AMD Shanghai is officially out

[Shintai]

Just a note...on desktop. a 9950 does 2930 in 1 CPU score at 2.6Ghz....

Was this a B2 TLB bugged one?

Q6700 does around 3100 and Q9450 3250 with DDR2.

All 1 CPU.

[Macadamia]

Just a note...on desktop. a 9950 does 2930 in 1 CPU score at 2.6Ghz....

Was this a B2 TLB bugged one?

Q6700 does around 3100 and Q9450 3250 with DDR2.

All 1 CPU.

Err, no.



9600 No TLB does score higher, and 9600 TLB might be low, but the rest of the marks don't match. A 2.6Ghz ES just gets 2742.


And that's supposed to be with a "faster" L3 I think (relative to server parts).

[Lightman]

Well I'm getting 3000 points on my Phenom clocked: 3000MHz CPU; 2277MHz L3 cache; 1102MHz DDR-2!!!!!!!
OS: Vista x64
Cinema4D R10 x64

[Shintai]

Err, no.



9600 No TLB does score higher, and 9600 TLB might be low, but the rest of the marks don't match. A 2.6Ghz ES just gets 2742.


And that's supposed to be with a "faster" L3 I think (relative to server parts).

http://www.bit-tech.net/hardware/200...-965-review/10

Just seems like the 2.3Ghz underperform with TLB patch as I said.

Your own score shows a 300 points up. So Im quite sure the 2.3Ghz in this case is burdened by the TLB patch.

So you contradict your own picture

[Macadamia]

It is 10. Only 10 has this kind of scoring.

[Hornet331]

It is 10. Only 10 has this kind of scoring.

so it makes it less valid?

You usually talk about the latest version when you refer to a benchmark, or does anyone still think, when talking about spec that someone refers to spec2000?

[Macadamia]

so it makes it less valid?

You usually talk about the latest version when you refer to a benchmark, of does anyone still think when talking about spec that someone refers to spec2000?

Maybe you should try decaf. I'm done talking with you.

[Hornet331]

Maybe you should try decaf. I'm done talking with you.

i dont drink coffee at all, it just taste horrible (same with beer).

[Chad Boga]

i dont drink coffee at all, it just taste horrible (same with beer).

Good Lord, you don't know what you are missing.

Official illy Coffee and Asahi Beer fanboy.

[Macadamia]

Here it is.


TR does same results in controlled environment. Good points for relative reference. This was before a TLB patch ever existed, so good try spinning.

[Shintai]

Here it is.


TR does same results in controlled environment. Good points for relative reference. This was before a TLB patch ever existed, so good try spinning.

http://www.bit-tech.net/hardware/200...-965-review/10

Read this again. I dont think the desktop version outperform the server version in rendering that much. Do you?

And you seem to mix scores alot from 9 and 10

[informal]

Macadamia is right.
The gains in Cenebench10 ,when going from K8 to K10 in single thread score were minimal,in the region of 3%!
The gains from K10->K10.5 are 23% clock for clock in 64bit version!Be it more L3 or a combination of IPC gains,the Shanghai simply blasts Barcelona core in this benchmark(it's a toy benchmark nonetheless).

The specint_base2006 is more telling(21% faster clock for clock:


@Shintel

Nice try.
Tech report used Cinebench10 64bit version.
http://techreport.com/articles.x/13176/2

We used the following versions of our test applications:

* SiSoft Sandra XI.SP4a 64-bit
* CPU-Z 1.40
* SPECjbb2005 with Sun Java 6 Update 2 Windows x64 edition
* Valve VRAD map build benchmark
* Cinebench R10 64-bit Edition
* POV-Ray for Windows 3.7 beta 22 64-bit
* CASE Lab Euler3d CFD benchmark multithreaded edition
* MyriMatch proteomics benchmark
* notfred's Folding benchmark CD 8/8/07 revision
* picCOLOR 4.0 build 598 64-bit
* The Panorama Factory 4.5 x64 Edition
* Windows Media Encoder 9 x64 Edition

[DoubleZero]

Thats due to cache.

