Amd bulldozer or sandy bridge 2600k to buy

[johns0]

I'm deciding on which system to upgrade to,looks like the bulldozer series will be delayed in coming out till september so i will have to wait 5 months or more and hope the cpus will be that good or go for the sandy bridge 2600k now.

[jagabo]

And 5 months from now Intel will be a few months from releasing their next generation. So you'll have the same dilemma then.

[aedipuss]

the unlocked version of the 2600k is pretty amazing...

[johns0]

I ordered the 2006k,i upgrade so often it isn't a dilemma for me.

[deadrats]

I'm deciding on which system to upgrade to,looks like the bulldozer series will be delayed in coming out till september so i will have to wait 5 months or more and hope the cpus will be that good or go for the sandy bridge 2600k now.

where did you hear that? last i heard it was due out june 7th...

[johns0]

http://www.brightsideofnews.com/news/2011/5/30/amd-delays-bulldozer-citing-performance-issues.aspx

[deadrats]

wow, i had been seeing reports from a number of sites that claimed that bulldozer wasn't all it was cracked up to be ("man with a shovel", that's a great line) but it's disappointing to hear that amd willbe pulling another "barcelona".

maybe amd should fire all it's engineers and product managers and go way old school: drop support for all the fancy extensions, such as avx and fma4 and instead just go with a 20 core cpu running at 1 ghz with butt loads of L2 cache (maybe 50+ mb), custom engineer alu's and fpu's that are 512 bit's in width, are programmed in traditional x86 code but have an internal scheduler that allows the alu's and fpu's to behave like traditional simd units, where 1 instruction can operate on multiple pieces of data simultaneously; in other words the hardware automatically vectorizes the code.

or they could release bulldozer as it is but release a specially coded driver, like gpu manufactures do, that is in reality a run time compiler that is capable of rearranging instructions so as to get maximum performance from bulldozer.

[deadrats]

I ordered the 2006k,i upgrade so often it isn't a dilemma for me.

since you're getting the 2600k, how about running a few tests with the latest media coder build, comparing the encoding speed and quality of x264 with intel quick sync?

[johns0]

I'll give that a try,just hope i get my cpu and mb before the canada postal strike.

[jagabo]

maybe amd should fire all it's engineers and product managers and go way old school: drop support for all the fancy extensions, such as avx and fma4 and instead just go with a 20 core cpu running at 1 ghz with butt loads of L2 cache (maybe 50+ mb)

Silicon real estate isn't free.

[deadrats]

maybe amd should fire all it's engineers and product managers and go way old school: drop support for all the fancy extensions, such as avx and fma4 and instead just go with a 20 core cpu running at 1 ghz with butt loads of L2 cache (maybe 50+ mb)

Silicon real estate isn't free.

amd's gpu's have up to 3072 stream processors; now i understand that x86 cores are different from gpu cores, but perhaps they could borrow a page from their gpu division and perhaps make a many core cpu with existing fabrication techniques.

[jagabo]

And lose their existing customer base (x86)?

[aedipuss]

the problem is real estate. cpu cores can only be so big, and adding additional cpu cores to a die also requires adding memory(caches), interconnects, and controllers. not likely to see huge gains all at once, moore's law still rules.

[deadrats]

And lose their existing customer base (x86)?

here's the thing; amd has been saying for a while, since at least 2009 when they first announced "bulldozer" that eventually they planned to introduce a cpu with a fully integrated gpu that handled all floating point math (bulldozer was originally planned to be the first cpu that did away with the fp registers and instead used the gpu to perform x87 math).

keeping that in mind and remembering that transmeta has had a cpu that can run any instruction set, including x86, it's not outside the realm of reality to think that amd could license their technology and make a processor based on existing radeon architectures; i.e. uses simpler risc-like gpu cores rather than existing x86 cores and still be able to run existing x86 software, think something along the lines of a gpu accelerated virtual machine.

the reality is amd desperately needs to do something to stay competitive with intel, the days of when the k6-2 was a compelling processor or when the K8 came, saw and conquered are long gone; there's no question that if money is no object intel cpu's are the way to go, amd has some great technology in their gpu division, they need to start tapping it in creative, unconventional ways.

[jagabo]

keeping that in mind and remembering that transmeta has had a cpu that can run any instruction set, including x86,

It ran ~1/20 the speed of a similarly clocked x86. That's why it was a flop. And Intel has already tried this with Itanium. Intel originally claimed Itanium would eventually replace x86 but it's x86 performance was similarly abysmal. Intel gave up on the idea of replacing x86 with Itanium. The i960 and i860 risc CPUs before it were flops too.

it's not outside the realm of reality to think that amd could license their technology and make a processor based on existing radeon architectures; i.e. uses simpler risc-like gpu cores rather than existing x86 cores and still be able to run existing x86 software, think something along the lines of a gpu accelerated virtual machine.

x86 emulation is probably one of the least vectorizable or parallelizable problems.

