AMD: Six-core Istanbul server CPUs moved up to May

[Yokozuna]

"During the early part of this afternoon's conference call with analysts this afternoon, AMD CEO Dirk Meyer told analysts -- one day ahead of a momentous product call scheduled for tomorrow afternoon -- that strong reception and testing for its Istanbul-architecture server CPUs will enable the company to start orders for its first six-core products next month. This will enable shipments of six-core systems from suppliers as soon as June, said Meyer."

Full story here:

http://www.betanews.com/article/AMD-Sixcore-Istanbul-server-CPUs-moved-up-to-May/1240349461

[v_rr]

wow, that was fast:
- December X4
- May X6

5 months diference only.

[roofsniper]

nothing new you can order them in may but they are still coming out in june like annouced a week or two ago.

[demonkevy666]

O_O pushing to limit on that 45nm production
hack on AM2+ for 6.0ghz 6 core pwnage.

[gumballguy]

"During the early part of this afternoon's conference call with analysts this afternoon, AMD CEO Dirk Meyer told analysts -- one day ahead of a momentous product call scheduled for tomorrow afternoon -- that strong reception and testing for its Istanbul-architecture server CPUs will enable the company to start orders for its first six-core products next month. This will enable shipments of six-core systems from suppliers as soon as June, said Meyer."

Full story here:

http://www.betanews.com/article/AMD-Sixcore-Istanbul-server-CPUs-moved-up-to-May/1240349461


Lol it certainly is "betanews". Their sentences are in beta form - that's terrible english!

[[XC] gomeler]

Wonder if this will be ported over to AM3 quickly.. it would help AMD compete against Core i7, especially if it'll do 6GHz like current Phenom II X4 processors. Phenom II X6 955 anyone? :p:

[Macadamia]

So... spectime!

2.4-2.5Ghz 75/95W ACP/TDP
2.6-2.8Ghz 105/125W ACP/TDP

As drop in or heck, even new builds (never underestimate OEM deals) they are definitely fine, but I suspect that Sao Paolo (Istanbul on Socket G3) would shine more with the improved memory system (which IS useful on 2+S configs)

[informal]

AMD also added to their price list a 40W EE Shanghai models,several 55W ones and 3.1Ghz 105W SE model :). With 6 core ready for June availability they seem to be executing their new plan pretty well.

@ [XC] gomeler

6 core Phenom II is not out of the question but it will take some time before it makes it to market.My guess is Q3 at best. Before we have 6 cores on desktop I'd expect a 3.4-3.5Ghz QC desktop models by years end(even if it means they will carry the "140W" rating).

[YukonTrooper]

Any architectural changes to Istanbul besides the added cores?

[B.E.E.F.]

http://mypetjawa.mu.nu/archives/MOAR.jpg

[informal]

Any architectural changes to Istanbul besides the added cores?

Probe filter is one pretty significant change which can add upto 15% clock for clock increase (apart from 50% core count increase) over Shanghai based Opterons.This feature is important and relevant in multisocket systems.

[YukonTrooper]

Probe filter is one pretty significant change which can add upto 15% clock for clock increase (apart from 50% core count increase) over Shanghai based Opterons.This feature is important and relevant in multisocket systems.
15%? That's significant. What type of environments?

[informal]

The Istanbul demonstration has been covered pretty well by techreport (http://techreport.com/articles.x/16448).Probe filter is called "HT Assist" in AMD's terminology.


The third demo was the most interesting for a couple of reasons. First, because it was intended to show how Istanbul can serve as a drop-in upgrade for Socket F systems. The only requirements: the system must support split power planes, and it must have a BIOS upgrade to operate with the new processors. Second, the demo was impressive because it included a performance test. Two otherwise-identical systems were situated side by side: one with a quartet of Shanghai Opterons, the other with four Istanbul chips. Both systems were running with HyperTransport 3 active—a capability coming soon to Shanghai Opterons but not yet available in current products. To illustrate the performance difference between the two boxes, the AMD tech ran a Stream benchmark. The 16-core Shanghai system produced throughput numbers in the range of 25,000 MB/s. The 24-core Istanbul box, by contrast, hit about 42,000 MB/s. The tech then swapped the processor-and-memory daughtercards between the two boxes, and of course, the performance characteristics moved with them.

