PCPER.com: 5 GHZ+ Phenom II Overclock on Dry Ice, 6Ghz on LN2

[justapost]

CB10 crashed on all machine it's not CB10 not stable.

You think it's a software issue? They do not mention it failed non overclocked. So why did they not verify stability with something else more reliable?
On the other hand I have not seen i920 prime95 proven oc's and it does not matter much for benching fun.

[Ghostbuster]

CB10 crashed on all machine it's not CB10 not stable.

It could be a CPU eratta of some sort (that happens at a certain frequency), seems that some code sequence in CB10 could have activated it..

[Calmatory]

Another fake? However this one doesn't look like it's been shopped.

http://www.vr-zone.com/articles/phen...shot/6228.html

What is with the 4 GHz on air sceptism by everyone?

[Hornet331]

You can see what AMD lists in IPC improvement list.Seems DDR3 will bring approx. ~5% more perf. when compared to AM2+ Denebs.So overall,going to AM3 Deneb from AM2+Agena one gets ~18% advantage per clock.AM2+ Denebs should be ~13% faster than Agena if these numbers are true(but this is kinda overall gain;somewhere will be higher(pov ray ie. ~21%),somewhere lower (Nero recode~5%) ).Shanghai seems to confirm these numbers so it may very well be true for desktop too.

I doubt the 5% IPC increasement due to faster ram, the only programs that would probably benefit form more bandwidth would be archivers. The other IPC gaines seem quite reasonable.

[Sparky]

That is impressive. Although i would have to invest in another form of heater though

i hate that type of socket though....so used to LGA
wish AMD would change FFS[/IMG]

Why? I don't mind the pins on the CPU, actually I rather have them on the CPU than in the socket, a bent pin on the CPU isn't a big deal like it can be in the socket.

But if you like LGA - which I have dubbed "lump grid array" - just get a dual opteron board, socket F is LGA and you'd have twice the CPUs to play with

[ownage]

I don't think Phenom2 will be faster then Phenom, or at least not much. Remember the AMD chart? Phenom2 was about 16% faster then the old Phenom while the Phenom2 was clocked 15% higher.

[Calmatory]

Read the article again.Crysis is,AFAIK,not stressing all 4 cores,hence it's not a full load.But yeah,you can rest assured these thing will run relatively cool(compared to other CPUs),even after heavy OCing and overvolting.Another big plus since nobody wants a melting socket/board or excess heat heating up other components.

Relatively cool with 125 W TDP? E.g. Pentium D 820 was 95 W and I recall it was "too hot!".

I don't think Phenom2 will be faster then Phenom, or at least not much. Remember the AMD chart? Phenom2 was about 16% faster then the old Phenom while the Phenom2 was clocked 15% higher.

There are tests of Shanghai compared to Barcelona. They show improvement. Besides, what is the point in triplefolding L3 if it brings no improvements? What about the other claimed improvements, E.g. improcved branch predictor, prefetcher, TLB...?

There is improvement, somewehere near 8-12 % average, depending a lot on programs used for measurement. Besides, which chart are you talking about? URL would be great.

[suddeath]

What is with the 4 GHz on air sceptism by everyone?

I'm not sceptic about Deneb doing 4GHz on air, IMO it's a real deal (though i'm not sure if it can be 24/7 stable with 1,55-1,6 core), but with all this fakes around how can we be sure which CPU-Z screenshot is authentic?

[ownage]

There are tests of Shanghai compared to Barcelona. They show improvement. Besides, what is the point in triplefolding L3 if it brings no improvements? What about the other claimed improvements, E.g. improcved branch predictor, prefetcher, TLB...?

There is improvement, somewehere near 8-12 % average, depending a lot on programs used for measurement. Besides, which chart are you talking about? URL would be great.

Show me

[Motiv]

Show me

Hardly a wide range but it's something...
http://www.anandtech.com/showdoc.aspx?i=3456&p=7

[Ghostbuster]

What is with the 4 GHz on air sceptism by everyone?

