AMD Releases Triple Threat and Price Cuts

[jkflipflop98]

Well, when your yields are at an abysmal 30-40%, you gotta do something to scrape some bucks together.

 

[jkflipflop98]

[citation][nom]Reynod[/nom]Kuka's point is a good one.AMD needs to bin the X2 gigher end parts like the 6400+ etc.Then more will show interest in the Phenom.Sounds good ... but when you need sales of "anything" then this falls down as an aarguement.Plus the 90nm parts (the high end X2's and old Opti's) are cheaper to make thant the Phenoms.The 65nm X2's also offer a wider powere envelope - albeit the only (Brisbane) mask they made was with the 512K (half the cache of the high end 90nm X2's) crippled cache.Some idiot should have made a full shrink of the 90nm (1Mb + 1Mb) mask and then they could have shut down the 90nm process alltogether.Idiots.[/citation]

It doesnt work that way Reynod. It appears once again that YOU'RE the idiot here. The main killer isnt the newer design, it's the crappy SOI process they're using.

 

[chaohsiangchen]

Nobody ever questions whether Intel "disables" bad cache block on some C2D chips and sell them as E2xx0, E4x00, E5x00 and E7x00 or fab them with different process. The failure rate of Phenoms doesn't seem to be high, so AMD can make more x3 instead of x2. Nothing is wrong with tri-cores, and people do buy them. The only complain I have to AMD is that they should have release 8750 BE in the first place.

 

[chaohsiangchen]

[citation][nom]Alien_959[/nom]Just asking. Is it a great idea to make propearty dual core based on Phenom something with 1 mb L3 cache an greater clock speed. Because the software is more optimised for dual core and greater clock speed would have greater impact on performance. Opinions?[/citation]

That's actually pretty bad idea. I've done some test with 9850 and it only out-take E4300 at 1.8GHz when data size is twice as large as E4300's L2. Granted that CnQ is on in those test, but at 1.8GHz same thing happens. Agena core is superior to K8, but still about 20% behind C2D clock for clock. Most of the deficiency can be attribute to failure to utilize enough CPU clocks in single threaded process. Phenom only shines when you can utilize at least 2 cores. In that case, C2D is no match to Phenom x3 or x4 clock for clock. Those SuperPi 1M test is an indication of the raw power of a CPU, and current AMD offerings are just not as good as Intel ones. However, if your code and data size is way larger than L2 cache, AMD starts to have an edge. But just look at the size of some Intel CPU's L2. They need to improve L3 size, IMC and also IPC in order to catch up Intel. From Phenom's architecture, cache area is actually relative small compare to Intel C2D. There is no point cutting L3 when most failure occurs on cores.

 

[ahslan]

intel's e5200 is a serious threat to what AMD has going on...the main reason why my rigs are AMD is because I can make a dirt cheap rig that provides more than enough performance for me...now that intel has introduced the e5200 ($90 45nm, 2.5ghz dual core...that overclocks well as well...jesus...), AMD really needs to do something...

 

[caamsa]

A quote from Anand Tech from April says it all.

"AMD doesn't have the resources to spin a dual-core Phenom die, so what better way of repurposing the quad-core die (especially if one core is defective) than to make a Phenom chip with less than four cores. Sure it's not the most efficient way to manufacture, but AMD doesn't have the luxury of producing a number of different Phenom die at this point. The triple-core Phenom strategy makes perfect sense if you're AMD, the question is: does it make sense if you're an end user?"

 

[WheelsOfConfusion]

*waits patiently for an updated CPU chart that can show the strengths of various dual-tri-quad cores*

 

[Alien_959]

I was reading rumors that AMD is going to outsource it's CPU business, if things are where they are now, it may be a good way to go( I know I don't like fabless AMD too), but in this way AMD can cut costs, resourses ect. AMD will have more focus on R&D. Anyway something must be done I hate to loose AMD as a competitor.

 

[dagger]

So, a 8250e@1.9ghz and e8400@3.0ghz both has 65w TDP. Assuming they performing the same per clock (which isn't true, Intel performs better), 3x1.9=5.7, 2x3.0=6.0. Even for programs that scale at 100% across cores, the AMD tri core is still slower.

 

[Guest]

"Even for programs that scale at 100% across cores, the AMD tri core is still slower."

What is even worse is put that 1.9GHz 8250 up against a 2.6GHz K8 which can be had for $59 (or an $87 2.8Ghz K8) both of which are 65Watt... the $59 dual core chip will be better on most applications. And on the rare applications which scale perfectly the performance will be very similar. Why would you buy an 8950 tricore, which is more expensive and generally slower than a 2.6GHz or 2.8GHz dual core K8? What new market is this ultra low end tricore serving?

Too many products... too many speed bins... too many 'black editions'... no cohesive market strategy. More products = more price points = less flexibility on product pricing.

 

[Guest]

Pricedroppinmadness: You obviously have no idea what the margins are on these. While they might make 5 more dollars on a quad and have 5x or so more volume, the actual dies cost near near $100. I makes much more sense to sell these at a ~10% margin and lose a bit on the high end than completely lose $100 for each failed die. Especially if yields are low, which makes sense if they are making triple core cpus.

I also like how you claim that AMD is more than likely using good quad dies because they need to fill the triple core demand. You state earlier in your post that there is hardly any demand. If this is the case, obviously even a reasonable yield rate could fill the orders.