That's one heck of an in-place upgrade, but the bigger question may be: Why the huge performance gain with the addition of more cores, given that Stream is typically considered, at least partially, a bandwidth-bound benchmark? And why the magnitude of the gain, with only 50% more cores and (although they were not disclosed) likely lower per-core clock frequencies for Istanbul?
Part of the answer, it seems, may be a feature new to Istanbul that AMD calls HT assist (presumably for HyperTransport assist). This feature is what the company calls a probe filter (and may more commonly be called a snoop filter) that functions to reduce traffic on socket-to-socket HyperTransport links by storing an index of all caches and preventing unnecessary coherency synchronization requests. Current Opteron systems use a broadcast-based probe protocol, sending probe requests to all sockets. Istanbul, instead, either knows that no probes are required or is able to do a directed probe to a single socket. (Although it may still use broadcasts in certain, specific situations.) Istanbul's probe filter stores its data in the processor's L3 cache. The amount of cache space dedicated to probe filter storage, AMD says, will be configurable in the BIOS, and the more space dedicated to probe filter storage, the more granular its operation will be.

AMD didn't handicap the exact performance impact of HT assist for us, but the quad-socket Stream test may have been an extreme case. The probe filter capability will be unique to the six-core Istanbul and will not be incorporated in a future revison of Shanghai. However, with fewer cores, Shanghai Opterons will be able to reach higher clock speeds within the same power envelopes as Istanbul, and AMD expects the clock frequency advantage to somewhat offset the lack of a probe filter. Regardless, one would expect Istanbul to be especially popular in systems with four or more sockets, where coherency traffic is a thornier issue and HyperTransport bandwidth is at a premium.

[[XC] Oj101]

Completely off topic, but informal, is your avatar Photochopped? 6.25x muliplier?

[Macadamia]

Completely off topic, but informal, is your avatar Photochopped? 6.25x muliplier?

Quarter multipliers are doable on AMD K10 systems with the use of FID I think.

[Lightman]

Quarter multipliers are doable on AMD K10 systems with the use of FID I think.

Correct! :rolleyes:
They were present even on old Phenoms.

Back on topic:
I would like to have 6 cores desktop equivalent of Istanbul. If OCed to 6GHz+ on all cores it should produce quite nice CPU score in Vantage. Also improved memory controller should add a little bit.
I'm still wondering why it is so quiet about new RB-C3 revision. Back in 2008 it was mentioned on AMD slides but now it's dead quiet about it. Will it appear around same time frame as Istanbul core? I can only hope :)

[Hans de Vries]

Istanbul in May, in systems in June.

Deliverable from today:

Shanghai Opteron 8393, 3.1GHz 105W
Shanghai Opteron 8389, 2.9GHz 75W
http://products.amd.com/en-us/opteroncpuresult.aspx?f1=Third-Generation+AMD+Opteron%E2%84%A2&f2=&f3=&f4=512&f5=Socket+F+(1207)&f6=C2&f7=45nm+SOI&f8=&f9=2200&f10=4&f11=HT3&

The latter suggests an Istanbul Opteron at 2.9GHz, 105W (1.4x75W)
A Phenom X6 @ 3GHz would be very nice. (even at 140W)

At the server front it's all about bandwidth now. Socket G34/Maranello is
42.6 GB/s versus Nehalem at 32 GB/s versus Socket 1207 at 12.8 GB/s
versus Xeon Core 2 Quads at 10.67 GB/2.

