Shanghai coming in October

[Idontcare]

AMD 45nm Opteron debut in October ahead of schedule to return to the server chipset market is no longer dependent on third parties

AMD originally planned to release in January 2009 the new 45nm server processor Shanghai, advanced to the final in mid-October, including a 2 way Opteron server processor 2376 (2.3GHz/2.0GT/s), 2378 (2.4GHz / 2.0GT / s), 2380 (2.5GHz/2.0GT/s), 2382 (2.6GHz/2.0GT/s) and 2384 (2.7GHz/2.0GT/s), as well as the 8-way server Opteron processor 8378 (2.4 GHz/2.0GT/s), 8380 (2.5GHz/2.0GT/s), 8382 (2.6GHz/2.0GT/s) and 8384 (2.7GHz/2.0GT/s) a total of 9 models using Socket F (1207 ), Built-in DDR2 memory controller, built-in 6MB L2, are all 75W TDP.

http://www.google.com/translat...ie=UTF8&sl=zh-CN&tl=en

 

[myocardia]

That will be good news for AMD's financial situation, at least. Well, hopefully it will be. It looks to me as if they're doing their best to "hook" a few customers, before Nehalem shows up. Am I the only one who thinks that?

 

[Zstream]

About damn time, I need to upgrade a few servers.

 

[Idontcare]

Originally posted by: myocardia
That will be good news for AMD's financial situation, at least. Well, hopefully it will be. It looks to me as if they're doing their best to "hook" a few customers, before Nehalem shows up. Am I the only one who thinks that?

Probably won't see volume shipments until 1Q09 though. So more like a confidence boost to would-be investors and corporate customers that they can deliver to their roadmaps even in these dire times.

New process node (45nm) and new CPU (Shanghai) spells double-trouble for ramp rates...consider how long it took them to ramp up to volume (hmmm, did they ever?) shipments of K10 last fall and that was with a mature 65nm process.

Hopefully we'll get an early Deneb preview based on initial reviews with Shanghai chips. Remember when Anand did this with Barcelona to give us a heads-up on how piss-poor Phenom's performance was going to be relative to the higher clocked Kentsfields at the time?

Only question on my mind about K10.5 is will the L3$ clockspeed be synchronous/asynchronous and how bad will the latency be?

 

[nerp]

I really really hope this works out well for AMD. They need to be as competitive as possible to survive this critical period. With the way the economy is looking right now, big corporate expenditures on technology might not be so bright a spot on the horizon.

 

[Viditor]

We should keep in mind that this is still just rumour...
That said, IF IT'S TRUE, the other part that's a very nice move for AMD is that they are finally putting out a new server chipset.

I found a complete listing (also just rumour) of the new chips over at xtremesystems:

Even numbers = HT1.0, Odd numbers = Newer Shanghais, HT3.0. They will come with the SR5600 Chipset in Q209.


AMD Opteron 8000 Series

Clock L2 Cache L3 Cache Core Socket B/W RAM Date due
105W
8386SE2.8GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Feb, 2009

75W
8385 2.7GHz 4x512KB 6MB 4 F(1207) 4.4GT/s DDR2 Q2, 2009
8384 2.7GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2008
8383 2.6GHz 4x512KB 6MB 4 F(1207) 4.4GT/s DDR2 Q2, 2009
8382 2.6GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2008
8380 2.5GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2008
8378 2.4GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2008

55W
8376 HE 2.3GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Feb, 2009
8374 HE 2.2GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Feb, 2009

AMD Opteron 2000 Series

Clock L2 Cache L3 Cache Core Socket B/W RAM Date due
105W
2386 SE 2.8GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Feb, 2009

75W
2385 2.7GHz 4x512KB 6MB 4 F(1207) 4.4GT/s DDR2 Q2, 2009
2384 2.7GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2009
2383 2.6GHz 4x512KB 6MB 4 F(1207) 4.4GT/s DDR2 Q2, 2009
2382 2.6GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2008
2380 2.5GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2008
2378 2.4GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2008
2376 2.3GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Oct, 2008

55W
8376 HE 2.3GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Feb, 2009
8374 HE 2.2GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Feb, 2009
8372 HE 2.1GHz 4x512KB 6MB 4 F(1207) 2GT/s DDR2 Feb, 2009


Sorry, forgot the link.