I could be a suicide shot on air.. Maybe with no applications loading the cores.

The voltage are almost inline with the other 4GHz CPUZ screenshot (from Coolaler forum), though a bit lower.

There are tests of Shanghai compared to Barcelona. They show improvement. Besides, what is the point in triplefolding L3 if it brings no improvements? What about the other claimed improvements, E.g. improcved branch predictor, prefetcher, TLB...?

There is improvement, somewehere near 8-12 % average, depending a lot on programs used for measurement. Besides, which chart are you talking about? URL would be great.

It could be this one http://en.hardspell.com/doc/showcont...58&pageid=3149 (its source is from Coolaler forum)

Another (old) source http://www.itocp.com/thread-12164-1-3.html

Hardly a wide range but it's something...
http://www.anandtech.com/showdoc.aspx?i=3456&p=7

Those are server side applications. They create lots of threads depending on the number of clients connected to it, thus the processor have to work on those threads (in memory) a lot. Opteron's high memory bandwdith and high speed interconnect benefits here.

[Macadamia]

Relatively cool with 125 W TDP? E.g. Pentium D 820 was 95 W and I recall it was "too hot!".

I don't think Intel was honest before Core 2 Duo.


And Phen II might be hot on stock/crappy HSF combos, but once you give it some muscle, it would be cool. Same applies to the 805.

[ownage]

Hardly a wide range but it's something...
http://www.anandtech.com/showdoc.aspx?i=3456&p=7

You can't compare server benchmarks with desktop/gaming benchmarks.
Deneb doesn't even looks better then Phenom1 on the benchmarks provided by AMD so the only thing I expect is a Phenom1 on higher speeds.

[Shintai]

Hardly a wide range but it's something...
http://www.anandtech.com/showdoc.aspx?i=3456&p=7

But thats mainly cache heavy and cache coherency numbers. (Abit like 65nm vs 45nm Xeons).

For the desktop is another ballgame. Even 5% faster than phenom clock for clock would be impressive. I think people forget the K8->K10 improvement. And 45nm K10 is only a very small refinement of it. Think yorkfield vs kentsfield. More cache, small tweaks. Here is your 5%.

[BrowncoatGR]

Actually, the 7% shrink is the gate length not gate thickness ... the gate thickness (oxide) was not mentioned.

My bad, didn't pay too much attention to that article. What effect does gate length have on transistor performance? I still think it's unlikely they reduced gate thickness much, considering they actually increased it for K10 on the 65nm process.

[Macadamia]

You can't compare server benchmarks with desktop/gaming benchmarks.
Deneb doesn't even looks better then Phenom1 on the benchmarks provided by AMD so the only thing I expect is a Phenom1 on higher speeds.

Yeah, so that 20% boost on a C0 Stepping Deneb on POV-RAY (HardSpell) is completely useless.

Funny when it's Cinebench, then people take it as "desktop". Ugh.

[ownage]

Yeah, so that 20% boost on a C0 Stepping Deneb on POV-RAY (HardSpell) is completely useless.

Funny when it's Cinebench, then people take it as "desktop". Ugh.

I never look at cinebench.

[Calmatory]

Show me

"One of the first enhancements, and perhaps the biggest change, is the increase of the L3 cache from 2MB to 6MB. On average, it's expected that this will increase performance by anywhere from 5-10%." + the claimed branch predictor, TLB and prefetcher improvements. Besides, as I said, it is hugely dependent on the programs used, there is no way of saying "average" really. For some people the average improvement might be 4-7, and for another 10-13% due to very different kinds of programs used. And after I think about it more troughly, I agree, it is a bit optimistic, yet possible.

This might be fake though, but it is all I found. As it suggests, the IPC improvement varies a lot between apps.

