AMD Phenom II Review Thread

[Mats]

"Thrustmaster-gripping livingroom pilots"

The fact that there are exceptions to what largon stated before isn't really interesting in this situation.
There are ALWAYS exceptions.

[jmke]

i went through a nice collection of reviews yesterday, put all the scores in one big pile and came up with this result chart, compared to the Phenom II X4 940



Full review list

[Mats]

Great work, jmke! Reviewing the reviews!

[AbelJemka]

Impressive work jmke!

[Shintai]

Nice jmke

[Hornet331]

Great work, jmke! Reviewing the reviews!

Indeed, nice work.

[Lightman]

What 12 core bulldozer are you talking about. There is a K10 based 12 core that gonna cost something around 2-4000$. Desktop "bulldozer only stands as 4cores with GPU and 4+ cores without.



The 6 core is server only. Plus there aint any desktop above 3Ghz on their roadmap for a good reason. The 6 cores die would be too big for any practical mainstream market. And yes, it is late 2009. Not middle.

I wouldn't say that.

Reason:
Barcelona die is 283mm2
Shanghai die is 258mm2
Barcelona single computing core (excluding L2 cache) is ~25mm2
Shanghai single computing core (excluding L2 cache) is ~15mm2
AMD 65nm SRAM takes ~12mm2/MB
Intel 45ns SRAM takes ~6mm2/MB

Assuming AMD 45nm SRAM is around as dense as Intels (comparing 65nm AMD SRAM is denser by 0.3mm2 than 65nm Intel SRAM, Intels cache is not so dense but performs better) then extra 2x512KB of L2 cache takes 12mm2 of die space.

It was already predicted by Hans De Vries than Istanbul will have die size of around 300mm2.

By simply adding 258mm2 of Shanghai + 2x15mm2 per extra cores + 12mm2 for 2x512KB L2 cache we are getting exactly 300mm2 (some soures were leaking die size of 292mm2, this would mean quite good die size optimization if true)

So reassuming, how big of a difference will make for AMD producing 292-300mm2 dies and selling them for $300.00 when they were producing 283mm2 and selling them for well under $200??

[Macadamia]

According to economics of scale (for AMD at least), the 6-core part definitely has to come to desktop.
AMD unlike Intel doesn't produce a mask like taking a dump everyday.

@Lightman L2 and L3 both are around the 7mm2/MB mark I think.

[Hornet331]

imho amd has to bring a hexa core to desktop if they what to stay competitive... just look how much "damage" a singelsocket bloomfield does in the dualsocket server market. In this market we talk about 4 cores + HT vs 8 real cores.

[RPGWiZaRD]

Great initiative jmke.

So according to that compilation PII is still ~10% slower clock for clock than Intel Yorkfield Q9xxx series (or roughly the same as Kentsfield). So probably ~4.0GHz PII is around the performance levels as Yorkfield @ ~3.6GHz unless they scale heavily different in overclocking ratio which I doubt. Pricing seems reasonable but it's only "as good option as Intel" and being so late to respond it's not gonna bring any very huge cash to AMDs account, hope AMD won't have to continue to catching up Intel and could put some decent competition still in future but it looks like it would need an a failure on Intel side and success on AMD side for it to happen. Not to mention i5 will arrive later this year and probably provide some decent performance advantage (which can be confirmed by i7 pretty much) and Intel will control the pricing marketing again (either sell CPUs very expensive/expensier than AMDs parts and squeesh as much $$$ as possible out of every CPU sold or sell a bit cheaper and put a pressure on AMDs lineup which forces AMD to do pricecuts depending what path they decide to go). As a consumer I of course would hope for the latter.

[leoy]

In overclocker gr's review @ 3.7 deneb yorkfield kentsfield the difference is less than 10% and if u take into account the nb clocks that can add some 2-3% performance beyond the 2-2.2 ghz(goes up to 2.7 most of them )the differece is even less.
Something that noone didn't notice is that kentsfiel and yorkfields performance are really similar(maybe 2-3 % difference overall).

[Torchon]

WOW, nice job jmke.

Can you please elaborate on how you got those scores? I want to use your chart as a reference point but to do so I need to know the science behind it.


Thanks.

[Rammsteiner]

imho amd has to bring a hexa core to desktop if they what to stay competitive... just look how much "damage" a singelsocket bloomfield does in the dualsocket server market. In this market we talk about 4 cores + HT vs 8 real cores.

I dont know what's with the i7 stuff all the time? What does it have to do with this is anyway?

Any idea how small the market for i7 is in desktops? Hell, even Phenom does have a bigger market than that. Said this multiple times before, PhII does basicly have the things in house to regain a whole lot of market share, especially for enthusiasts, but i7's place is only with the enthusiasts.

i5 could be a way bigger danger, however that's like 9 months away. AMD has time enough to release eventual new revisions to improve Deneb's performance even more. They better do though before they've to lower prices to humiliating amounts.

For the pure performance i7 is very good, but you pay for that. And unless you really have to live by efficient computing and/or calculating things ASAP, i7 has no real place since even a good clocked dual core would be more than enough to do its job.

AMD has to release a hexa(+) core for servers, that's their main priority and make the architecture more efficient. HT is just fading out inefficiency, it works, now offence, but it doesnt mean AMD is nowhere without HT.