How things change.. all the sudden spec_rate matters, and the all the sudden cache is a gimmick.

[informal]

Do i have to do math for you Hornet??

2.3Ghz Barc. scores 9.2 => 2.7Ghz Brac. assuming perfect scaling would score 9.2 x 1.17=10.76

2.7Ghz Shanghai(667Mhz DDR2) scores 13.1.Therefore:
13.1/10.76=1.21 or 21% clock for clock

It's simple math...

edit2: Ahh Hornet requested a delete of his post since it was a brain fart? Oh man lol
edit:
done the calculation for specint_ratebase this time,Shanghai now extends the lead over same clocked Barcelona by a few more percents,now it's a claimed 25% in rate test(base ,a singlethreaded one was 21% better clock for clock).So even here we see the claimed IPC increase stands...


Also ,there is 22% increase in singletreaded specfp_base2006 scores when compared to same clock Barcelona.Numbers are here.
In specfp_rate_base2006 ,a multhithread test,there is 26% clock for clock increase over Barcelona with when 667 DDR2 mem. is used and 33% per clock increase with DDR2-800!!

[Cybercat]

I really hope we get some solid information on Deneb's performance before January 8.

[Sly Fox]

Nice results! Great to see AMD do well server-side at least, that's where the money is.

Of course there will still be haters, but progress looks good here. I just hope Deneb drops Intel 45nm pricing.
Nothing like cheap Q9650's to make X48 scream!

[bedlamite]

These results are quite ok, but these... http://finance.google.com/finance?q=NYSE:AMD
I don't understand

[Shintai]

These results are quite ok, but these... http://finance.google.com/finance?q=NYSE:AMD
I don't understand

Its called recession plus a debt crisis and no cashflow (Loans, consuming etc).

[Helmore]

And probably a small reaction to Intel's lowered forecast. If Intel is warning everyone that they are not doing as great as they had hoped, then AMD must surely also be affected and that's probably another reason for the decrease.

[keithlm]

Deneb is january 8th (Atleast embargo) on the last AMD slide.

The slide you posted came out about April 9th. A lot of things may have changed in the 7 months since that slide was presented.

[fragmasterMax]

Can anyone find some reviews with oc results?
Those, i must find...........

[informal]

You are expecting to see someone overclock the server chip??
There won't be any OCing of Shanghai.Deneb if launched in late Nov or Dec time frame can show us the OCing potential.By the eetimes article i just read,Shanghai fixed leakage issue Barcelona had :

the transistor drive current for AMD's 45-nm devices is much lower than that of the Intel HKMG transistors. But power consumption is quickly becoming a high priority for server chips. AMD's transistors exhibit very low channel leakage. Our transistor benchmarks indicates that leakage current is less than one-third of the value measured on AMD's 65-nm process. It's also significantly lower than the Intel 45-nm HKMG process. In fact the Ion/Ioff ratio for AMD's PFET is nearly 10 times better than that for the Intel PFET.

Early C0 and C1 samples showed up to 4.2Ghz OCing(MSI engineer over at coolaler could run Super pi up to 4.1Ghz IIRC) which is a direct confirmation of the quote above.Also it appears Deneb ES run very cool and consume a lot less power than Agena(both at stock and OC).

[Shintai]

At low speeds yes. At 2.8 and 3Ghz its already 125W TDP again. The new limit just moved abit upwards. But it behaves just as with 65nm. Try a lower clocked 65nm Phenom/Barcelona and you see alot lower power consumption.

To compare you can buy 3.4 and 3.5Ghz 45nm Xeons.

1.5V Deneb C1 3.6G Stable air. Deneb C1 uses 1.368V for 3Ghz

And 1.6V for 3.8Ghz.

[Particle]

If I had a way to overclock my dual F system, I certainly would try it. I'm not sure my board supports 45nm yet though, and I don't know if the board has any OC options at all.

[Blaber]

Guys this might surprise you but according to Anandtech Deneb has 8MB OF L3 CACHE : http://www.anandtech.com/cpuchipsets...7&p=2#comments

WTF ; IF ITS TRUE THEN AMD fooled us all