It seems Maranello might be advanced, Probe Filter OK, HTC OK
If APML is OK as well then we might see a Sao Paolo / Magny Cours
demo today. The question of the day is:

Does Istanbul/Sao Paulo have a 288bit 4-channel DDR3 Memory
interface as the single die Sao Paolo in Socket G34 might suggest???

Just my old hobby:
http://www.chip-architect.com/news/Real_Sao_Paolo_Shanghai_Nehalem_1600x784_.jpg

Regards, Hans

[duploxxx]

Istanbul in May, in systems in June.

Deliverable from today:

Shanghai Opteron 8393, 3.1GHz 105W
Shanghai Opteron 8389, 2.9GHz 75W
http://products.amd.com/en-us/opteroncpuresult.aspx?f1=Third-Generation+AMD+Opteron%E2%84%A2&f2=&f3=&f4=512&f5=Socket+F+(1207)&f6=C2&f7=45nm+SOI&f8=&f9=2200&f10=4&f11=HT3&

The latter suggests an Istanbul Opteron at 2.9GHz, 105W (1.4x75W)
A Phenom X6 @ 3GHz would be very nice. (even at 140W)

At the server front it's all about bandwidth now. Socket G34/Maranello is
42.6 GB/s versus Nehalem at 32 GB/s versus Socket 1207 at 12.8 GB/s
versus Xeon Core 2 Quads at 10.67 GB/2.

It seems Maranello might be advanced, Probe Filter OK, HTC OK
If APML is OK as well then we might see a Sao Paolo / Magny Cours
demo today. The question of the day is:

Does Istanbul/Sao Paulo have a 288bit 4-channel DDR3 Memory
interface as the single die Sao Paolo in Socket G34 might suggest???

Just my old hobby:
http://www.chip-architect.com/news/Spec_Sao_Paolo_Shanghai_Nehalem_1600x784_.jpg

Regards, Hans

Roadmaps state Fiorano platform to be ddr2. Its Sao Paulo that should bring 4xHT and DDR3.

Istanbul is on this platform with 3 x ht3 (@4.4-4.8GT), but will add the APML, HT assist and IOMMU virtualization.

45NM yields must be very good and DVT from Istanbul has been introduced about 4months faster then schedule few months ago so that is very good, that is at least a small light in the dark.


So... spectime!

2.4-2.5Ghz 75/95W ACP/TDP
2.6-2.8Ghz 105/125W ACP/TDP

As drop in or heck, even new builds (never underestimate OEM deals) they are definitely fine, but I suspect that Sao Paolo (Istanbul on Socket G3) would shine more with the improved memory system (which IS useful on 2+S configs)

actually even a bit better spec :D the 105W starts from 2.8 and don't forget the HE parts.

[Hans de Vries]

Its Sao Paulo that should bring 4xHT and DDR3.


Sao Paolo and Istanbul are the same die.

Sao Paulo was Istanbul + probe filter + HTC + APML. The first two have been
reported working/demonstrated and are now brought to socket 1207 also.

http://techreport.com/r.x/2008q3/opteronroadmap.png

Regards, Hans.

[Hornet331]

ai.. to bad its a 8xxx series opteron, which puts it out of reach for mere mortals (price wise). :(

[duploxxx]

Sao Paolo and Istanbul are the same die.

Sao Paulo was Istanbul + probe filter + HTC + APML. The first two have been
reported working/demonstrated and are now brought to socket 1207 also.

http://techreport.com/r.x/2008q3/opteronroadmap.png

Regards, Hans.

not sure where you got this roadmap from, but i am afraid that it is no longer correct, actually from 08/2008 this was already wrong, i think this screenshot is dated from begin 2008.

there are way more differences in specs. To my opinion this is no longer the same die, but Magny-cours- sao paulo do fit together as a die.


ai.. to bad its a 8xxx series opteron, which puts it out of reach for mere mortals (price wise). :(

it's 2/8000 series but would love to see a FX from this :D

[EniGmA1987]

