AMD Updates CPU Launch Schedule - Phenom 9850 Due In a Few Weeks

[Rob Williams]

From our front-page news: According to industry-tracking web site DigiTimes, AMD have revised their roadmap to include a few new models, all of which should launch in the coming weeks. The Phenom 9050 models will include the 9850, 9750, 9650, 9550 and the 9150. Although the 9850 was initially due to launch with a 2.6GHz clock speed, difficulties have kept it from hitting that goal, so it will be released at 2.5GHz instead.

Triple-core 8x50 Phenom's will be launched in late April and include a TDP of 65W, and we should also be seeing dual-core Athlon 64-branded X2 Black Edition 5600+ and 4600+'s by the end of this month. For those awaiting high-end Phenom Quad-Cores, the company is still on track to launch those sometime in the third quarter.

<table align="center"><tbody><tr><td> http://techgage.com/images/news/amd_barcelona_core_large_logo.jpg
</td></tr></tbody></table>
For the dual-core CPU market, AMD will launch the Athlon 64 X2 5600+ Black Edition and 4600+ by the end of March and will follow with Phenom-based Athlon 6250 and 6050 CPUs by the end of the second quarter. The company has also informed customers that final orders for Athlon 64 X2 6400+, 6000+ and 5600+ (89W versions) will be taken at the end of March, the sources added.

Source: DigiTimes (http://www.digitimes.com/news/a20080317PD209.html)

[sbrehm72255]

I truely hope that no one is holding their breath on the release dates.

They (AMD) have been having nothing but problems getting stuff out the door lately.

And just think, I used to be one of the biggest AMD fans out there, not all that long ago really. My C2D system is my first ever Intel build.

[Rob Williams]

Same here. Before I built this Quad-Core, I was using a Socket 939, haha. Before that it was Socket A. I guess I lie... I know I had an Intel machine before... might have been Socket 7.

I do hope AMD can pull things together though. We need them, and right now, what they are doing just seems to be laughable. It's really too bad. I don't think anyone expected this to happen when C2D launched... since then things have been constantly going downhill for AMD.

[sbrehm72255]

Either AMD has had bad designs or plain bad luck, maybe a combination of the 2. But sooner or later things have to turn around for them, I'm really getting tired of this one sided tater sack race.........;)

[Krazy K]

How much slower does each core compute on a multi-core chip? It wouldn't make much sense to keep making single core if they did half the work of a similar clock speed dual core; price and power would be the main factors to make singles still.

Does this make sense?

[Merlin]

How much slower does each core compute on a multi-core chip? It wouldn't make much sense to keep making single core if they did half the work of a similar clock speed dual core; price and power would be the main factors to make singles still.

Does this make sense?
Think of it as one hand, two hands, three hands and four hands.
Sometimes if all you need to do is drink a cup of coffee, you only need one hand....Now if you had coffee and donuts, you may use two hands....one for the donut and one for coffee.
Now...if you wanted to type, eat a donut and drink coffee you would need three hands...and so on.
So you may only need one core to do one thing at a time, two cores to to do two things...and so on.
Now...how fast do you want to eat that donut?

:techgage::techgage: Merlin :techgage::techgage:

[Krazy K]

Well the computer doesn't need to enjoy doing it is whatever I tell it to do. That prime95 program will only use one core, would that run faster on a single core or dual core. Can one core of a multi-core do as much work as a single core is what I'm asking. SETI @home will utilize all the core, so there's 2-4 work units that I could do at one time. Are the cores equivalent or between chips or do they scale back they ones on the multi cores, and by about how much?

[Kougar]

How much slower does each core compute on a multi-core chip? It wouldn't make much sense to keep making single core if they did half the work of a similar clock speed dual core; price and power would be the main factors to make singles still.

Does this make sense?

Assuming you find a dual-core chip running at the same clockspeed and same L2/L3 cache as a quad-core chip, they will perform the same. Adding cores doesn't make the CPU slower unless they change something, such as lowering the clockspeed or removing some cache.

Well the computer doesn't need to enjoy doing it is whatever I tell it to do. That prime95 program will only use one core, would that run faster on a single core or dual core.

Prime 95 uses as many cores as you have available in the computer. ;) If you don't believe me, try: http://i111.photobucket.com/albums/n129/Chanur64/DQ6/12hour3GhzQuadOC.jpg

Can one core of a multi-core do as much work as a single core is what I'm asking. SETI @home will utilize all the core, so there's 2-4 work units that I could do at one time. Are the cores equivalent

Yes. ;)

[Krazy K]

OK, so your getting the same effectiveness as a single core, just times 2 or 3 or 4. Once again, considering all things equal, a dual core does twice the work as a single core and half the work as a quad core.

Gotcha!

[Kougar]

As long as the program can actually make use of 2, 3 or 4 cores equally, then yes. The catch is many programs still can't.

