6 core AMD chips info request

[Glow9]

Alright so haven't been on here much as of late so kinda been lacking on the updates.
So I'm kinda eyeing my upgrade route for the future here and I'd like to go back to AMD but my position:
I can just grab a Q9550 for $169. Or I can grab a spancy new i5 750 for $149. Both are cheaper than a 955 and the Q9550 is almost on par with it and the i5 can out perform the 955 if memory serves me correct.
So!!!! I figure since AMD is kinda losing in the price to performance war for me. I figure the 6 core AMD can do some damage. I have a dual core and I think jumping from 2 to 6 would be kinda wicked. Anyone know when these are planning to come out, and how much they will be? Or are these going to be wicked expensive like i7s? I just feel kinda meh.... about going with another intel rig considering 90% of my other rigs have been AMD.

[apt403]

If Xbit is to be trusted, sometime around late Q2, early Q3 2010.

[evilSuRGe]

I think it will be around the same time they release the new 890 chipset. Mid next year

[informal]

Glow,no offense,but you have been down this road before,many times.There were threads started by you,numerous ones,in which you asked about upgradeability of the AM2+/AM3 system,price,performance,upcoming CPUs etc. and you always asked if you should do this or that with AMD or go s775/whatever. There is nothing bad in evaluating your options,but it gets kinda boring and tiring that we have to read similar(almost identical) questions time and again.

I say you should have jumped on Q9550 LONG time ago and would be done with this issue.The chip should OC great(3.8-4Ghz on air) and should be able to handle anything you can throw at it in the next 3-4 years,IMO. i5/i7 and Phenom II are all similar CPUs performance wise where i7 is on top from this group since it excels in 3d rendering and a/v encoding due to SMT support. It's on average 20-25% faster than Yorkfield and Deneb and some 10-15% faster than i5(Nehalem sans SMT).
Thuban on the other hand should be able to match/outperform similarly clocked Nehalem due to 6 pure hardware threads,in apps that are well threaded(>4 threads). OCing wise Thuban will do similar clocks as Deneb does,so ~4Ghz air cooling,more with better cooling methods.Stock clocks ,I'd expect 2.6-3.2Ghz models,all within 140W TDP spec. Price won't be sky high,probably more aggressive(read lower) pricing when compared to similar performing intel MPUs. Availability is projected to be Q2 next year.

All of the platforms apart from s775 have good upgrade path,but s775 IS STILL great performer ,especially on desktop(at server space it sucks a lot compared to anything with direct connect/QPI).

So again,buy the cheap if not used/second hand Q9550 and enjoy.

[Nedjo]

LOL
Glow9 is again on his continuous upgrade mission.... man are we in some hilarious arguments

it amazes me how relentless is he in his viral attempts to show us how AMD is inferior in performance and price!

dude haven't you had enough in previous epic viral-busting threads??

please just go with another Intel rig, and please stay in this room and thread just for lots of LOLs... if there's anyone still eager to flame war with you

[Glow9]

Glow,no offense,but you have been down this road before,many times.There were threads started by you,numerous ones,in which you asked about upgradeability of the AM2+/AM3 system,price,performance,upcoming CPUs etc. and you always asked if you should do this or that with AMD or go s775/whatever. There is nothing bad in evaluating your options,but it gets kinda boring and tiring that we have to read similar(almost identical) questions time and again.

I say you should have jumped on Q9550 LONG time ago and would be done with this issue.The chip should OC great(3.8-4Ghz on air) and should be able to handle anything you can throw at it in the next 3-4 years,IMO. i5/i7 and Phenom II are all similar CPUs performance wise where i7 is on top from this group since it excels in 3d rendering and a/v encoding due to SMT support. It's on average 20-25% faster than Yorkfield and Deneb and some 10-15% faster than i5(Nehalem sans SMT).
Thuban on the other hand should be able to match/outperform similarly clocked Nehalem due to 6 pure hardware threads,in apps that are well threaded(>4 threads). OCing wise Thuban will do similar clocks as Deneb does,so ~4Ghz air cooling,more with better cooling methods.Stock clocks ,I'd expect 2.6-3.2Ghz models,all within 140W TDP spec.