Doesnt Istambul have a DDR3 memory controller? Shanghi is the server version of Phenom 2, so the improvement called Istambul should have it. So the only reason we cant use DDR3 on a server now would be the motherboard RAM slots are the DDR2 slots, right? Maybe ASUS will make a nice dual socket workstation MB and throw in some DDR3 slots for fun :)

[duploxxx]

Doesnt Istambul have a DDR3 memory controller? Shanghi is the server version of Phenom 2, so the improvement called Istambul should have it. So the only reason we cant use DDR3 on a server now would be the motherboard RAM slots are the DDR2 slots, right? Maybe ASUS will make a nice dual socket workstation MB and throw in some DDR3 slots for fun :)

well the 1 socket Buenos Aires with the new chipsets will support am3 so yes the normal ddr3 is there, just not activated i assume. However this is still a big difference with U/Rddr3 which is in 2p/4p servers

[informal]

Completely off topic, but informal, is your avatar Photochopped? 6.25x muliplier?

I see you already got your answer :),but no,it's not photochopped.It's done by OCing and having CnQ on... And it's leet :D.

[demonkevy666]

phenom II x6 would need another bios change on boards.

I don't' see why the can't make a bios on the board for MCM phenom II x8.

[ubuntu83]

How about 32nm Phenom II X6 compatible with AM3 socket?

[Mechromancer]

Is the memory controller on this hex-core DDR2/DDR3 like the current AM3 Phenoms? I'd love this to be backwards compatible to AM2+. I just built this system in January and would like an upgrade path to 4+ cores, instead of only 4. Why? Just for E-peen. My system doesn't even have to be upgraded for 3 years, but I'm a member of XS which means I do things in EXCESS!

[Hans de Vries]

not sure where you got this roadmap from, but i am afraid that it is no longer correct, actually from 08/2008 this was already wrong, i think this screenshot is dated from begin 2008.

there are way more differences in specs. To my opinion this is no longer the same die, but Magny-cours- sao paulo do fit together as a die.

This is the roadmap which anounces the Fiorano platform which was in
the middle of August 2008

http://www.eweek.com/c/a/IT-Infrastructure/New-AMD-Server-Platform-Due-in-2009/

Istanbul in this road map is the same six-core die as Sao Paolo but with
some new features disabled. Some of these features have been demonstrated
working already such as the probe filter which is now also included in the
Istanbul to be released in May,

Sao Paolo is one die with six cores in a 1974 pin G34 package.
Magny Cours is 2 interconnected Sao Paulo dies in a G34 package


it's 2/8000 series but would love to see a FX from this :D

:)

Regards, Hans

[informal]

AMD webcast said that AMD is already demonstrating (live!) Magny-Cours system to the interested parties.This is amazing news since G34 and MCM is practically done deal and sampling with partners in Q4 with product launch in Q1 2010! Also,they mentioned the "Interlagos" CPU which is a Bulldozer core(12-16 cores on 32nm).Also mentioned is a so called segmentation of sockets with G34 being the socket for high perf. 2P-4P(8-16 cores per socket,45nm and 32nm with both present and future uarch. supported) while C32 socket is for energy efficient 1P-2P systems (4-8 cores per socket,45nm and 32nm with both present and future uarch. supported).

I will post some screenshots from the webcast.For now ,a real die shot of Istanbul may serve as a teaser :D :
http://img91.imageshack.us/img91/6990/1240423884310.jpg

[YukonTrooper]

I wonder how it will compare to i7 w/HT in a crunching environment. That's what I'm most interested in. I feel the addiction of crunch coming on and I'll be needing the best bang-for-your-buck CPU's.

[Face]

More slides here:

http://www.pcper.com/comments.php?nid=7062

http://img178.imageshack.us/img178/7802/bulldozerperf.jpg


And this little gem was near the end. AMD is very bullish about the performance of the Bulldozer architecture that they will introduce in 2011. Note the faded line, as they are not entirely sure where it will end up... but they believe that it will be very competitive to what Intel has at that time. Note that "Interllagos" is a 16 core product.