 

[VirtualLarry]

So much for this then. As I expected.

 

[Viditor]

Originally posted by: VirtualLarry
So much for this then. As I expected.

Keep in mind that
1. these are rumours still
2. these are Shanghai based Opterons and not desktop based Denebs

Edit:I'm not agreeing with him, just saying that nobody really knows anything yet...

 

[Idontcare]

I am impressed that with AMD's 45nm process they are seeing good enough yields in the power consumption range of 75W TDP that they are willing to list a 2.7GHz SKU at that TDP.

 

[soccerballtux]

markets seem to like it, AMD is up.

 

[BLaber]

I think Shanghai and Deneb could be what doctor ordered for AMD...

 

[OCGuy]

Originally posted by: BLaber
I think Shanghai and Deneb could be what doctor ordered for AMD...

From leaked benchies.....not really. Now theres a chance that translated chinese benchmarks can be wrong, obviously......but from what ive seen, they will still be behind Yorkfield, and not even closet to i7. Unless that is what AMD wants everyone to think.

 

[BLaber]

Originally posted by: Ocguy31

Originally posted by: BLaber
I think Shanghai and Deneb could be what doctor ordered for AMD...

From leaked benchies.....not really. Now theres a chance that translated chinese benchmarks can be wrong, obviously......but from what ive seen, they will still be behind Yorkfield, and not even closet to i7. Unless that is what AMD wants everyone to think.

Lolololo... Hold on. As I understand u think translating a website to make it more readable and understandable will change the numbers shown in benchmarks run's , then why translate u can easily make sense of these bar charts even without translating it from Chinese to english (I mean at least u can give Google web translator that much respect for doing its job right).:shocked:

I never said that Deneb will the next top performing part from AMD to beat intel to defeat, no that's not happening in near future especially with Deneb.

But i think Deneb should be able to match or better C2D in performance and as for as core i7 goes , well see what Deneb has to offer when it runs on DDR3 a.k.a AM3 socket , that will be a fair comparison i.e apples to apples b/w Deneb and Core i7.:beer:

Some more bs from Fudzilla for Laughs :laugh:

Won't compete with Nehalem

We learned that AMD won't go after Nehalem with its 45nm promised land called Deneb. Deneb in DDR2 variation and 45nm process with decent clock won't even try to compete with the Core i7 Nehalem generation. Our sources are confident that Deneb at 3.0GHz and even 2.8GHz should be enough to compete with the Core i7 920 (Bloomfield) that is clocked at 2.66GHz. Clock to clock AMD doesn?t stand any chance as according to all of the news we've gathered so far, Deneb can compete with Yorkfield on the same clock, i.e. the Core 2 Quad generation of CPUs, but not with Core i7. The good news is that this might indicate that all Deneb cores will cost close to $300-400, if not less including the fastest clocked one. The Core i7 940 at 2.93GHz will sell for $562 and as you can imagine AMD?s 3.0GHz Deneb has to be priced lower than that as it will end up slower. Deneb is still in dreamland and with a lot of luck, we might actually see some CPUs in November or December. The good news for AMD is that most of the CPU market is in the sub $300 range and this is where AMD has a chance to make some money with its new 45nm parts.

 

[Idontcare]

Originally posted by: BLaber
But i think Deneb should be able to match or better C2D in performance and as for as core i7 goes , well see what Deneb has to offer when it runs on DDR3 a.k.a AM3 socket , that will be a fair comparison i.e apples to apples b/w Deneb and Core i7.:beer:

Are you talking Kentsfield or Yorkfield when you make reference to Deneb beating C2D?

Given the IPC gap between Phenom and Kentsfield (>5% IIRC) and the IPC gap between Yorkfield and Kentsfield (5-10% IIRC) we really need to be realistic about how much IPC improvement we can justifiably expect from a minor core revision as Agena to Deneb is purported to be.

I have no issue believing that the publicly speculated architecture improvements to Agena for Deneb (increase L3$, etc) will increase IPC by 5% or so, thus closing the IPC gap to Kentsfield...but can we really expect the IPC improvement in Deneb to be 15% (versus 5%) as needed to close the IPC gap to yorkfield?