All of the platforms apart from s775 have good upgrade path,but s775 IS STILL great performer ,especially on desktop(at server space it sucks a lot compared to anything with direct connect/QPI).

So again,buy the cheap if not used/second hand Q9550 and enjoy.

I hear ya guy I do believe me, but when I was eyeing a new quad over the last while I was eyeing it all out prepping for this summer to buy, then ran into some money troubles with moving and such. So now I'm looking for a bit of an update for when the 6 core chips will be coming out. Since I haven't been following I figured an update would be good to catch me up. So since it might be Q3 I have a while to save up. Why is it that AMD needs to add an entire new core to do the same amount of threads as i7's? Can't AMD do their own version of hyperthreading or what not?

[apt403]

IIRC, AMD is working on SMT for Bulldozer.

[Particle]

Thuban will be out some time next year. A Q9550 is your cheapest upgrade solution and will provide competitive performance to the alternatives. It's certainly the price/perf winner for you since you already have the motherboard.

[FlanK3r]

I think 2Q 2010 and some newest in fo here:
overclokers.net

[informal]

SMT can't be bolted to any design,it takes A LOT of time and money to add some feature like that to your own design.Also it takes a lot of time to validate such a tech on your existing platform.
In reality,SMT gives you,on average 15-30% more performance for 30% larger die size.Not bad,but not ideal too. With real cores,especially if they are small enough(like Deneb's,which are a lot smaller than Nehalems,think 15.3mm2 Vs 24mm2 per core),it's easier to add more of them than to invest a lot of time and money into a problem which is more software related(pipeline bubbles) than hardware related. With well optimized software like Linpack,you have zero or negative gains with intel's SMT tech turned on. While general software gets better and better with time(optimized close to the level of one Linpack for example),SMT like technologies will bring smaller and smaller gains. SMT is NOT a bad technology for solving before mentioned software problem,but it's not ideal solution either.It also is not cheap power wise,MPUs using it do consume considerably more power while the gains may not be as expected so the feature must be turned off(hardware reset only for now),so you can see there are some caveats to it.

AMD has some big plans for it's next gen Bulldozer cores. It presumably is using some other approach to multithreading(so called clustered multithreading) than the one other big players are using(SMT).Basic idea is to "cluster" each future bulldozer 4-way core into some separate blocks,especially the integer execution units .Those units(2-way units) would be very efficient and could work independently and share data,so that the execution of either single or multithread code could be as efficient as possible. The core will still have one front end and one SSE/FPU unit that will be shared among 2 int clusters making it similar to SMT like designs in that regard. The BD cores will also suport wider SIMD execution similar to next gen. intel cores codenamed Bridge cores. All of this can be found on Dresdenboy's blog online.

[Glow9]

AMD has some big plans for it's next gen Bulldozer cores. It presumably is using some other approach to multithreading(so called clustered multithreading) than the one other big players are using(SMT).Basic idea is to "cluster" each future bulldozer 4-way core into some separate blocks,especially the integer execution units .Those units(2-way units) would be very efficient and could work independently and share data,so that the execution of either single or multithread code could be as efficient as possible. The core will still have one front end and one SSE/FPU unit that will be shared among 2 int clusters making it similar to SMT like designs in that regard. The BD cores will also suport wider SIMD execution similar to next gen. intel cores codenamed Bridge cores. All of this can be found on Dresdenboy's blog online.

It sounds like if they chose to do that since it would be on par with a 6 core chip they could instead go from 6 to 4 and save cash. Maybe I'm wrong but 6 seems to feel the gap for now but maybe no the most cost effective for the long haul if you can get way more performance out of these new cores.

[informal]

Bulldozer for desktop will go from 4 to 8 cores and will slide in AM3r2 socket(presumably AM3r2 CPUs can go backwards in "old" AM3 boards,but not vice versa). Server parts will do double of that(8 to 16 cores per one CPU,done with MCM over direct connect).