[informal]

The pic of Magny-Cours CPU that's being also demoed today!
http://img230.imageshack.us/img230/6303/clip2g.jpg

Planned DCA roadmap for 2010:
http://img222.imageshack.us/img222/7397/clipn.jpg
Projected improvement for Magny-Cours over 2009 Istanbul :
http://img15.imageshack.us/img15/2504/clip3w.jpg

Improvements to AMD-V and AMD-P in 2010:
http://img222.imageshack.us/img222/5108/clip5.jpg

A 2010 Maranello(G34) and San Marino(C32) platforms supporting next gen BD cores(apart from supporting Sao Paolo and Magny-Cours):
http://img222.imageshack.us/img222/5815/clip6.jpg
http://img15.imageshack.us/img15/7992/clip7.jpg


Bulldozer is scheduled for Q1 2011 launch(servers and desktop) with projected int improvement of 1.35-1.4x over Shanghai.Core count will be in range of 12-16 for G34 platform(6000 series) and 4-8 for C32(4000 series).Here is the projected speed increase over Magny Cours(12 cores).AMD predicts the jump from Magny Cours to Bulldozer based "Interlagos" cores will be as substantial(or even bigger) than a jump from Istanbul to Magny Cours,relatively speaking:
http://img145.imageshack.us/img145/9599/clip8t.jpg

Last slide is self explanatory :) (Q1 '01 launch for Maranello platform ;)):
http://img27.imageshack.us/img27/9238/clip9m.jpg

[lockee]

So Sao Paolo = 2 Shanghai MCM, Magny Cours = 2 Istanbul MCM?

[spicypixel]

12 Core MCM ship sampling already? Thats pretty impressive.

[informal]

So Sao Paolo = 2 Shanghai MCM, Magny Cours = 2 Istanbul MCM?

I think that Sao Paolo=Istanbul for G34 while Magny-Cours will have from 8 to 12 cores as can be seen below:
http://img15.imageshack.us/img15/7992/clip7.jpg

[Hans de Vries]

I think that Sao Paolo=Istanbul for G34 while Magny-Cours will have from 8 to 12 cores as can be seen below:


Sao Paulo seems to have disappeared as a codename somehow...
There's now Interlagos instead, the F1 circuit in Sao Paulo.

Regards, Hans.

[informal]

Yep,but new stuff is that "Interlagos" is now a BD based MPU and done on 32nm while featuring up to 16 cores(we have no frequency targets but we have a rough estimate about throughput performance(i guess spec rate tests) from here (http://www.amdzone.com/phpbb3/viewtopic.php?p=156737&sid=06dc601a7f5255ffc092287a6df16f34#p156737);this is user generated chart on the basis of the official chart):
http://www.enumae.com/images/PBD-SPEC.jpg

Note in the chart the increase in spec_int rate(?) is a lot higher(~190) when compared to Shanghai model at ~113 giving us ~1.7x increase ,while spec fp rate(?) is 1.38x better than in Shanghai's case(memory bw limited maybe?).

Also,we now have a Cinebench10 x64 (yuk :p: ) score for Istanbul (http://blogs.amd.com/work/2009/04/22/amd-opteron-6th-anniversary-press-kit/),running side by side with Shanghai 2P system and running within the same power envelope as Shanghai system so I guess it's 2.4Ghz and 75W rated ES part compared to 2.7Ghz 75W Shanghais in the second system:

Direct linky: http://www.youtube.com/watch?v=XDdvZkBo4JE

2P 2.4Ghz or 2.5Ghz(estimated to fit in 75W ACP envelope) Istanbul scores 25863(34s)
2P 2.7Ghz(75W ACP envelope,since we know it is 2.7Ghz by looking here (http://techreport.com/articles.x/15905/7)) scores 18619(47s)

The net result is 38% better performance with ~12 or 8% lower clock speed,depending if the 75W ACP ES Istanbul models were running at 2.4Ghz or 2.5Ghz.This translates to linear speed increase of 1.5-1.56 over the same clock Quad Core Shanghai system which is expected since it has exactly 50% more cores.

