Magny-cours server chips shipped out

[yannifb]

Here's the news:

http://www.xbitlabs.com/news/cpu/display/20100219070650_AMD_Begins_Revenue_Shipments_of_Twelve_Core_Opteron_Microprocessors.html

I guess 2.4 Ghz for 12 cores isn't bad. Also that quad channel memory controller seems pretty nice. How much bandwidth do you guys think this chip will have? Also, do you think AMD's Fusion APU and Bulldozer will have this quad channel controller?

If this holds its own against the 8 core Xenon then good news for AMD.

 

[jennyh]

2.4ghz is amazing for 12 cores.

I know a lot of intel fans will say stuff like 'yeah so what, 6 core + hyperthreading + turbo will beat it sometimes', which is fair enough in one regard.

On the other hand, turbo boost and hyperthreading take up a tiny amount of silicon real estate. AMD will have these features in Bulldozer, along with 12 or even 16 cores. Can you imagine how powerful Bulldozer is going to be? Intel really needs to catch up on the core count before too long.

 

[cjl]

Jenny: This won't be competing against the 6 core Intel CPUs. It'll be competing against the 8-core Beckton chips.

 

[jennyh]

It wont be, they are completely different markets as MC is targeted at 2P (of course they will compete at 4p, but that's still a relatively small market in comparison). Also I still don't see Beckton and I've been hearing about it for well over a year...and I'll be surprised if the clocks are a lot higher than 2.4ghz.

Give credit where it is due. 2.4ghz on a 12 core cpu is pretty amazing considering 45nm, no HKMG...it's almost unbelievable.

 

[cjl]

Oh, absolutely it's impressive (and I have to agree that Beckton will be ~2.2-2.4 GHz, at least until they switch it to 32nm). I'll be curious to see how they compare though (both to Beckton and to Gulftown).

 

[hunter315]

wow that seems like quite a few cores even for a server, hopefully AMD will be able to switch to 32nm soon as it would give them the potential for higher clock speeds which would definitely help it to take on an 8 core intel with hyper threading and turbo. Good to see AMD is making significant progress on server chips, more money they take in from there the more they can drop into desktop CPU R&D

 

[someguy7]

Well done AMD. Intel made the double cheeseburger and AMD comes back with the baconator. That is just a joke by the way. Don't need the AMD lovers getting all upset.

 

[BadTrip]

Well Done with the 2.4 ghz. Anxious to see some numbers.

 

[yannifb]

I can't wait to see the 12 and 16 core server chips based on the Bulldozer arch.

 

[BadTrip]

You will be waiting for quite a while.

 

[yannifb]

q1 2011. Not too long. And IIRC on the Intel is doomed thread (closed now) someone posted a link that said AMD is releasing the BD chips in q4. Of course this was before that thread got massively off topic.

 

[BadTrip]

q1 2011. Not too long. And IIRC on the Intel is doomed thread (closed now) someone posted a link that said AMD is releasing the BD chips in q4. Of course this was before that thread got massively off topic.

You will not see Bulldozer before 2011, and I seriously doubt 12 and 16 core Bulldozer chips will be released before August 2011. Probably later.

 

[yannifb]

You will not see Bulldozer before 2011, and I seriously doubt 12 and 16 core Bulldozer chips will be released before August 2011. Probably later.

And you're so adamant about this why?

 

[BadTrip]

New Node + New Arch = Big Trouble

and I am not saying Bulldozer wont be Badass, it's just going to have problems.

 

[BadTrip]

If Intel tried the same, I would be saying the EXACT same thing.

 

[jennyh]

Yup, a new architecture and node at the same time is enough to break a company.

AMD will be ok, but I think Bulldozer is a lot more likely Q2 2011, almost best case.

 

[BadTrip]

Yup, a new architecture and node at the same time is enough to break a company.

AMD will be ok, but I think Bulldozer is a lot more likely Q2 2011, almost best case.

I agree. AMD will definetly be OK. Bulldozer will not happen before 2011, and you are probably right with Q2 predictions.

 

[yannifb]

I agree with you on those points, but q1 is a tad late. I'd say q1 is very likely, since global foundry is already testing 32nm process, and b/c BD has been in the works for a very long time. However even if it does come out in q2, its going to be something special. This is because it is supposedly going to have a significantly higher IPC and integer perf, the two places AMD lacked in against the i7. Also it has a new pipeline design (the Ph II still uses an old one, but modified). Heres the source of all this info- a link to many links:

http://groups.google.com/group/comp.arch/browse_thread/thread/759bcccbfa0b8b07/3cd3bfa93b736a56?q&pli=1

Many things, but interesting nonetheless. Oh and some of the FAD 2009 slides are ones that I haven't seen before, which was pretty interesting.

 

[jimmysmitty]

It wont be, they are completely different markets as MC is targeted at 2P (of course they will compete at 4p, but that's still a relatively small market in comparison). Also I still don't see Beckton and I've been hearing about it for well over a year...and I'll be surprised if the clocks are a lot higher than 2.4ghz.

Give credit where it is due. 2.4ghz on a 12 core cpu is pretty amazing considering 45nm, no HKMG...it's almost unbelievable.

