Istanbul (6 core Opteron) to be released this week

[Viditor]

Softpedia

"According to a recent report, the chip maker is planning the release of 'Istanbul' for later this week, which will allow it to gain more market share from its main rival, Intel. The report suggests that the company's first 6-core Opteron processors will be launched as part of the new Opteron 2400 series"

Fudzilla

"AMD plans to launch Calisto dual core K10.5 45nm in one week, on the first day of Computex. Sources confirmed that AMD plans to launch Phenom II X2 550 Black edition at Computex and this CPU runs at 3.1GHz and has a massive 7MB total cache size. The CPU is Black Edition which means it will come with an unlocked multiplier and you should be able to overclock it nice and easy. The launch will take place on Tuesday June 2nd, in Taiwan and the rest of the world"

 

[Idontcare]

Istanbul's gross margins are ensured, no comment there.

But what kind of GM's is a dual-core Phenom II X2 550 going to have? Are they all harvested, meaning functional yields for the X4's are really that bad, or is AMD actually churning 250mm^2 chips to sell for $100 X2's?

 

[soccerballtux]

Originally posted by: Idontcare
Istanbul's gross margins are ensured, no comment there.

But what kind of GM's is a dual-core Phenom II X2 550 going to have? Are they all harvested, meaning functional yields for the X4's are really that bad, or is AMD actually churning 250mm^2 chips to sell for $100 X2's?

Do you work anymore? 😛 I'm just worried you're spending too much time on these boards.

This sounds very motherly. I do not mean to be your mother 🙂

 

[Idontcare]

Benefits of working for oneself :thumbsup: I get done what I want and when I want. As it is now, homebound with a 2yr old, there's only so much playdoh playing I can do without going stir crazy. The forums are definitely a convenient way to interact with people older than 2yrs in the 2-3 minute increments my kids give me here and there throughout the day.

 

[Phynaz]

Mr. Mom!

 

[Phynaz]

Fudzilla

"AMD plans to launch Calisto dual core K10.5 45nm in one week, on the first day of Computex. Sources confirmed that AMD plans to launch Phenom II X2 550 Black edition at Computex and this CPU runs at 3.1GHz and has a massive 7MB total cache size. The CPU is Black Edition which means it will come with an unlocked multiplier and you should be able to overclock it nice and easy. The launch will take place on Tuesday June 2nd, in Taiwan and the rest of the world"[/quote]

You're a little late on the trigger there. The specs/benches of the 550/250 were posted last week.

 

[soccerballtux]

Originally posted by: Idontcare
Benefits of working for oneself :thumbsup: I get done what I want and when I want. As it is now, homebound with a 2yr old, there's only so much playdoh playing I can do without going stir crazy. The forums are definitely a convenient way to interact with people older than 2yrs in the 2-3 minute increments my kids give me here and there throughout the day.

Not so sure about that; 😀 have you been to Off Topic or L&R recently? CPUs seems pretty insulated though 🙂

 

[alyarb]

I don't see how AMD can keep selling these 90+ watt dual cores. I understand harvesting is great, but are their yields so bad that they can get an entire product line out of trashed quads? I'd much rather look into a 45nm K8 with big L2.

 

[lopri]

Originally posted by: alyarb
I don't see how AMD can keep selling these 90+ watt dual cores. I understand harvesting is great, but are their yields so bad that they can get an entire product line out of trashed quads? I'd much rather look into a 45nm K8 with big L2.

I first thought these dual-core K10's were quad rejects, but now I do not know until reviews are out. Istanbul is a 6-core / 6MB L3 part. So that naturally makes me wonder where the other 3MB of L3 is? From the look of Deneb wafer, it looks like there is 3MB of L3 per every two core so it could be that AMD can chop the wafer whichever way it wants. If that's the case these dual-cores may not be quad-rejects.

Back on Instanbul: IMO it is a pointless CPU at least on current platform. I was shocked to learn that majority of AMD's server platform (Socket F) is still nForce (!) with HT 1.0. There surely are codes that can benefit from more cores, but for majority of server apps cores need to be backed by enough bandwidth. Until HT 3.0 and DDR3 arrive on AMD's server platforms, these 6-core parts will be severely bottlenecked. I guess server people are really not an upgrading type? HT 2.0/3.0 have been out for years on desktop..

