AMD Shanghai is officially out

[ryboto]

But ye, finally 65nm Core 2s got competition. And one with much greater power consumption than the Q6600

What was K10 then? K10, on average isn't that far behind Kentsfield. A user here made a compilation of benchmarks, K10 was ~5% on average behind Kentsfield clock-clock. Also, you're suggesting, based on ONE power measurement from first revision sample in a system we don't know anything about that the 45nm parts consume more than a Q6600? Or are you basing this off of TDPs for Shanghai? You're assuming TDP=Actual power draw?

[Shintai]

What was K10 then? K10, on average isn't that far behind Kentsfield. A user here made a compilation of benchmarks, K10 was ~5% on average behind Kentsfield clock-clock. Also, you're suggesting, based on ONE power measurement from first revision sample in a system we don't know anything about that the 45nm parts consume more than a Q6600? Or are you basing this off of TDPs for Shanghai? You're assuming TDP=Actual power draw?

5% is about whats 45nm K10 is faster than 65nm.

And you read the power consumption the opposite way. 45nm K10 got MUCH better power consumption. Just like yorkies. Q6600 is quite bad in that area.

[Katanai]

http://www.intel.com/intel/location/Denmark.htm

Ooh I see.

[Calmatory]

What I don't get is why the people that rag on shintia don't realise that they are just as biased and as negative to Intel. Pot, kettle, black.

It is not just "Shintel having an opinion". Shintels opinion is that AMD is bad and he tries to prove this over and over again in any thread which would imply AMD doing something good.

It looks like a religious belief. Look at his statements, he states that "look omigad Deneb sucks 1.7V for 4 GHz or so!" and somehow he does not notice that it is C0 revision which he knows that is not as good as C1 or C2. Basically he is not neutral, and does just want to show Deneb in bad light as said. This, then is so called fanboyism, and after it happens over and over again, it really gets on nerves. He is just running around "lol AMD sucks. " but keeping it "within the rules".

[Shintai]

It is not just "Shintel having an opinion". Shintels opinion is that AMD is bad and he tries to prove this over and over again in any thread which would imply AMD doing something good.

It looks like a religious belief. Look at his statements, he states that "look omigad Deneb sucks 1.7V for 4 GHz or so!" and somehow he does not notice that it is C0 revision which he knows that is not as good as C1 or C2. Basically he is not neutral, and does just want to show Deneb in bad light as said. This, then is so called fanboyism, and after it happens over and over again, it really gets on nerves. He is just running around "lol AMD sucks. " but keeping it "within the rules".

I didnt say it sucked. It just didnt grow into heaven like some want it to.

oh..and its a C1...Plus it was 1.5V for 3.6Ghz and 1.6V for 3.8Ghz.

Now you just make things up...

http://www.xtremesystems.org/forums/...4&postcount=54

[cegras]

I didnt say it sucked. It just didnt grow into heaven like some want it to.

oh..and its a C1...Plus it was 1.5V for 3.6Ghz and 1.6V for 3.8Ghz.

Now you just make things up...

http://www.xtremesystems.org/forums/...4&postcount=54

No one says it grows in heaven until you point it out and indirectly accuse people discussing the cpu of being hormone filled fanboys. God, get it through your head that you're flamebaiting.

Who's reality are you trying to correct? You're trying pretty hard though, I'll give you that.

[informal]

5% is about whats 45nm K10 is faster than 65nm.

And you read the power consumption the opposite way. 45nm K10 got MUCH better power consumption. Just like yorkies. Q6600 is quite bad in that area.

Yeah and yorks are 5-6% on average above Kentsfield..So what's your point?
BTW,shouldn't you wait for official retail Deneb reviews to draw a definite conclusion on IPC and power draw?
Also remember that Deneb will get a DDR3 treatment in Q1 that could give a few % boost over DDR2 scores now that we have fast low power DDR3 1333 modules on the market.

[ryboto]

5% is about whats 45nm K10 is faster than 65nm.

And you read the power consumption the opposite way. 45nm K10 got MUCH better power consumption. Just like yorkies. Q6600 is quite bad in that area.

5%? You know that for certain? Where did you see definitive proof of this? Please share?

This is what you said regarding power consumption,

And one with much greater power consumption than the Q6600

How was I supposed to interpret that as a positive remark?

[Vapor]

Cut the crap guys. Closed for 15 minutes.

Both sides of the argument need to tone it down and stop the trolling/flamebaiting. Staff has no hesitation to lock any of you out of the Xtreme News section and forget the key.

[Vapor]

Re-opened....and my eyes are on all the threads with this group of members participating...

[Katanai]

You've heard the man. Now play nice guys.

[justapost]

A german website pointed to an new power saving feature of shanghai called "smart fetch"

Source: http://www.golem.de/0811/63524.html

Google Translated

One of the tricks of this is a function called "smart Fetch" . Each of the four cores can independently almost completely power off. To maintain data integrity, the contents of the L1 and L2 caches are written in the L3 cache. So other cores can access the data without the sleeper to awake. This option can be disabled in the bios and if enabled only brings minimal losses in computing power, but a total of 21 percent power savings.

Found that quite interesting explains why the delta between idle and load is relative high(100W) compared to L5430 (60W), beside the obvious that the L5430 is a low power version.

[Voodoo²]

What about HT 3.0? I think Tyan has a board right now.

