AMD Readies ?Thuban? Six-Core Desktop Processor

[MODEL3]

Originally posted by: Idontcare

Originally posted by: MODEL3

Nevertheless, based on currently available information, AMD Thuban is due only in Q3 2010

Damn!
I thought AMD would launch them in Q2 along with leo.

The microprocessors will be compatible with socket AM3 infrastructure and will have integrated dual-channel PC3-10600 (DDR3 1333MHz) memory controller

Since Xbitlabs is saying Q3 2010 shouldn't the memory controller support DDR3 1600?

Q3 2010? Hmmm...that's getting dangerously close to when we'd expect a 32nm release timeline...Could the story here really be that Thuban is 32nm 6-core Phenom III?

I don't know.
Xbit labs is saying 45nm SOI, but that part may be Xbitlabs speculation!

I just remember that Anand said that the high end Phenom II X4 965 must in september fall to around 199$ to be competitive with intel's offerings.

AMD up until now launch per quarter a higher (Q1 09 3GHz, Q2 09 3,2GHz, Q3 09 3,4GHz) clocked Phenom X4 version!

If in September Q3 2009 the highest AMD offering go to 199$, can they release for a year (up to Q3 2010) higher clocked (64bit OSes) Phenom X4 versions? (and have the quantity it takes to fill the 199$ market?) (my guess is no)

I think, if they want to increase their ASP, they have to release a 45nm Phenom II X6 version! (Along with leo it could be a good timing? irelevant but why they delayed leo one quarter?)

 

[alyarb]

istanbul is 45nm. AMD has no reason to wait until 32nm for a 6-core AM3 part.

 

[drizek]

The semi-inebriated conspiracy nut inside me tends to wonder if there isn't some manner of mobo-support ecosystem "spread the wealth" effects going on here.

I can see that. On the other hand, lower chipset prices give these companies a little more room to play with margins.

While I am sure manufacturers love selling high-priced X-series motherboards from Intel, at the same time they have to redesign the things every few months and validate them against bleeding edge, half-broken(judging from the anand labs blogs) ram and other components. And because they are so expensive, they get very little volume. AMD boards require quite a bit less RnD. The 7 series north/south bridges required little if any redesign between revisions(minor changes to the SB, most of the changes were just higher IGP clock speeds) and provided good performance across the board. Whats not to love?

 

[Idontcare]

Originally posted by: alyarb
istanbul is 45nm. AMD has no reason to wait until 32nm for a 6-core AM3 part.

Agreed, but then they also have no technical reason to wait until Q3 2010 to release such a part either.

Originally posted by: MODEL3
If in September Q3 2009 the highest AMD offering go to 199$, can they release for a year (up to Q3 2010) higher clocked (64bit OSes) Phenom X4 versions? (and have the quantity it takes to fill the 199$ market?) (my guess is no)

Excellent point, and I agree with your guess too.

AMD needs something to fill the year that is going to exist between Intel's release of 45nm Lynnfields and 32nm Clarkdales versus AMD's release of whatever is going to be their first 32nm CPU (be it a PhenomIII shrink, a Bulldozer part, or Bobcat).

Given the TDP's required to hit 3.4GHz on PhII X4 with GF's 45nm process it doesn't seem too likely that AMD is going to be releasing >4GHz X4 SKU's in 2010 without busting the 140W TDP budget.

And since AMD would definitely know this already if it is true (and I am merely speculating, so it might not be true) then they would most certainly already have something planned and in motion to address Q4/09, Q1/10, Q2/10, Q3/10, Q4/10, and Q1/11 (earliest AMD's 32nm desktop CPU's would be available for purchase at Newegg by my estimate, could be wrong).

 

[Phynaz]

Originally posted by: alyarb
istanbul is 45nm. AMD has no reason to wait until 32nm for a 6-core AM3 part.

Except AMD has said as recently as yesterday that Istanbul is server only.

Don't forget a 2.8Ghz Istanbul is 140W, a lot to put in a desktop.

If these are on the drawing board they will have to be based on Lisbon.

 

[drizek]

140W is not too much for a desktop, not if they want to compete on the high end. For socket AM3 owners who are jealous of Core i5, being able to upgrade their CPU to a 6 core might be a better value than dumping their system and getting a brand new Intel.

I think they can get a 3ghz to fit in the 140W TDP. Maybe even 125W if they push it. Enthusiasts and enthusiast motherboards should have no problem pushing it further.

 

[OCGuy]

Sweet! But just like i7, it is worthless over C2Q/PHII for almost everyone.

 

[Phynaz]

I didn't say 140W is too much for a desktop, I said it's a lot for a desktop.

As far as 125W at 3.0Ghz, AMD says not soon.

Postby JF-AMD on Wed Sep 02, 2009 6:31 pm

Process improvements come in 100-200MHz increments every 12-18 months. How are you expecting Istanbul to get to 3.4GHz for instance? That is a 600MHz jump. Or, at typical cadence, 3-5 years.

 

[alyarb]

Originally posted by: Phynaz

Originally posted by: alyarb
istanbul is 45nm. AMD has no reason to wait until 32nm for a 6-core AM3 part.