[Krazy K]

As long as the program can actually make use of 2, 3 or 4 cores equally, then yes. The catch is many programs still can't.

They can still run background processes, and when they aren't being used I would guess that they wouldn't draw any power. So why not keep them in your pocket until they need to be used. But like you said, nothing can really take advantage of the cores so is it really worth investing in. I guess that's a personal question.

[Krazy K]

Why stop at 4 cores?

Read here. (http://www.pcworld.com/businesscenter/article/145608/amd_jumps_to_12core_chip_skips_8core_chip_plans.ht ml)

[Rob Williams]

Why stop at 4 cores?

Read here. (http://www.pcworld.com/businesscenter/article/145608/amd_jumps_to_12core_chip_skips_8core_chip_plans.ht ml)

It seems a little strange that AMD wants to skip over Octal-Core CPUs entirely, since it seems to fit. Some servers might be able to use 8 cores, but 12 cores is quite a sizable jump.

Sounds like they really, really don't want to put money into things where it's not absolutely necessary.

[Kougar]

Seems more like they are specifically targeting the HPC and supercomputing markets, although turning two/four socket boards into a single socket might also be useful?

Those are the only markets AMD is still competitive atm anyway. Or should I say "ftm", for the moment.

[Rob Williams]

Those are the only markets AMD is still competitive atm anyway. Or should I say "ftm", for the moment.

This is true. The fact also remains... Quad-Core processors are more than enough for anyone, unless they have a specific need. So making 12-core processors for a specific market does make good business sense. An 8-core would have limited use, and no use in the consumer side of things.

Good points.

[Kougar]

Always three sides to a coin! I try to find them... :)

Silly or not, still would love to have a 12-core computer. I already shut down my old pre-Pentium 4 era Folding@Home boxes due to a very bad power/performance ratio... would love to condense my current systems into just ~3, and retire both of my Pentium 4 machines as those are very slow as well, very power inefficient to boot. I wish I could "fold" my systems into the ground, but they simply last way to long... I've not actually had a PC completely die on me to date, or nothing a quick swap didn't fix.

[Rob Williams]

I'm not saying that desktop users don't have the need for a 12-core, but they are very few. Folders would make up the biggest portion, I believe.

I agree about the P4 thing. Those beasts just suck down so much power and don't deliver good enough results. It means a higher power bill for no reason.

You need a budget Skulltrail ;-)

[Kougar]

Hah, I'd rather throw together two or three Quad systems for the same price... :) The Skulltrail board and FB-DIMMs are simply expensive and I don't feel Skulltrail was ever tuned as well as most enthusiast motherboards.

It's to bad Nehalem-EX (Becktown) will use FB-DIMMs and a custom LGA1567 socket... 8 cores with 16 thread processing would let me condense the entire household into a single computer. :D

[moon111]

It seems a little strange that AMD wants to skip over Octal-Core CPUs entirely, since it seems to fit. Some servers might be able to use 8 cores, but 12 cores is quite a sizable jump.

Sounds like they really, really don't want to put money into things where it's not absolutely necessary.

Wasn't AMD's 'triple' core actually a neutered Quad-Core unit with one core that didn't work cut off? If so, could they just dump manufactering defects from the 12 core unit into different packaging. The could sell 12 core units, 11 core units, 10 core...

[Rob Williams]

Hah, I'd rather throw together two or three Quad systems for the same price... :) The Skulltrail board and FB-DIMMs are simply expensive and I don't feel Skulltrail was ever tuned as well as most enthusiast motherboards.

It's to bad Nehalem-EX (Becktown) will use FB-DIMMs and a custom LGA1567 socket... 8 cores with 16 thread processing would let me condense the entire household into a single computer. :D

That does make sense. The benefit of Skulltrail though is that it's all in one rig and would use less power than two different Quad-Core rigs. The ASUS Z7 WS board, as far as I know, retails for under $500, but I could be wrong. Then there are the Xeon's, like the ones I used in my article, which are quite affordable as well. Also, FB-DIMMs are not much more expensive than DDR2, but I'll admit they are not as fast, either. I don't think fast RAM affects folding too much though.

The biggest downside of Skulltrail, to me, is the noise. It's nuts! At least the one here is. But I do see your point... it might just make more sense. BUT, do remember with a Skulltrail, it might cost a little more, but you only need one PSU, one GPU, one HDD, etc etc.

Wasn't AMD's 'triple' core actually a neutered Quad-Core unit with one core that didn't work cut off? If so, could they just dump manufactering defects from the 12 core unit into different packaging. The could sell 12 core units, 11 core units, 10 core...

That's an interesting point, but for the sake of marketing and market saturation, they wouldn't release different CPUs like that. You raise another question though, because isn't the Cell processor in the PS3, most often a 7-core chip, instead of the 8-core chip it's supposed to be, because of defects?

