mean AMD CPU + Radeon GPU optimized software?
Software developers have been doing it. I think it's nothing new.
i.e. CUDA vs AVIVO, it's already there.

This post has been edited by ALeUNe: Jul 28 2011, 12:37 PM

I don't know how would you define AMD having an inefficient cores compared to Intel. There never was a comparison between inefficiency vs efficiency. A single BD core does not take as much of a die space as a single SB core. If you were to compare die space which is a rough measurement of transistor count, a single SB core will be as big as a single BD module (2 cores) minus L3 cache. So therefore I can safely say that if a process can only use single thread, SB is a clear winner here but if it could use two, BD has an advantage.

Of course some might claim that it is just splitting transistors to make it two cores but if it is never that simple to do that. There are downsides of having more cores definitely but I'd rather get a SB that has 4 cores instead of using the same transistor budget for one huge SB core. There are diminishing returns if you want to make a processor more complex. Designing a smaller core and copy paste it 4 times is also much more economical than designing a single huge monolithic core.

What AMD wants with BD is that it is like Intel's HT but with HT you're looking like a <20% gain in a best case scenario. There is no replacement for physical cores and AMD's approach is to have two cores that perform 80% as good due and have them work in parallel. So in a two threaded workload, a module should be better than a single SB core with HT. I won't say that AMD's approach is good or bad but there will be a market where parallel workloads are important, mostly involving servers. The statement below is not written by me but it should give you a clear idea of what I mean.
Not Fusion optimized, Avivo is dealing with videos and what I mean by OpenCL or what AMD calls it AMD APP, you can use the GPU for just about any parallel workloads and it does not have to be video based only.

AMD used to charge USD$1000 for the FX-60 years ago. That is when they were highly profitable.

Based on the leaks which I would consider viable, I don't think so even in heavily threaded applications. I'll have to reserve judgement until some real benchmarks and reviews are published. Many things could have happen after that (e.g. bug fixes, clock speed fixes, etc)

Not true. Some compilers and programming APIs already support multi-core (including ICC). Intel is also trying to introduce MIC (Many Integrated Cores) to simplify porting of applications (easily by adding a few lines of pragmas and directives). However if you try to port applications to OpenCL or CUDA, that would require a major re-write. That is why the number of GPGPU applications for normal tasks are so few, while you will find most of them in specific HPC applications (mostly in CUDA).

Being highly parallel intensive computation and multithreaded is not the same thing. Similar to "C"? Not really, more of an "API". Have you seen how the codes look like?

As mentioned before, porting applications to GPCPU usually requires major re-writes, re-code and re-compile. That is why there is so few of them. Check True Fusion: AMD A8-3800 APU Review. Page 20: GPGPU Applications (seems to be the only review that test the very few GPGPU applications around).

Not all programs can be GPGPU accelerated, because GPGPU hardware still has many limitations since they work on fixed sets of data but not with dynamic data (those with lots of data dependency for example, such as ray-tracing which is still until today best done with CPUs).

CUDA is already highly popular in HPC, you can see that from the HPC hardware (majority HPC machines use Tesla, including the latest Cray supercomputer). Unfortunately OpenCL is still not that popular (probably due to having some platform specific APIs for different GPGPU architecture from each manufacturer, AMD have their own coding structure while NVIDIA have also their own coding when it comes to optimizing for fastest compute tasks). Then there's also DirectCompute, another recent competing API for GPGPU from Microsoft. Thus often programmers usually tend to stick to one already familiar API (such as CUDA). It also does seem NVIDIA Tesla hardware is more popular because it perform specific tasks better than AMD Stream (e.g. Folding@Home).

OBR the joker? IMHO take his results with lots of salt or ignore them altogether (after the stunts he pulled).

This post has been edited by lex: Jul 28 2011, 04:30 PM

They have fallen behind quite a bit but we cannot compare what most consumers are willing to pay nowadays since most are quite thrifty after the recession.

I don't have much info about compatibility of IB with the current Panther Point but most do suggest that there is compatibility but I'll wait till Intel gives the green light.

C language is not optimized for multithreaded so a rewrite to OpenCL is definitely necessary. They are few now but I'm sure they will be more popular as it gains momentum. I'll see how C++0x deals with parallelism.

I wish I have extra time to read through Nvidia's 61 pages and AMD's 142 pages of the OpenCL programming guide.

Give it some time. AMD is a small player and when Intel adopts OpenCL, we'll see a lot of OpenCL programs. In fact Intel has their OpenCL SDK already.

I agree that not all can be GPGPU accelerated but some workloads can be done by the GPU. Scaling for CPU performance with # of cores is pretty bad especially if the code written for it is not optimized for multicore.

Nvidia is the current leader for such standards no doubt about that. AMD is just starting with their HD7000 series or their current iteration of their GPU architecture is more compute based.



This post has been edited by dma0991: Jul 28 2011, 05:01 PM

The profit from just one of those is roughly equivalent to sales of 20 to 100 cheaper processors. Imagine even the sales of just a few of these high end CPUs.

Like I've mentioned before, some compilers are already optimized for multi-threaded (even simplified). Watch SGI Picks the Intel MIC Swim Lane for Exascale by 2018 from 5:32 onwards (this is ICC).

A quick look at OpenCL on Wikipedia, you can see what I meant.

Currently, Intel's implementation will be based more on AVX and future hardware (possibly with LRBNi extensions).

Scaling of CPU performance with increasing number of cores will depend on the hardware itself. For example a 2 socket systems scales better than 4 socket (this is well known, which is why most supercomputers are based on stacks of 2 socket systems). Then there's the MIC which scales very well because of its tight integration of simple x86 cores (no need for QPI or Hypertransport links).

Finding AMD Stream in HPC field is very rare as it is mostly dominated by NVIDIA Tesla. IMHO, Intel's MIC will be the next big player in this field.

This post has been edited by lex: Jul 28 2011, 05:51 PM

I won't know for certain how well Intel MIC will perform in real world yet but from my understanding, if a portion of a code is not parallel enough, it will be subjected to Amdahl's Law. Without the code being as parallel as the hardware, it will reach to a maximum number of cores and there is no performance gain afterwards. I am not sure how parallel is Intel's implemention with MIC but I think if the programming language from the ground up is made for parallel, it should be able to benefit the most. This is my opinion only though and might not reflect how well MIC will perform.

AVX lol, another 500+800 pages from Intel that I have not read. AVX is also implemented with BD. I do not know much about Larabee but from what I have heard others have said that it is not really awesome. Like I said, market dominance can change with time and since AMD is going towards that direction, Nvidia is going to have a little bit more competition than they used to have. Intel MIC, maybe.

I am quite curious why AMD Llano did not release a desktop 35w TDP version yet. Everybody know that Bobcat platform is great for HTPC, but itsnt great enough to be called perfect. no Blu-Ray 3D support

Intel SB 2100T is here already, all we need is a AMD LLano + better ATI driver to complete this.

This post has been edited by billytong: Jul 28 2011, 06:55 PM

AMD Confirms Pricing of Eight-Core FX Processors: Approximately $300