Edit: On a slight off-topic, I keep hearing this 'mysterious' performance boost on current Phenom II. The mysterious performance improvement is supposed to be a software thing, or so I heard, but I wonder how such a 'fix' is possible? Does anyone have any inside-info? (It may well be a BS, though)

 

[Idontcare]

Originally posted by: lopri
I first thought these dual-core K10's were quad rejects, but now I do not know until reviews are out. Istanbul is a 6-core / 6MB L3 part. So that naturally makes me wonder where the other 3MB of L3 is? From the look of Deneb wafer, it looks like there is 3MB of L3 per every two core so it could be that AMD can chop the wafer whichever way it wants. If that's the case these dual-cores may not be quad-rejects.

Lopri you really lost me here.

What other 3MB of cache are you talking about? 😕

Deneb has a unified and shared 6MB of L3$ which is fully accessible by all cores, same as Istanbul.

In addition to the shared L3$, each core has a dedicated 512KB of L2$ which is not shared with the other cores (although its contents are accessible via snoops, which is where Istanbul's new improvements come in).

Originally posted by: lopri
Back on Instanbul: IMO it is a pointless CPU at least on current platform. I was shocked to learned that majority of AMD's server platform (Socket F) is still nForce (!) with HT 1.0. There surely are codes that can benefit from more cores, but for majority of server apps cores need to be backed by enough bandwidth. Until HT 3.0 and DDR3 arrive on AMD's server platforms, these 6-core parts will be severely bottlenecked. I guess server people are really not an upgrading type? HT 2.0/3.0 have been out for years on desktop..

Edit: On a slight off-topic, I keep hearing this 'mysterious' performance boost on current Phenom II. The mysterious performance improvement is supposed to be a software thing, or so I heard, but I wonder how such a 'fix' is possible? Does anyone have any inside-info? (It may well be a BS, though)

Actually you highlight one of the hardware improvements in Istanbul which have been highlighted by AMD as a reason to upgrade to Istanbul. The improved snoop filter actually enables the server to do even more with less HT bandwidth.

 

[ShawnD1]

Originally posted by: lopri
I guess server people are really not an upgrading type? HT 2.0/3.0 have been out for years on desktop..

Correct. One place I worked for in 2007 still had a lot of Pentium 3 computers with Windows NT 4. Our main Citrix server was using a dual 2ghz Xeon with a whopping 8gb of ram (spread across 50+ simultaneous logins). It was actually a good setup. I never bitched about lag or slow load times. Something like Crysis at max settings uses more CPU and memory power than 50 people using Citrix Desktop.

Another thing is that socket compatibility is a big deal when it comes to servers. Changing the CPU is a simple upgrade with no software changes, but changing a motherboard opens the door to a whole mess of problems. Have you heard of people needing to reinstall Windows after changing the motherboard, or how not doing so can cause blue screens and boot errors? Stuff like that can be the deciding factor on whether or not the server should be upgraded.

 

[lopri]

I was thinking of something like this? I don't know about engineering so it's a pure arm-chair guess.

http://spreadsheets.google.com...Jb_vtwANuQ&output=html

 

[lopri]

ShawnD1: LOL (@PIII in 2007) thanks for sharing a real-world experience. Yes, it's understandable that platform changing is a huge hassle in server world. In contrast, it's so ironic that we have HT 3.0 on desktop (which really does nothing for performance, other than compatibility)

 

[Idontcare]

Originally posted by: lopri
I was thinking of something like this? I don't know about engineering so it's a pure arm-chair guess.

http://spreadsheets.google.com...Jb_vtwANuQ&output=html

Wait, are you meaning more than one type of chip on the same wafer?

As in the same wafer has both Istanbul chips as well as Deneb chips at the same time?

If that is what you mean then yeah, it doesn't happen, ever. Way too complicated for mask-sets and layout to do that. Every chip gets its own maskset and dedicated wafer.

 

[lopri]

Yeah that's what I was thinking. Since that's not possible, the dualies are quad-rejects, then.

 

[alyarb]

Why are kuma 65nm though, and why are they being released just now? It's like AMD created an ocean of failed Agena chips and now they want to sell them to us.

 

[richierich1212]

Originally posted by: lopri
Yeah that's what I was thinking. Since that's not possible, the dualies are quad-rejects, then.

Not the 250, (Regor) since it has 2 x 1mb L2 cache and no L3 cache. 550, maybe.

 

[alyarb]

we'd have to see a die.

 

[BLaber]

From what I have read Regor and Callisto are not PII x4 gone bad & harvested in to X2's , they are different dies.But no links to conform this yet.