[wuttz]

i know bdozer isnt final yet, but they should have stuck with original (rumored)design: 2 int units and 2 fpu units. just so superpi fanboys would have their fill.

[haylui]

skip all these and wait for BullDozer...aka K11....

[demonkevy666]

i know bdozer isnt final yet, but they should have stuck with original (rumored)design: 2 int units and 2 fpu units. just so superpi fanboys would have their fill.

wait what ??
but I think that still would have been slow as deneb then.

[Glow9]

I'm getting confused now, so bulldozer will be out in 2011?

[Chumbucket843]

SMT can't be bolted to any design,it takes A LOT of time and money to add some feature like that to your own design.Also it takes a lot of time to validate such a tech on your existing platform.
In reality,SMT gives you,on average 15-30% more performance for 30% larger die size.Not bad,but not ideal too. With real cores,especially if they are small enough(like Deneb's,which are a lot smaller than Nehalems,think 15.3mm2 Vs 24mm2 per core),it's easier to add more of them than to invest a lot of time and money into a problem which is more software related(pipeline bubbles) than hardware related. With well optimized software like Linpack,you have zero or negative gains with intel's SMT tech turned on. While general software gets better and better with time(optimized close to the level of one Linpack for example),SMT like technologies will bring smaller and smaller gains. SMT is NOT a bad technology for solving before mentioned software problem,but it's not ideal solution either.It also is not cheap power wise,MPUs using it do consume considerably more power while the gains may not be as expected so the feature must be turned off(hardware reset only for now),so you can see there are some caveats to it.

linpack is BLAS so you will never reach that level of arithmetic intensity or parallelism with real world programs. something like file compression which runs almost 50% faster with HT is a good example. the server world is another huge fan of hyperthreading because all programs are already multithreaded.

yes, with OoO cores the benefit of HT is diminishing but thats typical of any technique to achieve high ipc. AMD said CMT will use more die space but provide bigger gains in return. if we were to compare an in order cpu like the atom HT can easily double the speed while only increasing the die size by 30%. its also hard to put down hyperthreading when CMT is only on paper and intel will continue to optimize SMT with newer architectures.

[Glow9]

So I read the article here http://www.xbitlabs.com/news/cpu/dis...s_to_2011.html It says its being built from the ground up. How will it really differ from deneb? Besides die size and adding exrta cores what else really makes it stand out or are those the big ones? Too be honest I'm excited to see what the new mobos are going to look like to host these kinda curious if they are going to be bringing anything new to the table (besides usb 3.0) Anyone know?

[demonkevy666]

So I read the article here http://www.xbitlabs.com/news/cpu/dis...s_to_2011.html It says its being built from the ground up. How will it really differ from deneb? Besides die size and adding exrta cores what else really makes it stand out or are those the big ones? Too be honest I'm excited to see what the new mobos are going to look like to host these kinda curious if they are going to be bringing anything new to the table (besides usb 3.0) Anyone know?

more ingeter units and fpu's pre-core with CMT cluster Multi-thread a 4 way core bigger then deneb supposedly.
(you should read Dresdenboys' blog: AMD Bulldozer for explanation of CMT)
hopefully SATA 3.0 on SB850 and hyper transport 3.1.

[Glow9]

I read this in a post however I am not sure if it is true. Someone mentioned this chips will be able to be dropped into sb750 boards but only ddr3 ones? Is this true?

[informal]

Yes,Zambezi is AM3 compatible and is DDR3(what is sure as of now,DDR2 has not been mentioned so I say disregard the support for it for now).

And it's funny how AMD just confirmed all the speculation dresdenboy and Hans De Vries fueled last couple of years.

Small correction in my post from November the 3rd. Each bulldozer minicore inside the module will be 4way wide(or 2way if FP/SIMD is being done resulting in 2x2way for one SIMD unit) ,not 2 way as I said in my post.
The FPU/SIMD unit will get 4 instr/cycle from fp dispatcher(so also 4way).

PS You also need BIOS support for Zambezi...

[FlanK3r]

But i think, Zambezzi will maybe only with AMD800 chipset compactible