[Bo_Fox]

16 cores on a 32nm processor, already??? Wow!!!

Would we also be seeing 16 cores on desktop CPU's? I mean, not only for servers, right?

[duploxxx]

Yep,but new stuff is that "Interlagos" is now a BD based MPU and done on 32nm while featuring up to 16 cores(we have no frequency targets but we have a rough estimate about throughput performance(i guess spec rate tests) from here (http://www.amdzone.com/phpbb3/viewtopic.php?p=156737&sid=06dc601a7f5255ffc092287a6df16f34#p156737);this is user generated chart on the basis of the official chart):


Note in the chart the increase in spec_int rate(?) is a lot higher(~190) when compared to Shanghai model at ~113 giving us ~1.7x increase ,while spec fp rate(?) is 1.38x better than in Shanghai's case(memory bw limited maybe?).

Also,we now have a Cinebench10 x64 (yuk :p: ) score for Istanbul (http://blogs.amd.com/work/2009/04/22/amd-opteron-6th-anniversary-press-kit/),running side by side with Shanghai 2P system and running within the same power envelope as Shanghai system so I guess it's 2.4Ghz and 75W rated ES part compared to 2.7Ghz 75W Shanghais in the second system:

Direct linky: http://www.youtube.com/watch?v=XDdvZkBo4JE

2P 2.4Ghz or 2.5Ghz(estimated to fit in 75W ACP envelope) Istanbul scores 25863(34s)
2P 2.7Ghz(75W ACP envelope,since we know it is 2.7Ghz by looking here (http://techreport.com/articles.x/15905/7)) scores 18619(47s)

The net result is 38% better performance with ~12 or 8% lower clock speed,depending if the 75W ACP ES Istanbul models were running at 2.4Ghz or 2.5Ghz.This translates to linear speed increase of 1.5-1.56 over the same clock Quad Core Shanghai system which is expected since it has exactly 50% more cores.

look at the demo again @1min17sec he states that they run identical speed and thermal enveloppe


I think that Sao Paolo=Istanbul for G34 while Magny-Cours will have from 8 to 12 cores as can be seen below:


if they go the MCM way it is clear that a magny 8 core is a double 4 core lisbon and same for the 6-12 mcm

[mstp2009]

Nice to see AMD executing. This is what we all like to see: competition.

[informal]

look at the demo again @1min17sec he states that they run identical speed and thermal enveloppe

Look at the score of the Shanghai system(18619 pts).It is the score that only a 2.7Ghz 2P Shanghai can produce as can be seen in the techreport link here (http://techreport.com/articles.x/15905/7).
If you imply that AMD managed to squeeze out 2.7Ghz from a six core Istanbul and still fit it in 75W then i must disagree(common logic at work).Plus the score would be roughly 1.5x better than the SHanghai score if they were running at the exact same clocks and wouldn't be 1.38x better(which fits perfectly with 8-12% lower clocks on Istanbul).

Maybe Ron Myers made a mistake and said that while he meant same power envelope(which he stated several times during a demo).

[Lightman]

Look at the score of the Shanghai system(18619 pts).It is the score that only a 2.7Ghz 2P Shanghai can produce as can be seen in the techreport link here (http://techreport.com/articles.x/15905/7).
If you imply that AMD managed to squeeze out 2.7Ghz from a six core Istanbul and still fit it in 75W then i must disagree(common logic at work).Plus the score would be roughly 1.5x better than the SHanghai score if they were running at the exact same clocks and wouldn't be 1.38x better(which fits perfectly with 8-12% lower clocks on Istanbul).

Maybe Ron Myers made a mistake and said that while he meant same power envelope(which he stated several times during a demo).

