Some guys noticed that amd ran an OPENCL BENCHMARK with FMA4 INSTRUCTION SET.
http://www.donanimhaber.com/islemci/...onuclari_8.htm
My knowledge telling me, accelerate opencl with any instruction set is MIRACLE.
Printable View
Some guys noticed that amd ran an OPENCL BENCHMARK with FMA4 INSTRUCTION SET.
http://www.donanimhaber.com/islemci/...onuclari_8.htm
My knowledge telling me, accelerate opencl with any instruction set is MIRACLE.
Please, can you elaborate?Quote:
Originally Posted by undone
Maybe he does not know OpenCL can be run on CPU too...Quote:
Originally Posted by Lokinhow
well the way i see it if its only just beating a 4 core chip like the 2500k in multithreaded tasks with its 8 cores then its performance vs 2500k in games that only make use of 1-4 cores will be about halfQuote:
Originally Posted by jimbo75
the only way those charts could turn out to show bd is better at gaming is if the reason it beats the 2500k and not the 2600k is due to the extra 4 cores scaling much the same or worse than ht over the first 4 cores but this seems unlikely
so im realy hopeing those charts are a load of crap or this cpu may be another dud
that chart showing some odd games performance suggests for some resson crossfire is working better on the amd platform than intel in some cases
if that is the case it may still do ok in games that are gpu bottlnecked when running crossfire and at the moment these games do outway the number of games that are cpu bottlnecked so this could be a fairly good thing
it's a 4 core chip with CMT really, though AMD decided it would be cool to label it an 8 core its more like a 4 core with better scaling than HT.Quote:
Originally Posted by dasa
I saw 5x scaling in VR-Zone cinebench leaks, did anyone else not?
before your edit you made mention of us being lost and really with all the crap thats been posted about bd over the last year+ is it suprising?
hopefully we will soon get a heap of real reviews we can pick threw and become a little less lost
Ouch. 200 (~$270usd) euros? In the US $249 USD can buy a 2600k @ Microcenter.Quote:
Originally Posted by Hornet331
As for the benchmarks... I find it odd that my 3 year old i7 860 @ 3.6ghz scores 6.23 on Cinebench 11.5.
It's a brand new architecture, a major change that rivals that of Thunderbird core from the K6IIs. It will have some hickups but it has a long way to go before it is saturated, expect with every stepping and every new revamp a lot of improvements. This is just the beginning.Quote:
Originally Posted by dasa
Phenom II 1090T does 7.24 at 4.3Ghz and 8.63 with 4.5Ghz, some tweaking and easy 9+, that is 24/7 machine crunching at the moment at WCG.Quote:
Originally Posted by Andrew LB
Im just waiting for end of... to release some benchmarks on Zambezi, no idea if its same as retail sample.
Yes, I did...though I came off quite rude so the edit was necessary.Quote:
Originally Posted by dasa
But yes, there is so much misleading info and so many numbers pulled from thin air it's rediculous. People like you (no offense!) and others are trying to read all these made up numbers and make sense out of it.
For example we had people talking about how low BD single thread performance is by dividing Cinebench by 8.
I think those leaked slides from AMD actually indicate good performance for zambezi bulldozer. I don't see there being a performance or price problem for AMD.
the problem is that they are a year late. the x58 system compared in the slides is about to be replaced. the current sandy bridge chips are 6 months from being replaced.
I don't question anymore if AMD will compete well with bulldozer. I question if they will compete well for longer than a few months. ONLY time will tell. the wait for piledriver begins.
Quote:
Originally Posted by BeepBeep2
none taken i can agree with that :)
Is anybody ever running any "normal" benchmarks anymore?
I mean if I want to link a 200 object file C++ shared library I use no MMX or SSE instructions, I don't multithread, I just wait like an idiot. I think it's sad that pretty much all the benchmarking we have seen for Bulldozer and Sandy Bridge before is just deliberately pointing at a single trick this pony does better and then brings forward performance numbers that are completely unreproducible for anything anybody in the real world does.
Just running cinebench single-threaded would be a huge advantage. As has been pointed out, dividing a result by number of cores in a CPU as complex as bulldozer with all kinds of shared caches, buses and other resources, and god knows how each of these access RAM under what conditions, is nothing short of useless.
Exactly my point. Don't those scores seem strangely low considering the chip has 8 cores. I would have expected scores at least 30% higher when compared clock/clock with a Phenom x6 considering the supposedly better architecture and 2 extra cores. If it were only a single leaked slide, I probably would have said they were fake, but the quantity of slides leads me to believe they're legit.Quote:
Originally Posted by rintamarotta
Run on CPU but with no instruction set support, otherwise it won't be called OPENCL. But I could be wrong, compiling opencl with any instruction set seems impossible.Quote:
Originally Posted by Mechanical Man
No, that OPEN just means that the standard is OPEN, anyone can write implementation of it. That would be just very stupid that you could not write software with instructions giving lot of performance.Quote:
Originally Posted by undone
Limited GPU :rofl:
http://img546.imageshack.us/img546/7190/30396861.jpg
http://img706.imageshack.us/img706/1104/59666881.jpg
HD 6870 AND 1920X1080 LOL :D
yeah, looks like AMD was taken over buy some 8 year old kid :( IF those are real ;)
One thing is use 6870 GPU, but another thing is using $170 GPU with $1k CPU. I would have thought that kid buying a system with $1k CPU would afford to get himself two 6990s or 590s :/
AMD has it's own OpenCL ICD for both GPU's and CPU's. This means they can and already did implement Bulldozer XOP and FMA4 extensions to extract more performance of the CPU path.Quote:
Originally Posted by undone
OpenCL ICD does exactly same job as OpenGL ICD, catches OpenCL commands and translates them to target numerator being it optimized CPU path or GPU path.
Hope this helps.
Well its was clear form the first slide that it was gpu limited.. but using a HD6870.. come on... also using doubled priced memory for the intel system.. wut?Quote:
Originally Posted by cold2010
then i don't really understand how the cinebench score can be so low, when the Wprime 32 a lot faster of the 2600K..
The WPrime score seems low for the 2600K, the % difference between the 2500K and 2600K should be 40+%:Quote:
Originally Posted by Lanek
http://www.bit-tech.net/hardware/cpu...ridge-review/7
http://www.hardocp.com/article/2011/...ssors_review/3
If you want to test the CPU on heavy loads, thats how it is done. You need to cut FPS numbers with a slower GPU on areas where FPS are high (where there isn't much work done to process the image)Quote:
Originally Posted by cold2010
and the slide clearly says to spend more on graphics with the money you save, so they couldnt max that outQuote:
Originally Posted by gosh
:D:rofl::DQuote:
Limited GPU
Attachment 120499