From what I understand, most Cell chips do indeed have a non-functioning core, so I have to wonder how AMD will manage to pump out 12-core processors that are faultless. I am no fab worker, so I'm really not going to guess at that.

[Kougar]

Yeah, it would likely be more power efficient I'll grant, but I could toss together some pretty old parts to make them too. And server boards aren't very power efficient, or made for overclocking dual CPUs.:)

You raise another question though, because isn't the Cell processor in the PS3, most often a 7-core chip, instead of the 8-core chip it's supposed to be, because of defects?

From what I understand, most Cell chips do indeed have a non-functioning core, so I have to wonder how AMD will manage to pump out 12-core processors that are faultless. I am no fab worker, so I'm really not going to guess at that.

Man, the answer is even more complicated than I thought. I quoted IBM's documention directly below. This was my understanding, that one is usually disabled due to the extremely low yields Cell processors had. It was around 15-25% yields during the first year iirc. Apparently however they built in a spare to begin with.

The Cell BE processor is a chip consisting of nine processing elements (note the PS3 has one of them disabled, and one of them reserved for system use, leaving seven processing units at your disposal).

I'm sure if AMD retains enough market share until they launch 12-core products, then they will eventually launch a version with a few cores disabled to sell off the mostly good chips. 100% yields is not possible, although Intel has been darn close with some of their better fab facilities.

[Rob Williams]

Oh, that's interesting. I thought the Cell was an 8-core chip with one disabled... I didn't know about the spare chip solely for system use. So I guess AMD could do the same thing. Produce a 13-core processor and up the yields, and disable one core if it happens to come out perfect, haha.

Good find.

[madmat]

Yeah, it would likely be more power efficient I'll grant, but I could toss together some pretty old parts to make them too. And server boards aren't very power efficient, or made for overclocking dual CPUs.:)



Man, the answer is even more complicated than I thought. I quoted IBM's documention directly below. This was my understanding, that one is usually disabled due to the extremely low yields Cell processors had. It was around 15-25% yields during the first year iirc. Apparently however they built in a spare to begin with.



I'm sure if AMD retains enough market share until they launch 12-core products, then they will eventually launch a version with a few cores disabled to sell off the mostly good chips. 100% yields is not possible, although Intel has been darn close with some of their better fab facilities.

The quad core Phenoms are 100% yield chips since the Phenoms are a pure quad core design versus Intel with their dual dualcore approach, although they've taken to selling the failed quads as triple cores.

[Merlin]

Maybe we can see a little competition between Intel and AMD.
But AMD is going to have to do better.

:techgage::techgage: Merlin :techgage::techgage:

[Kougar]

The quad core Phenoms are 100% yield chips since the Phenoms are a pure quad core design versus Intel with their dual dualcore approach, although they've taken to selling the failed quads as triple cores.

Well, I was talking about 100% yields in terms of wafers. Technology is just not there yet. That was why IBM designed the extra core into the Cell chip, to increase yields. If you want to get extremely technical, even if a Intel/AMD Quad CPU "works" when fabbed it doesn't make it a 100% yield, because it likely won't meet the top tier bin target, as far as quality goes.

Rob, I also thought it was an 7 or 8 core chip with one core disabled, was the first I'd ever heard of it being 9 cores. Makes me wonder if they use the same policy for their Cell Blades?

[Rob Williams]

Rob, I also thought it was an 7 or 8 core chip with one core disabled, was the first I'd ever heard of it being 9 cores. Makes me wonder if they use the same policy for their Cell Blades?

This is a bit confusing, to say the least. From what I can gather, each Cell CPU contains nine cores, but ten threads. There are eight SPEs, each of which are single-threaded, and then the Power Processor Element, which is unlike the other cores, but is still considered to be one. So, it's nine cores, but ten threads.

http://www.research.ibm.com/cell/images/cell-diephoto.jpg

http://www.research.ibm.com/cell/heterogeneousCMP.html

So I am assuming that the PPE is exclusive for system use, while the eight SPEs are what's used for all of the required calculation.

"The Cell B.E. processor is a multi-core broadband processor based on IBM's Power Architecture. A single chip contains a Power Processing Element (PPE) and eight Synergistic Processing Elements (SPEs). Each SPE has an on-chip local store. The PPE has an on-chip L1/L2 cache. Also on the chip are memory and I/O controllers.

The PPE is a general purpose 64-bit RISC processor with 2-way hardware multithreading. L1 cache for the PPE is 32KB instruction and 32KB data. Each of the eight SPEs is a special purpose RISC processor with 128-bit SIMD (Single Instruction Multiple Data) capability and 256KB local memory."

http://www-03.ibm.com/technology/splash/qs20/pdf/cell_be_systems_whitepaper.pdf

From what I can see, Cell Blades and the PS3 share the identical processor, and both probably use them in a similar manner.