At the end of the demo it says 2x2.6GHz Opteron 2382.
I believe AMD squeezed 2.6GHz Istanbul into 75W ACP.
Cinebench scaling is not linear by any means and 1.38x from 50% more cores is absolutely spot on with what I would expect. Remember that core improvements are mostly for bandwidth intense scenarios. Cinema4D is far from bandwidth hungry application, what counts there is pure math processing.

If 2.6GHz is under 75W then 2.9-3.0GHz Istanbul within 105W ACP should be possible very soon. Give AMD 2-3 months to stock pile better cores before release. :up:

[informal]

Thanks Lightman,2.6Ghz it is then.Probably a bit better score on thse than techreport managed with 8384s back in the day.
If 2.6Ghz Istanbul fits in 75W ACP envelope than,as you said,105W envelope can bring as high as 2.9Ghz SKUs.Quite an achievement :up:

[v_rr]

That 12 core Magny-cours should be a 600mm2 monster :D
Istanbul should be arround 300mm2.

Anyway AMD is ramping more cores and more clock at a good rhythm. If they started this 6 months earlier.....

[DoubleZero]

Bulldozer seems meh


Note the faded line, as they are not entirely sure where it will end up...

Hope so, cause until the faded part the numbers are less then stellar.

[informal]

Bulldozer seems meh

Hope so, cause until the faded part the numbers are less then stellar.

Actually the numbers are quite good,given the fact we have no frequency target projection for BD cores.But all this is "projected" so we will have to wait and see.

Before BD arrives,a picture of a socket G34 server running 48 cores in 4P configuration(a Magny-Cours in action :D):

http://www.flickr.com/photos/amd_unprocessed/sets/72157617192384144/

http://farm4.static.flickr.com/3582/3465930377_57842cc0f4.jpg?v=0

48 real cores in task manager :D,looks like povray is running in background.

[Macadamia]

Holy **** in every aspect faster (timeline) and better (CPUs) than I thought.

Looks like AMD's been much more smoother with silicon lately. Shanghai pulled ahead, Istanbul launching much sooner, Magny Cores launching in Q1 instead of a vague 1H 2010.

Oh, and the new platform is even sexier. Crossgrades FTW.

Now that there is no more seperate 1-socket Opteron platform, 2P CPUs should come close/closer to 1P prices, which was supposed to be the way it is since Gainestown pricing is also rather cheap.

[informal]

"Dark Major" wasn't kidding when he said AMD was about to do some major regrouping inside the company^^ :).They are indeed firing on all cylinders.Ahh,just look at that task manager hehe :D.

[B.E.E.F.]

Oh, and the new platform is even sexier. Crossgrades FTW.

What is that :confused:.

[lockee]

Istanbul is running at 2.6GHz here.

http://www.youtube.com/watch?v=XDdvZkBo4JE

edit; already posted above

[Lightman]

I love to play with that 4P Magny-Course system :shocked:

BTW where is Movieman? Has he seen this pic already? Is he alright? :p:

[kl0012]

So Istanbul still can't reach in specint/specfp even Xeon E5530 (2.4GHz, 80W TDP) which has score 192/158 for dual socket cfg.
And seems that Magny having 3x cores more then Xeon W5580 not much faster in specint - 290 vs. 260

[NightCrawler™]

One has got to love those race track names... Seems AMD is up for a race too.. :D
They have definitely picked up a good path silicon wise.. I never expected to see them go for the hex core this fast..

[duploxxx]

So Istanbul still can't reach in specint/specfp even Xeon E5530 (2.4GHz, 80W TDP) which has score 192/158 for dual socket cfg.


that's because spec benchmarks like HT and the turbo mode:), so that's 6 real core against 4+4 turbo in the same tdp package. Will i choose reality or spec, hmm difficult choice i'll just beta both and see what's best for my applications and still istanbul is residing on the old platform so this is holding the performance back.


