Can AMD survive Bulldozer's disappointing debut?

[Guild]

Originally Posted by carop

The numbers are from the usual suspect. They should have given him credit for those numbers. He does seem to have the silicon. The last time I heard about this CPU, it was on sale at e-bay.

It seems Intel is not without problems. There are problems with the virtualization engine and SAS ports. However, the real problem is with the power that is required by Sandy Bridge-EP. The clock rate of the high end workstation model will reach 3100 MHz with a very generous 150 TDP. I think that is a lot of power for a CPU to be acceptable:

http://www.cpu-world.com/news_2011/2...ries_CPUs.html

When I look at the table above, I see that most of the Sandy Bridge-EP Xeons will not run above 2500 MHz. They may make sense as Nehalem replacements from 3 years ago, but I would rather wait and see how they stack up against the Westmere-EP.

If only AMD could have delivered "35% more performance with 33% more cores" ...

To be fair, AMD's top 16C CPU will be cheaper than any 8C Xeon E5. It's more of a competitor to 6C/4C. TDPs aren't too bad either - most are 110W.

[mmarq]

Originally Posted by carop

Cray will probably be one of the factors for using PCIe.

Cray's next-generation Cascade supercomputer is scheduled to be introduced sometime between 2012 and 2014. It is partially funded by the High Productivity Computing Systems program of DARPA.

One of the key components of Cascade is a new interconnect code-named Aries. Aries replaces the HT links of the current Gemini interconnect with PCIe links. This suggests that future AMD Opteron and Intel Xeon processors as well as GPU co-processors will have embedded PCIe circuits.

For example, Cray has signed a multi-year, multi-phase contract with the High Performance Computing Center at the University of Stuttgart. The delivery of the Cascade system, is expected to be completed in the second half of 2013:

http://www.prace-ri.eu/More-capacity...-PRACE?lang=en

There is also some discussion that the Intel MIC architecture could use QPI rather than PCIe. It is possible that they will have QPI variants. The MIC architecture targets the HPC segments, and it seems that Cray is the major factor for using PCIe instead of QPI.

Those PCIe links in the "server" CPU die ?... ummm.. not likely, G34 is already 1974 pins and is BIG... for up to 5 HT links and 4s systems... G2012 is perhaps more links for 5 modules 10 cores chips(20 core Interlagos) but the difference most probably doesn't warrant to put there a PCIe controllers, though 2012 most probably doesn't indicate the number of pins...

Nevertheless AMD needs HT for CPU to CPU interconnect with cache coherency... PCIe HAS NONE OF IT... and so why spoil a good thing with stupid crap bloating PCIe ?

Cray already does their chipsets, perhaps those PCIe links are from their chipset HUB, after all practically NO DIFFERENCE at all in a performance POV, if those PCIe links are die born or chipset born... PCIe is not really prepared for this jobs...

Doubt very much ARIES replaces HT on AMD chip dies.. unless ARIES is a the TM of Cray for HT4.0.. perhaps some new functionality at the CPU chipset I/O for HT4.0.. after all HT is already much more and easily extensible than PICe.

As for Intel and QPI.. i think its correct, the future is complete closed platforms, if Intel pulls something reasonable as a GPGPU, they might use QPI slots(develop them) and completely close the platform... and then try extortion in the future, Radeon or GeForce for QPI slots ?... they have to pay Intel good!... when it should be the other way around!...

(oh! boy i couldn't think of a better reason to show'em a finger than this.. after all HT is already better than QPI and it could be much better.)

http://en.wikipedia.org/wiki/List_of...sed_.282012.29

And funny those lists already speak of a 28nm HKMG for server parts ?... server half node ? ... SOI ?

[pTmd]

Originally Posted by mmarq

http://en.wikipedia.org/wiki/List_of...sed_.282012.29

And funny those lists already speak of a 28nm HKMG for server parts ?... server half node ? ... SOI ?