It seems a big stretch to me. Hope is one thing, but listing out the architecture improvements to justify the hope is an entirely different thing.

I can hope Deneb will have 15% IPC improvements such that it is IPC comparable to Yorkfield (then it's up to the fabs to pump out the chips with the right GHz to be absolute performance competitive with yorkfield) but I haven't seen a list of architecture improvements that could possibly generate a 15% IPC improvement. (not that it isn't impossible, I have no insider info or access, but publicly we have no data to substantiate our hope)

But Nehalem...damn. Bulldozer won't be out until Westmere is out (at best) or Sandy Bridge (worse). What a bleak horizon.

 

[BLaber]

I totally agree with you as my guess right know is as good as yours on how much of IPC improvement Deneb will bring.

Well lets wait it out my friend and see what we get :clock:

 

[Idontcare]

Dangit!

You were supposed to come back and disprove my assertions with a nice compiled list of Deneb's architecture improvements over Agena and prove that at least a 15% IPC improvement was plausible and on the table.

You weren't supposed to confirm my perception that it is not plausible for Deneb to beat Yorkfield clock-for-clock

I don't want my perception to be plausible; I want to be dis-proven and have Deneb show up with a blow-out 30-50% IPC improvement

 

[Martimus]

Originally posted by: Idontcare
Dangit!

You were supposed to come back and disprove my assertions with a nice compiled list of Deneb's architecture improvements over Agena and prove that at least a 15% IPC improvement was plausible and on the table.

You weren't supposed to confirm my perception that it is not plausible for Deneb to beat Yorkfield clock-for-clock

I don't want my perception to be plausible; I want to be dis-proven and have Deneb show up with a blow-out 30-50% IPC improvement

I would expect Deneb to have about the same jump in performance over Agena as Yorkfield did over Kentsfield. They are both a die shrink with minor improvements, and a large increas in cache. Deneb will probably also increase the IMC frequency as well, to help performance, but I don't know by how much if at all. At least the rumors are putting the initial clockspeed estimates at 2.8 and 3.0 GHz, so it should actually be a worthwhile product to purchase. I still doubt it will have any significant market penetration though, because it is too late to the party, and the performance will probably be just adequate - which isn't enough to make people consider switching platforms.

 

[nyker96]

I hope they do more than just a 45nm shrink, they need more than that to compete with Intel 45nm.

 

[Nemesis 1]

Rather confusing thread . The most I got from it was. Including the linked thread to the betting thread is that . A new socket is required meaning a new M/B is required. Which is a costly upgrade.

 

[Viditor]

Originally posted by: Nemesis 1
Rather confusing thread . The most I got from it was. Including the linked thread to the betting thread is that . A new socket is required meaning a new M/B is required. Which is a costly upgrade.

Ummm...same F1207 socket and same mobo (no change required). The next change will be the new chipset offerings next (Q2) year.

 

[SlowSpyder]

Originally posted by: Ocguy31

Originally posted by: BLaber
I think Shanghai and Deneb could be what doctor ordered for AMD...

From leaked benchies.....not really. Now theres a chance that translated chinese benchmarks can be wrong, obviously......but from what ive seen, they will still be behind Yorkfield, and not even closet to i7. Unless that is what AMD wants everyone to think.

It's certainly a step in the right direction. AMD has shown that they didn't need to have the best overall performance with their current graphics card line up to still sell a lot of product. If they can have 'good enough' performance at a decent price (without losing too much profit margin) they can probably do pretty well. Not to mention the smaller manufacturing process to help with that. I'm sure AMD will enjoy selling $300+ Denab processors that cost them less to manufacture then their current line up of $180 Phenoms that cost more to manufacture.

 

[Nemesis 1]

Originally posted by: Viditor

Originally posted by: Nemesis 1
Rather confusing thread . The most I got from it was. Including the linked thread to the betting thread is that . A new socket is required meaning a new M/B is required. Which is a costly upgrade.

Ummm...same F1207 socket and same mobo (no change required). The next change will be the new chipset offerings next (Q2) year.