 

[Guest]

hecksagon....just think for a second - AMD sells over 13mil CPU's/qtr...over 1/2 are desktop meaning somewhere probably around 7mil/qtr. AMD sells what at most a million desktop quad/qtr (which is probably on the high side) and unless yield rate is 50% probably on the order of 100-200K tri-cores (again this is probably on the high side). This would leave ~6mil or so dual cores.

200K tricore x $120/chip = $24mil. This is the equivalent of charging $4 more per dual core something which would probably be EASILY achievable if there was no tri-core pushing down dual core prices. Then add in a small potentially lost # of quad core sales and it should be rather simple to see that it might be wise just to throw tri-cores away completely - especially if you consider the act of packaging, selling and stocking these chips is not 'free'.

Again people are speaking with no sales and marketing background (or concept of ROI) and keep thinking better to sell them then throw them away - the logic is so simple you would think an idiot like me would understand it, no? AGAIN...the problem is the flawed assumption in that logic is that the sale of tri-cores has no impact on dual core pricing or on volume of dual core or quad-core sales. You are treating tri-core as some sort of new market, which it isn't... it is simply a converted dual core or quad core sale (think of it this way if tri-core didn't exist would that person not buy a dual or quad CPU instead?). When you consider the impact this has on pricing, then you may begin to understand scenarios where it is indeed better to throw them away.

The fundamental problem right now is AMD has so many products that there is something like $5-8 increments between parts so if tri-core has an impact on the high end dual cores this quickly cascades down thru the entire pricing structure. If there were fewer parts, AMD would be able to limit the damage to pricing on just a few bins.

And your yield rate assumption is ridiculous - don't assume most yield failures means that you get a tri-core part instead of a quad core. There are all sorts of yield failure modes which would kill the part altogether... people have this vision of a yield issue = just a bad part of 1 core. For example a 60% yield for quad core could mean 0% tri-core parts or some very small %... it certainly doesn't mean anywhere NEAR the 'leftover' 40%. So in fact a reasonable yield rate may yield very few tri-cores - I was just throwing out the idea that not all tri-cores HAVE TO BE non-working quads... this again is one of those convenient assumptions everyone is making.

 

[ZootyGray]

[citation][nom]caamsa[/nom]A quote from Anand Tech from April says it all. "AMD doesn't have the resources to spin a dual-core Phenom die, so what better way of repurposing the quad-core die (especially if one core is defective) than to make a Phenom chip with less than four cores. Sure it's not the most efficient way to manufacture, but AMD doesn't have the luxury of producing a number of different Phenom die at this point. The triple-core Phenom strategy makes perfect sense if you're AMD, the question is: does it make sense if you're an end user?"[/citation]
If you wish to destroy your own credibility, then keep on believing and quoting shrimpeenanand tik. Known for anti-amd, innaccuracy, bogus testing with nothing but anti-amd bias. Spintel employboy.

 

[ZootyGray]

No one has said this.

Athlon 64x2 is socket AM2.
Phenom x3 or x4 is the gateway to Socket AM2+.

A Phenom x3 or x4 makes available the newer tek features of Socket AM2+. Do your own research!
DDR2 1066 ram!
Faster Hyper Transport Link!
Vista's DirectX 10.1!
etc.

The 3rd core covers the possibility of multi-core software - and most software is still for dual cores - but there is some multicore software.

Clock speed can only do so much for performance. Cores can do more.

Much of the talk of oclox = a small, often insignificant performance increase. Most people do not oclock.

There is a great deal of rumour and opinion and guessing in these comments - if you do your own research, you will know that, or you will know better!

see www.amdzone.com.

 

[ZootyGray]

[citation][nom]caamsa[/nom]A quote from Anand Tech from April says it all. "AMD doesn't have the resources to spin a dual-core Phenom die, "[/citation]

Bias and total BS!!

AMD RELEASED A DUAL-CORE PHENOM ABOUT 2 WEEKS AGO!!!!

the "6500+ x2" based on Phenom tek!

Again - all you will get about AMD from shrimpee nanand tik is bogus rumour, slur, and bias. THIS IS WELL KNOWN TO AMD PEOPLE! they are a bunch od spintel fanboys, and they will discourage you with bs.

see amdzone.com

 

[Guest]

Zoot - the comment is actually accurate as the 6500 is not a re-spun die; it is likely a cut down quad core (whether it be thru defects or intentional) - hence no resources were needed for respin.

This chip has also not been "released" or "launched", it was merely referred to as an upcoming product in a power point slide. If you think this is a "native" dual core - take a look at the power and speed #'s and ask yourself - why is the speed so slow with such a high power bin considering it has 2 less cores than the Phenom.

AMD initially had a K10 dual core planned for Q4'07 (this was before the BARCY pseudo launch in Q3'07. This was then changed to Q2'08 and is now Q4'08? And given the very low speeds and relatively high power on the 6500 dual core - it may simply be similar to a tri-core, meaning another quad core piece of Silicon cut down to 2 working die.

 

[Guest]

you peopel rretarted, all I ketpt reading was "they need to get rid of the dual cores" and "why dont they drop the 3 cores for quad only". Im not sure why all you people read is gossip and not reviews. The single cores get a better rating on HD movies than the dual cores, the dual cores better than the quad cores. Therefore higher speed dual cores are btter for the people who have thrown away their TV sets and moved onto the future. These scores are posted here at this site.
Second the quad core AMD were faulty and the fix was to shut one core down, hence where the triple cores came from.

AMD is like the Aldi of computer parts, it may be cheaper by a few pennies, but you get what you pay for (usually less with most amd's offerings), I know I went with them last PC I built.