 

[Phynaz]

Originally posted by: BLaber
From what I have read Regor and Callisto are not PII x4 gone bad & harvested in to X2's , they are different dies.But no links to conform this yet.

That is my understanding also.

 

[Idontcare]

Originally posted by: Phynaz

Originally posted by: BLaber
From what I have read Regor and Callisto are not PII x4 gone bad & harvested in to X2's , they are different dies.But no links to conform this yet.

That is my understanding also.

Derived from the quad-core Deneb processor that has two disabled cores, Callisto is AMD?s latest, highly-anticipated 45nm dual-core offering. AMD is expected to unveil two processors ? the Phenom II X2 550 and the Phenom II X2 545 ? based on this platform at Computex.

http://www.overclock3d.net/new...om_ii_x2_at_computex/1

Could be complete BS, but I have yet to see AMD make any comments to the contrary.

So really all we have to go on is the known policy of harvesting X3's from X4 die with expectation that X2's could be made the same way versus our desire for logic to prevail and have the X2's be smaller die with strictly dual-cores in the maskset.

Until I see AMD say otherwise, I am inclined to believe they are doing what they know works, ala the X3's.

 

[heyheybooboo]

Originally posted by: lopri
ShawnD1: LOL (@PIII in 2007) thanks for sharing a real-world experience. Yes, it's understandable that platform changing is a huge hassle in server world. In contrast, it's so ironic that we have HT 3.0 on desktop (which really does nothing for performance, other than compatibility)

AMD server chipset "Fiorano"

Fiorano is consisted out of AMD's own SS7100 Southbridge and SR5690 or SP5100 Northbridges, depending on how many Sockets are present on the motherboard. You'll notice that these chips are based upon existing desktop platforms, e.g. Spider and Dragon. SR5690 is actually a server chipset building on features implemented in 790FX and GX chips, bringing support for IOMMU - key feature for seamless virtualization [hardware virtualization support - virtual machines can isolate their own hardware], while SP5100 does not feature IOMMU. We expect that SP5100 will target workstations, while SR5690 will head to the server arena.

On the other side, SS7100 Southbridge shares similarities with SB750 Southbridge chip - SATA 3Gbps, USB 2.0, eSATA are all supported. As a native part of Fiorano platform, supported processors [Socket F, e.g. 1207] will work either on HyperTransport 1.1 or latest 3.0. In the case of Opterons, we suspect that AMD beefed up the HT base clock, expanding the available bandwidth beyond the desktop platforms and their Phenom II processors.

Long Live Socket F ! 😀

Rolls out next month with Istanbul. IIRC there are additional HT links directly between CPUs. It will be interesting to see how advances in NUMA work out with the new AMD/Intel enterprise platforms. That's where instructions are critical. Page faults are typically off the charts when a CPU has to address a DIMM bank other than its own.

4-channel DDR3 is supposed to roll with the new Socket G34 1H2010 with the six-core "Sao Paolo" or twelve-core "Magny-Cours" but I'm not sure where it really stands.

 

[LoneNinja]

Phenom II (Calisto) is harvested from Deneb chips, that explains the massive L3 cache. Athlon II (Regor) is a different die that doesn't possess any disabled cores, 1mb L2 cache per core instead of 512kb, and is a 65W part. AMD made it public knowledge before Phenom II even hit the market that the new 45nm Athlons to launch in the 2Q09 were to be native dual cores rather than quad rejects.

 

[Phynaz]

Originally posted by: LoneNinja
Phenom II (Calisto) is harvested from Deneb chips, that explains the massive L3 cache. Athlon II (Regor) is a different die that doesn't possess any disabled cores, 1mb L2 cache per core instead of 512kb, and is a 65W part. AMD made it public knowledge before Phenom II even hit the market that the new 45nm Athlons to launch in the 2Q09 were to be native dual cores rather than quad rejects.

I think we have a winner 🙂

 

[imported_SLIM]

Originally posted by: Phynaz

Originally posted by: LoneNinja
Phenom II (Calisto) is harvested from Deneb chips, that explains the massive L3 cache. Athlon II (Regor) is a different die that doesn't possess any disabled cores, 1mb L2 cache per core instead of 512kb, and is a 65W part. AMD made it public knowledge before Phenom II even hit the market that the new 45nm Athlons to launch in the 2Q09 were to be native dual cores rather than quad rejects.

I think we have a winner 🙂

This was my understanding as well.