Sorry viditor but I am still struggling with this. I was tring to find link to something I read about a month ago that said that when AMD goes to DDR3 that a new socket would be required. Can't find that dam link . I thought it book marked .

 

[Nemesis 1]

Originally posted by: SlowSpyder

Originally posted by: Ocguy31

Originally posted by: BLaber
I think Shanghai and Deneb could be what doctor ordered for AMD...

From leaked benchies.....not really. Now theres a chance that translated chinese benchmarks can be wrong, obviously......but from what ive seen, they will still be behind Yorkfield, and not even closet to i7. Unless that is what AMD wants everyone to think.

It's certainly a step in the right direction. AMD has shown that they didn't need to have the best overall performance with their current graphics card line up to still sell a lot of product. If they can have 'good enough' performance at a decent price (without losing too much profit margin) they can probably do pretty well. Not to mention the smaller manufacturing process to help with that. I'm sure AMD will enjoy selling $300+ Denab processors that cost them less to manufacture then their current line up of $180 Phenoms that cost more to manufacture.

This thread has me so confused as to whats being stated. Deneb is a dual core processor is this not true. If it is infact true . I don't see how AMD can ask $300 for it. As I don't believe for one second that it will beat penryn dualies at the same clock. No intel dual cores sell for $300 except maybe 8600. Surely you don't believe AMD has an= to 8600.

I believe AMD could try to sell 2 core cpu for $300 but will fail miserably trying to do so . Also intel simply has to lower its pricies on the top model. If deneb is released shortly its not a long wait to see. But $ 300 for a dually seems to be a stretch . I believe $200 might be OK . in the low / middle area . But Nehalem is coming . and the last stepping is very good. This allows intel room to lower prices on 2core penryns.

I hope your right but I just not bying it. Ok I see deneb is a 4 core processor. Still I can't see it beating Intels lowend Q penryn.

 

[jones377]

Originally posted by: Nemesis 1

Originally posted by: SlowSpyder

Originally posted by: Ocguy31

Originally posted by: BLaber
I think Shanghai and Deneb could be what doctor ordered for AMD...

From leaked benchies.....not really. Now theres a chance that translated chinese benchmarks can be wrong, obviously......but from what ive seen, they will still be behind Yorkfield, and not even closet to i7. Unless that is what AMD wants everyone to think.

It's certainly a step in the right direction. AMD has shown that they didn't need to have the best overall performance with their current graphics card line up to still sell a lot of product. If they can have 'good enough' performance at a decent price (without losing too much profit margin) they can probably do pretty well. Not to mention the smaller manufacturing process to help with that. I'm sure AMD will enjoy selling $300+ Denab processors that cost them less to manufacture then their current line up of $180 Phenoms that cost more to manufacture.

This thread has me so confused as to whats being stated. Deneb is a dual core processor is this not true. If it is infact true . I don't see how AMD can ask $300 for it. As I don't believe for one second that it will beat penryn dualies at the same clock. No intel dual cores sell for $300 except maybe 8600. Surely you don't believe AMD has an= to 8600.

I believe AMD could try to sell 2 core cpu for $300 but will fail miserably trying to do so . Also intel simply has to lower its pricies on the top model. If deneb is released shortly its not a long wait to see. But $ 300 for a dually seems to be a stretch . I believe $200 might be OK . in the low / middle area . But Nehalem is coming . and the last stepping is very good. This allows intel room to lower prices on 2core penryns.

I hope your right but I just not bying it. Ok I see deneb is a 4 core processor. Still I can't see it beating Intels lowend Q penryn.

Deneb is their 45nm quadcore for desktop

 

[Martimus]

Originally posted by: Nemesis 1
This thread has me so confused as to whats being stated. Deneb is a dual core processor is this not true. If it is infact true . I don't see how AMD can ask $300 for it. As I don't believe for one second that it will beat penryn dualies at the same clock. No intel dual cores sell for $300 except maybe 8600. Surely you don't believe AMD has an= to 8600.

Kuma is the Dual Core processor. Deneb is the 45nm shrink of Agena. It is the desktop version of Shanghai. The 45nm quad core with 6MB of L3 Cache. It is the equivalent of the Kentsfield to Yorkfield change in processors on the AMD side.