That's based on a set of slide about Opteron platform and it was presented in July IIRC. In the future platforms roadmap, Dublin and Macau were marked to be manufactured on 28nm.

After that, GlobalFoundries demonstrated a slide in GTC 2011, suggesting the cancellation of 22SHP and its replacements as 28SHP and 20SHP. Since processes with SHP suffix usually means to be more advanced and with SOI, we can assume that Dublin and Macau will be manufactured on GlobalFoundries 28SHP.

[fatty2nd]

Apparently AMD will not survive this Bulldozer thing:-

http://news.softpedia.com/news/AMD-N...e-237103.shtml

At least not as the company as we know them. Say hello to paying out the nose for products that aren't as good as they could be.

[James]

Some really bad assumptions being made over one comment in that.

[BibianaHunter]

I am sure the ram idea precedes Mr. read. Plus, it launched three months ago which was just after he came on board. It seems everyone just loves to bash AMD. But AMD will.survive.

[flippin_waffles]

Originally Posted by BibianaHunter

I am sure the ram idea precedes Mr. read. Plus, it launched three months ago which was just after he came on board. It seems everyone just loves to bash AMD. But AMD will.survive.

^Exactly. But not just 'everyone' loves to bash AMD, there appears to be a very select group that gets the ball rolling.

As for the RAM, it appears to have gone over well in another market where the idea was tested. Apparantely North America is just too good to have this option, liberties be damned. What a culture lol.

[mmarq]

Originally Posted by pTmd

That's based on a set of slide about Opteron platform and it was presented in July IIRC. In the future platforms roadmap, Dublin and Macau were marked to be manufactured on 28nm.

After that, GlobalFoundries demonstrated a slide in GTC 2011, suggesting the cancellation of 22SHP and its replacements as 28SHP and 20SHP. Since processes with SHP suffix usually means to be more advanced and with SOI, we can assume that Dublin and Macau will be manufactured on GlobalFoundries 28SHP.

No the GTC2011 doesn't seem to talk about SHP... only 28nm LP variants and those are "bulk"... or at least they speak of HKMG for 32/28nm and since 32nm is SOI exclusively, then 28nm SHP must be SOI also, i think... more so because there is HP, LP ad SLP for bulk.. why another one ?

Is GF and AMD leaving SOI tech for now ? ... an Intel dream!... when its reported that UTB-SOI at 20nm is the right way to go ? .. or is it 28nm SHP SOI also ?

Originally Posted by fatty2nd

Apparently AMD will not survive this Bulldozer thing:-

http://news.softpedia.com/news/AMD-N...e-237103.shtml

At least not as the company as we know them. Say hello to paying out the nose for products that aren't as good as they could be.

ah!.. the same hogwash before... the best u-arch to date for multithreading is abandoned to pursue low power bobcat things... in the meanwhile bobcat evolution wichita and krishna are canned, and their project leader leaves AMD to Samsung taking with him part of his crew, yet AMD and its new CEO are doubling down on this low power u-archs ??? .. when they have already a Trinity bin that can go down to 17W (and Trinity IS BD u-arch in essence!) ???

What part of "as good as they could be" doesn't fit this scenario ?.. all of it perhaps ?

[ if this BD u-arch starts to get easily over 4ghz and kick major buts, there's gonna be a lot of ppl highly surprised!... when now they can be disappointed.... ]

[mmarq]

Originally Posted by rich wargo

Working on it, mmarq! Hey, Tim, get da axe and chop dem timin' crystals in half, we need ta double da speed on dem damn chips!

maybe you guys should apply for a patent " chip chop apparatus for speed'em up " lol

.. go ahead buddy you have all my support, keep-on-pushin even if you have to knock the god damn building out of its track (don't worry 'll hold).. it will be hi-tech fabing on the move lol

[fatty2nd]

Originally Posted by mmarq

snip

What part of "as good as they could be" doesn't fit this scenario ?.. all of it perhaps ?

snip

I meant that if Intel had no competition that is what we would get.