[RPGWiZaRD]

In overclocker gr's review @ 3.7 deneb yorkfield kentsfield the difference is less than 10% and if u take into account the nb clocks that can add some 2-3% performance beyond the 2-2.2 ghz(goes up to 2.7 most of them )the differece is even less.
Something that noone didn't notice is that kentsfiel and yorkfields performance are really similar(maybe 2-3 % difference overall).

Between Kent and York its around 5~6% difference in average (out of 2~13% or so min/max in the comparisions I've seen and depending on app etc), not 2-3%. Also done my own comparision with E6750 vs E8400 and in my favorite application I used to betatest for (PCSX2 ps2 emulator) it was 7.0~7.5% faster at EXACTLY same settings used with both CPUs). Don't forget with Yorkfield you can also run them at a lower CPU multi => higher FSB => higher performance. On my own setup 500x8 (4.0GHz) that I use corresponds to around 4.10 ~ 4.15GHz with 9x multi and 456~461 FSB, that's also like 3% difference.

[Metroid]

Phenom II is good enough from a competitive perspective, only the price is a bit salted.

The only way that AMD could be even more competitive in the CPU market would be the same approach it used against Nvidia concerning the GPU market:

Answer: Die Shrink step ahead.

Is it possible against Nvidia? Yes

Is it possible against Intel? Certainly not

Fact, AMD depends of third party companies, Intel does not.

[Motiv]

Fact, AMD depends of third party companies, Intel does not.

In what way does AMD depend on 3rd party companies and Intel not?

[Hornet331]

I dont know what's with the i7 stuff all the time? What does it have to do with this is anyway?

i5 could be a way bigger danger, however that's like 9 months away. AMD has time enough to release eventual new revisions to improve Deneb's performance even more. They better do though before they've to lower prices to humiliating amounts.

You gave the answer to yourself why i7 is mentioned all the time -> i5 is the exact same cpu and gives the exact same performance on desktops apps, hell some preliminary test show its sometimes faster then a i7.

As it stands now AMD would need a 3.2-3,4ghz deneb to match a i7/i5 @ 2,66ghz.

AMD has to release a hexa(+) core for servers, that's their main priority and make the architecture more efficient. HT is just fading out inefficiency, it works, now offence, but it doesnt mean AMD is nowhere without HT.

HT is not inefficient, actually its a menthod to increase efficincy, so if amd wants to increase efficency of there architecture they dont come around SMT.

A CPU with SMT and the same core count will always be more efficent (aka does more work) then a CPu without SMT and same core count.

SMT comes with relative low silicon budget, compared to a real core.

[Jamesrt2004]

AMD uses tmsc and ibm for die shrinks n' stuff iirc

[Shintai]

In what way does AMD depend on 3rd party companies and Intel not?

AMD (now?) uses the new foundry company to make its CPUs.

[jmke]

WOW, nice job jmke.

Can you please elaborate on how you got those scores? I want to use your chart as a reference point but to do so I need to know the science behind it.


Thanks.

those CPUs compared by multiple sites, each benchmark separately ranked in % compared to PIIX4-940, and then average from all of those

[Slay0r]

I wish fanboys from both sides would just poke eachother in the eyes when this kind of threads come up.. I mean, the thread title says "review thread" not "schoolyard ing".

good sum up jmke, by the way.

[Donnie27]

For the pure performance i7 is very good, but you pay for that. And unless you really have to live by efficient computing and/or calculating things ASAP, i7 has no real place since even a good clocked dual core would be more than enough to do its job.

AMD has to release a hexa(+) core for servers, that's their main priority and make the architecture more efficient. HT is just fading out inefficiency, it works, now offence, but it doesnt mean AMD is nowhere without HT.

Explain Pay for that as in the i7 940 selling for less than the 3800+ in its prime. I'm glad Ph2 920 has a common sense price because those mentioned would have made it an even worse deal. Never mind that though as Q9550 is less as well. Tell that Pay for it BS to folks who bought not only Intel X6800 but many here who both $669 4400+ and $850 FX-55. The whole system cost might be legit but trying to single just i7 is nothing but Fanboi Banter.

One of the more interesting benchmark results we got was what happened to the Core i7 965 in DIEP Chess when HyperThreading was disabled, namely a 20some % drop in performance. Since it really doesn’t matter which setting is enabled when it comes to show the “absolute winner” maybe it is time for AMD to take another look at that technology and potentially complement it with separate L1 caches for the logical CPUs. Just food for thought…

Some folks think AMD could benefit from Hyperthreading. Dirk might have worked with it while at DEC

[duploxxx]

You gave the answer to yourself why i7 is mentioned all the time -> i5 is the exact same cpu and gives the exact same performance on desktops apps, hell some preliminary test show its sometimes faster then a i7.

As it stands now AMD would need a 3.2-3,4ghz deneb to match a i7/i5 @ 2,66ghz.

is there any decent chart of performance tests done for several apps regarding your i5/i7 statement accept for the super pi and wprime results?

[saveus222]

NDA lifts today, the first reviews pop in.

http://www.dinoxpc.com/Tests/articol...dex.asp?id=866


Especially interesting comparisons to Phenom 1.




Seems like scaling from 2.6Ghz to 3Ghz sometimes brings slightly more performance than the 15% you would expect.



With much better power consumption than Phenom 1, it's starting to make more and more sense to buy an AMD cpu again.
AMD back in the game.

would you be able to compile a list of links in ur first post should help people view review links instantly..

[Vozer]

those CPUs compared by multiple sites, each benchmark separately ranked in % compared to PIIX4-940, and then average from all of those

Great work! Thank you very much, jmke.