[Shintai]

"One of the first enhancements, and perhaps the biggest change, is the increase of the L3 cache from 2MB to 6MB. On average, it's expected that this will increase performance by anywhere from 5-10%." + the claimed branch predictor, TLB and prefetcher improvements. Besides, as I said, it is hugely dependent on the programs used, there is no way of saying "average" really. For some people the average improvement might be 4-7, and for another 10-13% due to very different kinds of programs used. And after I think about it more troughly, I agree, it is a bit optimistic, yet possible.

This might be fake though, but it is all I found. As it suggests, the IPC improvement varies a lot between apps.

Not fake. Acurate is another matter tho.

Plus its yet another thing they steal from a chinese site and post as their own. That site should be banned here.

[Donnie27]

"One of the first enhancements, and perhaps the biggest change, is the increase of the L3 cache from 2MB to 6MB. On average, it's expected that this will increase performance by anywhere from 5-10%." + the claimed branch predictor, TLB and prefetcher improvements. Besides, as I said, it is hugely dependent on the programs used, there is no way of saying "average" really. For some people the average improvement might be 4-7, and for another 10-13% due to very different kinds of programs used. And after I think about it more troughly, I agree, it is a bit optimistic, yet possible.

This might be fake though, but it is all I found. As it suggests, the IPC improvement varies a lot between apps.

QFT! Some folks will always pick what fits their opinion or view though. Hell, some will even change what they pick/think is important. Look at the whole SuperPi flap!? There will always be apps that will run better on a certain architecture/s. Servers and Desktops comparatives should be looked at only as a reference, nothing more.

[justapost]

Plus its yet another thing they steal from a chinese site and post as their own. That site should be banned here.

Hmm xtreview you mean?

Associate published a review of the 45- nm processor phenom X4 on deneb core . For the tests Associate used the motherboard MSI K9A2 platinum, good for transfer from 65- nm to the 45- nm core and also 2 GB DDR2-800 memory form corsair productions and the graphic accelerator radeon HD 4850.

Nothing wrong here.

[Shintai]

Hmm xtreview you mean?

Nothing wrong here.

They dont seem to have anything together with the random chinese sites they take it from. And all the translations is pure google translations with all the errors. They always claim to be near, if not having it themselves.

http://en.hardspell.com/doc/showcont.asp?news_id=3858 (With more numbers)

Some from hardspell. Others from pconline.cn etc.

But hey..its not like the circle doesnt go around:

http://en.hardspell.com/doc/showcont.asp?news_id=4511
http://en.hardspell.com/doc/showcont.asp?news_id=4509



Just so much junk "journalism" around to get ads revenue.

http://reviewage.co.uk/index.php/Lat...ld-Beater.html

There's said to be two chips in the pipeline, the FX-80 at 4Ghz and the FX-82 at 4.4GHz.
The FX-80 at stock out performs a 5.0GHz Kentsfield.

[Ghostbuster]

Yeah, so that 20% boost on a C0 Stepping Deneb on POV-RAY (HardSpell) is completely useless.

Funny when it's Cinebench, then people take it as "desktop". Ugh.

Why not?

Those uses one thread per core (like most desktop applications), doesn't rely on graphics hardware just pure processor power. Scales very nicely from 1 core to 4 cores on all processors, meaning cache coherency has little affect on it... And its based on real life applications (POV Ray and Cinema4D). It does show the processor's pure processing speed in 3D rendering which is FPU/SSE intensive.

[informal]

"against an Intel Core i7 965 beat us the available model excellent"

What?


-----
Babelfish version!



So it's competitive AGAINST the i7 at 4 vs 3.2. They had to be convinced by AMD employees that it was?

I don't think they meant 4 vs 3.2.Translation is awful.And those 3.2Ghz for core i7 are really 3.33 or often 3.46 .

@Calmatory
You'll just have to wait and see how "hot" PhenomII is ,be stock or OC.I tell you it will be a lot cooler chip than core i7(which often reaches high eighties under air cooling in full load).

[malice85]

They do not take clockspeeds into account for that comparison. They just state that it does outstandingly (hervorragend) well compared to the i7 965. In addition to that they say that they had the chance to approve what the engineers claimed.