So Istanbul still can't reach in specint/specfp even Xeon E5530 (2.4GHz, 80W TDP) which has score 192/158 for dual socket cfg.
And seems that Magny having 3x cores more then Xeon W5580 not much faster in specint - 290 vs. 260

not much faster? keep in mind this is still the same k10 based architecture we are talking about, the latest k10 version was only able to get close to the old microarchitecture we known it lacks int against new intel from the beginning.

Since there are isn't much known data from magny i would hold the comments here, it's not 3x more cores again the HT advantage in spec benchmarks, which is a full core for that.
it is still 15% faster specint performance at a much lower clock speed and normal rated tdp level but off course you can expect that Nehalem will have a few higher bins on 32nm, for 45nm they probably have 1 more speedgrade in future but then again from 2,4 to 3,2 ghz the specint increased from 192 to 260, that is linear clockspeed/specint increase so they will require a 3,6-3,7Ghz for the same specint performance then the shown magny.


This is the roadmap which anounces the Fiorano platform which was in
the middle of August 2008

http://www.eweek.com/c/a/IT-Infrastructure/New-AMD-Server-Platform-Due-in-2009/

Istanbul in this road map is the same six-core die as Sao Paolo but with
some new features disabled. Some of these features have been demonstrated
working already such as the probe filter which is now also included in the
Istanbul to be released in May,

Sao Paolo is one die with six cores in a 1974 pin G34 package.
Magny Cours is 2 interconnected Sao Paulo dies in a G34 package



:)

Regards, Hans

Hans since you now have seen the core design, any clue on possible die size of istanbul?

[enumae]

Please note that this slide I made is my own, and the numbers are not based on anything provided by AMD.

http://www.enumae.com/images/PBD-SPEC.jpg

The numbers are based on scaling the slide to the highest recorded result recorded at SPEC for 2008, which was the Opteron 2384. These numbers could be a decent projection, but they may not be.

[kl0012]

off course you can expect that Nehalem will have a few higher bins on 32nm, for 45nm they probably have 1 more speedgrade in future but then again from 2,4 to 3,2 ghz the specint increased from 192 to 260, that is linear clockspeed/specint increase so they will require a 3,6-3,7Ghz for the same specint performance then the shown magny.


Probably it will have to complete with Beckton 8 core (16 threads). I suppose it will be a hard time for Magny.

[Jacky]

"Dark Major" wasn't kidding when he said AMD was about to do some major regrouping inside the company^^ :).They are indeed firing on all cylinders.Ahh,just look at that task manager hehe :D.
I am very positively surprised by those events, but so far it's still (mostly) slide-ware and early server silicon being shown off. A lot can still go wrong. While it looks rather promising, we should wait and see.
I'm assuming they still need to deliver on the desktop and notebook front if they want to get out of the red, by the way.

[Macadamia]

AMD also said that their dual core Congo platform is shipping this quarter (late Q2), just when Intel's single core CULVs come to market.

[Nedjo]

die:

http://farm4.static.flickr.com/3657/3463961733_79c8a0e4ff_b.jpg http://farm4.static.flickr.com/3521/3464778308_f3c5bf832a.jpg

again Hans proved being right:

http://www.chip-architect.com/news/AMD_6_core_Istanbul.jpg

all that power under 300 sqmm!

[perkam]

Thx for the slides.

For all those confused. Sao Paolo is now called Lisbon, and two of these make a Magny Cours.

Perkam

[hurleybird]

Not quite. Core 4 and 5 are backwards ;)

[Helmore]

Not quite. Core 4 and 5 are backwards ;)

And he initially had a different design, where the cache was in the middle with the cores surrounding it, if I remember correctly.

[mAJORD]

hmm, well not far off my prediction / mockup back in Feb ;)

http://img152.imageshack.us/img152/8784/istanbulfinalcr.th.jpg (http://img152.imageshack.us/my.php?image=istanbulfinalcr.jpg)

http://www.xtremesystems.org/forums/showthread.php?t=218718&highlight=istanbul

They Made the die wider and shorter than my mock-up though with L3 going right to the wider edges. and core 5-6 sitting out for whatever reason..