[deadrats]

It ran ~1/20 the speed of a similarly clocked x86. That's why it was a flop. And Intel has already tried this with Itanium. Intel originally claimed Itanium would eventually replace x86 but it's x86 performance was similarly abysmal. Intel gave up on the idea of replacing x86 with Itanium. The i960 and i860 risc CPUs before it were flops too.

that's not even close to being true:

http://www.vanshardware.com/articles/2001/september/010921_Transmeta_v_C3/010921_Transmeta_v_C3.htm

in the above circa 2001 comparison a Via C3 was tested against a Transmeta TM5600

and here they tested a 600 mhz Crusoe against a 600 mhz P3:

http://www.heise.de/newsticker/meldung/Crusoe-Nicht-der-Schnellste-aber-sparsam-29762.html

as far as the Itanium is concerned you conveniently choose to leave out certain key details; for instance itanium was originally supposed to employ x86 emulation in hardware, intel dropped that and decided to use a software emulation layer called IA-32EL instead.

furthermore, both intel and microsoft saw ia32/ia64 as the future of pc's and microsoft even released a 64 bit version of win 2k specifically for the itanium.

i can only find one reference to performance concerning emulated x86 code running on itanium and while the results are poor the benchmark itself is from all the way back in 2001; i think in the 10 years since, had intel so desired, they could have figured out how to run emulated x86 code in real time; hell intel's x86 processor have had virtualization hardware (VT-x) since the P4 days; don't you think had intel wanted to they could have used that technology and included hardware x86 emulation with their itaniums?

intel initially wanted to move away from x86 because their x86 patents were about to expire (patents only last 10 years in the u.s.); they didn't anticipate such a backlash from all their customers that didn't want to lose use of their legacy software. intel then decided to offer support for x86 on the itanium but purposely made sure that it would run slower as a means of promoting a forced upgrade (kind of what apple did when they moved from the power pc to the intel processor). when that backfired intel pulled another dirty trick; they made sure that the floating point capabilities of the P4 were weaker than the P3's on a clock for clock basis and they, with the help from microsoft, declared x87 deprecated and instead encouraged developers to use sse2fp, a technology that they would hold the patents to for the next 10 years.

fast forward 10 years and intel, again facing the specter of their technology moving to the public sector introduced 256 bit avx extensions, and surprise surprise, made said extension float only and once again they encourage developers to use avx exclusively; now granted a very convincing case is made for using avx over sse or x87, at least from a performance standpoint, but that doesn't mean any meaningful conclusions about the underlying technology can be drawn by looking at benchmarks as amd has shown that it's possible to created a 256bit non-vectorized floating point unit.

x86 emulation is probably one of the least vectorizable or parallelizable problems.

i don't see how you can say that with a straight face; x86 is an out of order execution instruction set, if anything should be easily parallelized or vectorized it should be instructions that can be fetched, decoded and executed in a data governed order.

it seems logical to me that mapping an x86 abstraction layer to 3000+ gpu cores should be a task that requires among the least amount of thought as to how to spread the work out among so many cores; if anything i would think the more cores the better.

[johns0]

To get back onto the main topic i got my 2006k installed and winxp 7 oem activated again on the phone with no problems and apps run faster such as 1gb rar files that were taking 42 seconds to unrar on the same hdd now take 18 seconds and multiavchd used to take 8 seconds to open now takes 4 seconds.

To test intel quick sync i will have to have the hd graphics as my main graphics but don't have it setup yet.

[deadrats]

To test intel quick sync i will have to have the hd graphics as my main graphics but don't have it setup yet.

great, can't wait to see the results, just remember to enable "working with D3D surfaces" for max performance with quick sync.

[johns0]

Found out i cant use the builtin graphics cause i got a p67 mb,need a h67 mb for that,i bought the msi p67 for gaming which i do.

[deadrats]

Found out i cant use the builtin graphics cause i got a p67 mb,need a h67 mb for that,i bought the msi p67 for gaming which i do.

i'm confused, why do you think you need a p67 mb to game on? h67 boards support discrete graphics solutions.

[johns0]

Cause p67 are geared for gamers.

[deadrats]

Cause p67 are geared for gamers.

care to expand on that; what features does P67 offer that H67 doesn't that is a benefit to gamers?

[johns0]

Better overclocking.