Except AMD has said as recently as yesterday that Istanbul is server only.

Don't forget a 2.8Ghz Istanbul is 140W, a lot to put in a desktop.

If these are on the drawing board they will have to be based on Lisbon.

you're splitting hairs between two very similar architectures. istanbul is placed on socket 1207 for 2/4-way servers with FB-DIMMs. AM3 parts have to work with a thinner, higher-frequency HT bus and DDR3 memory. implementing these little differences is not going to be a problem, as it's already been done with shanghai/deneb. currently there are 40W and 75W 6-core opterons, but you do run into high-wattage as you approach the 3 GHz mark.

 

[MODEL3]

TechReport's news report about Xbitlabs article!

http://www.techreport.com/discussions.x/17520

If Thuban exists, it could be a derivative of not Istanbul, but Lisbon, AMD's next-gen six-core offering

Some info for lisbon (it is also 45nm):

http://blogs.amd.com/work/tag/lisbon/

Next year, we plan to introduce the ?San Marino? platform, featuring our ?Lisbon? (C32) processor. We expect to have a platform that is specifically tuned to the needs of these very low power environments. It may not be a stretch to say that with next year?s optimized platforms, we may be able to provide even lower total power consumption than we see today with these AMD Opteron EE processor-based platforms.

Some info for AMD's server roadmap (San Marino/lisbon, etc...) :

http://www.techreport.com/discussions.x/16797


And some info for the current Six-core Opteron line:

http://www.techreport.com/discussions.x/17210

http://www.techreport.com/discussions.x/17502

 

[Idontcare]

Is there any reason to suspect the core architecture and isa for lisbon to be different from that of istanbul (or shanghai for that matter)?

I'm not getting what the relevant differentiation is here to say Thuban is Lisbon-derived versus Istanbul derived. Could be that I just don't know what the differences are between lisbon and istanbul. What are they?

 

[MODEL3]

Originally posted by: Idontcare
Is there any reason to suspect the core architecture and isa for lisbon to be different from that of istanbul (or shanghai for that matter)?

I'm not getting what the relevant differentiation is here to say Thuban is Lisbon-derived versus Istanbul derived. Could be that I just don't know what the differences are between lisbon and istanbul. What are they?

I'm not getting it either!

Techreport only says about the intergrated memory controller:

The Istanbul design only supports DDR2-800 memory, because AMD has kept the same Socket F platform on the server side since the dual-core days. However, it will release new platforms and new, DDR3-supporting Opterons next year

From AMD's site:

Integrated DDR2 DRAM Memory Controller: Low-Power Memory To Reduce Power Consumption
AMD's Integrated Memory Controller works exclusively with high bandwidth, energy-efficient DDR2 memory. It incorporates Memory RAS for increased fault tolerance to reduce system downtime and increase system reliability

Other than the intergrated memory controller, i can't find another major difference.
The core architecture must be the same.

 

[Idontcare]

Yeah that makes sense, update Istanbul IMC to be AM3/DDR3 compatible = Lisbon.

OT, sorry OP , but I just really don't "get" the drive for more and more threads above and beyond 4 for dekstop markets. The point of diminishing returns hits such a hard-stop when you go beyond 4 cores/4 threads for anything but the most challenging of rendering/encoding apps.

What we need is higher clocks so our 1-4 threads process even faster. I really think Intel "gets" this and that is why we see clarkdale coming out in a 2C/4T config. Given how i7 D0's overclock. I will be flabbergasted if clarkdales don't hit 5GHz on air. Which is exactly what we want/need for anything but the most embarrassingly parallelized application.

 

[Viditor]

Originally posted by: Idontcare
Yeah that makes sense, update Istanbul IMC to be AM3/DDR3 compatible = Lisbon.

Lisbon is also scheduled to be ULP. The goal is to have higher clockspeeds with lower total power consumption than today's Istanbul...
I imagine that this will be part of the IMC change...they did that before (dramatically reduced power but increased performance) when they released the Venice core.
Techreport

"The Venice core typically offers a bit of a performance advantage over the previous Newcastle core, and it does so while consuming less power at idle and under load than any of the processors we've tested alongside it. In some cases, like 3DMark05's CPU test, Sphinx speech recognition, and video encoding with Windows Media Encoder Advanced Profile, the Venice version of the Athlon 64 3800+ even outruns the older-rev Athlon 64 FX-55"

 

[ilkhan]

Originally posted by: Idontcare
Yeah that makes sense, update Istanbul IMC to be AM3/DDR3 compatible = Lisbon.

OT, sorry OP , but I just really don't "get" the drive for more and more threads above and beyond 4 for dekstop markets. The point of diminishing returns hits such a hard-stop when you go beyond 4 cores/4 threads for anything but the most challenging of rendering/encoding apps.

What we need is higher clocks so our 1-4 threads process even faster. I really think Intel "gets" this and that is why we see clarkdale coming out in a 2C/4T config. Given how i7 D0's overclock. I will be flabbergasted if clarkdales don't hit 5GHz on air. Which is exactly what we want/need for anything but the most embarrassingly parallelized application.

As someone who plays games on the same computer that later encodes fraps videos of said games...more thread, more clocks, more better.