Edit: Here it is.

http://www.tyan.com.tw/card/2008AMD45nm/index.htm

Sorry, it says here that it will be available for Q1 2009

http://www.tyan.com.tw/newsroom_pres...l.aspx?id=1300

[Katanai]

What about HT 3.0? I think Tyan has a board right now.

Edit: Here it is.

http://www.tyan.com.tw/card/2008AMD45nm/index.htm

Sorry, it says here that it will be available for Q1 2009

http://www.tyan.com.tw/newsroom_pres...l.aspx?id=1300

That board looks fine. Those PCIE slots are really good and the 8 DIMM's per CPU are awesome. It would be a virtualization monster except for the LAN, that's the only weak link I see there.

[Stukov]

Yeah and yorks are 5-6% on average above Kentsfield..So what's your point?
BTW,shouldn't you wait for official retail Deneb reviews to draw a definite conclusion on IPC and power draw?
Also remember that Deneb will get a DDR3 treatment in Q1 that could give a few % boost over DDR2 scores now that we have fast low power DDR3 1333 modules on the market.

I personally don't think that DDR3 will be a big performance booster for Deneb even at the fastest levels; that and DDR3 will still be really expensive. If there is anything nice going to DDR3 it will be reducing power consumption for the ram which should be nice once DDR3 prices decline.

It looks as though their immersion lithography or whatever it is called is a really good process, I just wished AMD could have got the Deneb out sooner, I waited a year before finally just buying a C2D, it looks as though I would have had to wait a full 2 before getting the processor I was waiting on.

[justapost]

More benchmarks (source)

[Nedjo]

wtf! EVEREST benchmarks for server part?!

I'm still awaiting Virtualization benchmarks from Johan!

[justapost]

wtf! EVEREST benchmarks for server part?!

I'm still awaiting Virtualization benchmarks from Johan!

Definately more usefull for shanghai but I'm interested in IPC improvements.


Wrote down results, calculated scaling with cores. Beside photoworks (unsure about 7zip) all those benchmarks scale neraly perfect linear with cpu frequency (at least that's what i found on my 9950BE). Photoworks scaled 3,5% with 23% cpu increase on ym 9950BE so I use a factor of 14% here. Based on those assumtions i estuimated shanghai speed at 2.8GHz and compared the results with an 2.8GHz Xeon. On the last two row's I removed the best and worst results.
So based on ZDNet's test shanghai might be still abit behind xeon in ipc but now with much better power consumption. Too bad they have not posted K10 results.

[Rammsteiner]

Indeed. I dont get the reason of the epic comparisons between AMD and Intel right away instead of at least including Barcelona

Scaling is very good though! But without Barcelona results included there isnt a lot to discuss about IMO. Power consumption is pretty clear, but what about IPC improvements? Scaling improvements?

It's a shame nowadays reviews still ain't as good as they should be.

[informal]

Good work justapost
Seems Shanghai is a very good improvement judging by single core scores.
Now wee need some Deneb loving' with 1066 ram

[Rammsteiner]

Good work justapost
Seems Shanghai is a very good improvement judging by single core scores.
Now wee need some Deneb loving' with 1066 ram

Indeed

[Hornet331]

Indeed. I dont get the reason of the epic comparisons between AMD and Intel right away instead of at least including Barcelona

Scaling is very good though! But without Barcelona results included there isnt a lot to discuss about IMO. Power consumption is pretty clear, but what about IPC improvements? Scaling improvements?

It's a shame nowadays reviews still ain't as good as they should be.

seems like Barcelona to Shanghai got ~the same increase in IPC as Kentsfield to Yorkfield.

Cause if we look at phenom vs. kentsfield, Phenom was avg. ~5% slower then
Kentsfield.

Now it seems Shangai is avg. ~5% slower then Yorkfield.

[demonkevy666]

seems like Barcelona to Shanghai got ~the same increase in IPC as Kentsfield to Yorkfield.

Cause if we look at phenom vs. kentsfield, Phenom was avg. ~5% slower then
Kentsfield.

Now it seems Shangai is avg. ~5% slower then Yorkfield.

shanghai is running on ecc ram 800mhz and old HT 1,000mhz. phenom (agena) is faster then Barcelona by some.

[Rammsteiner]

seems like Barcelona to Shanghai got ~the same increase in IPC as Kentsfield to Yorkfield.

Cause if we look at phenom vs. kentsfield, Phenom was avg. ~5% slower then
Kentsfield.

Now it seems Shangai is avg. ~5% slower then Yorkfield.

You've 'a' point there, however a server CPU with only DDR800 support... It's not like a solid comparison. Also keep in mind that's tested with programs the majority doesnt even know about, let alone use.

Although Im not so sure if it will really matter in 'normal' programs really, but you never know. In the end P4 was a lot better in things as Video encoding/decoding than A64 was. We'll see though, I know I wont get Deneb that fast anymore, maybe later on. Right now Im waiting for my new system and going to RMA some things

[justapost]

Good work justapost
Seems Shanghai is a very good improvement judging by single core scores.
Now wee need some Deneb loving' with 1066 ram

Hmm from all those everest benchmarks only photoworks benefits from faster NB and mem speed. On my 9950BE I got 8% better results by increasing the NB speed from 2 to 2.4GHz with cpu at 3.2GHz, so expect round 4% going to 2.2GHz. Going from DDR2 800 to 1066 gave me 30% better results. So I expect ~35% better results for photoworks. CB10 also does not benefit from mem and nb speed. Unsure about povray and 7zip (winrar did not).