Beckton is planned for 2p+. Its been slated for a 2010 release, never sooner. The 6 core is a different area, and probably mainly for 1P Xeon servers that people can afford and DT.

TBH, 12 cores, 16 or even just 8 are not impressive. I mean considering that intel has already made a working 80 core and plans a 48 core for Cloud Computing it kinda makes these seem like the drawing it out part and making us wait for something more amazing than we should have to. I am sure Intel could easily put the 48 core into full production.

And its nice to see AMD using quad channel like Intel will for Beckton. Might give them a boost needed to subdue Intels market share grabbing or at least save what they have.

 

[jimmysmitty]

I agree with you on those points, but q1 is a tad late. I'd say q1 is very likely, since global foundry is already testing 32nm process, and b/c BD has been in the works for a very long time. However even if it does come out in q2, its going to be something special. This is because it is supposedly going to have a significantly higher IPC and integer perf, the two places AMD lacked in against the i7. Also it has a new pipeline design (the Ph II still uses an old one, but modified). Heres the source of all this info- a link to many links:

http://groups.google.com/group/comp.arch/browse_thread/thread/759bcccbfa0b8b07/3cd3bfa93b736a56?q&pli=1

Many things, but interesting nonetheless. Oh and some of the FAD 2009 slides are ones that I haven't seen before, which was pretty interesting.

Intel was testing 32nm silicon back in 2007. Normally testing it means 1+ years away at least. Intel was testing 22nm in 2009 and thats slated for 2011.

I would like to see the pipeline design and see if they have a shorter or longer pipeline than K10. If they can shorten it while pushing K10.5 speeds, they will have great performance boosts. Then again Intel could do the same.

The nest 2 years will be interesting in the CPU market. probably boring in the GPU market. its still "ATI leads..... no wait nVidia leads......wait ATI......nV....." ect.

 

[Reynod]

Awesome+

I can see all of those database guys sporting erections at least half the size of wingy's !!

 

[MU_Engineer]

Here's the news:

http://www.xbitlabs.com/news/cpu/display/20100219070650_AMD_Begins_Revenue_Shipments_of_Twelve_Core_Opteron_Microprocessors.html

I guess 2.4 Ghz for 12 cores isn't bad. Also that quad channel memory controller seems pretty nice. How much bandwidth do you guys think this chip will have? Also, do you think AMD's Fusion APU and Bulldozer will have this quad channel controller?

If this holds its own against the 8 core Xenon then good news for AMD.

Well, we know how much bandwidth there will be. The Magny-Cours have four DDR3-1333 memory controllers per socket, for a memory bandwidth of 41.6 GB/sec per socket. There are also three 6.4 GT/sec cache-coherent HT3 links, which are good for 38.4 GB/sec full-duplex. And there is a 5.2 GT/sec non-coherent HT3 link to talk to the chipset, good for 10.4 GB/sec full-duplex.

AMD's Fusion APU as far as the roadmaps have said are going to be desktop and notebook parts and fit into variations of existing sockets (S1 and AM3), so they will be dual-channel DDR3 parts. Socket G34 Opterons are the only quad-channel parts and the roadmaps say there will be Bulldozer-based parts (Interlagos) but no mention of an APU.

 

[Reynod]

MU how does the bandwidth compare to the new Nehalem based server parts?

 

[MU_Engineer]

MU how does the bandwidth compare to the new Nehalem based server parts?

Memory bandwidth will supposedly be pretty similar as the new Nehalem-based server parts (Beckton, aka Nehalem-EX) have four DDR3 memory interfaces. I don't know what speed Intel will decide to run them at in any specific chip, but it would be fairly safe to say that DDR3-1333 should be available in the fastest parts.

As far as platform I/O bandwidth, it varies considerably but is generally less than what the Magny-Cours Opterons have, if my source is correct. Cisco accidentally leaked a brief overview of these chips and somebody put it on Wikipedia, which is all I have to go off of. The Xeon 6500-series Becktons that are targeted at DP operation have two 4.8 or 6.4 GT/sec QPI links. One would go to the neighboring CPU and one would go to the chipset, so the 6.40 GT/sec units have identical CPU-to-CPU bandwidth as a dual Magny-Cours setup but a little more CPU-to-chipset bandwidth. The 4.8 GT/sec unit has less of both. The MP Becktons have either three QPI links at 4.8 GT/sec or 5.86 GT/sec or four QPI links at 4.8, 5.86, or 6.4 GT/sec. The ones with four QPI links at 6.4 GT/sec are slightly ahead of the Magny-Cours (three at 6.4 GT/sec + one at 5.2 GT/sec) while the rest fall behind the Opterons.

Again, that's a sketchy source and I feel there is some information missing.

 

[jimmysmitty]

^From what I have read for Nehalem, there are 2 QPI links. 4.8GT/s and 6.4GT/s. Only the EE DT and server ships for 2P+ have the 6.4GT/s.

Beckton itself is going to support quad 6.4GT/s QPI links with Quad DDR3-1333.

It will be interesting to see both in action. But at this rate, we will never have enough time between CPU gens to actually see which one is better.