Crysis CPU benchmark: QX6850 VS QX9650 VS PhenomX4

[Shintai]

Intel breaks the 130W barrier after 3GHz on Penryn at 1.216-1.248V, so what would 1.536V 3GHz quad get? Must be high but they're planning to retail 2.8GHz this year, so we'll see what they have to offer then.

No, 3Ghz Yorkfield Xeons are 80W or less.

[Epsilon84]

I'm eager to see AMD fans spin this one.

nn_step / informal, time to step up and be counted! LOL j/k

On a serious note, these results are disappointing from many angles.

OK, so Phenom performance is obviously sub par, but what about Penryn? And QC? Hardly no gains over C2D at all! Is it true that the Crysis SP demo is not multithreaded? It better be, otherwise all this talk of 'QC is needed for Crysis' is just total BS!

I guess the 'good' news is that Phenom is capable of overlclocking to 3GHz on the B2 stepping... but it doesn't do much good when C2D/C2Q is superior clock for clock and clocking to 4GHz+ (in the case of Penryn) or 3.5GHz+ with current 65nm Conroes/Kentsfields.

[Brother Esau]

scratches head
gotta google search myself i guess

Dino.....thats sounds Kinky what prey tell are you boys doing over there in OZ

[savantu]

That's bad but something I expected. Thats a B2 step core as well there and thats about the same voltage it takes an ES QX9650 to do 4.7GHz IIRC.

As voltage increases, power increases quickly and the electrical resistance will also increase along with it so you get less MHz the higher V you give (low resistance junctions and interconnects are worst affected) and high TDP.

Power consumption increases with the square of V.

Theres only two ways to take for CPU design when it comes to metal gate AFAIK and ones for high performance/high TDP and the other is for low TDP/v.low standby TDP preference. Intel chose the high-k dielectric metal gate which means you get maximum clock speeds (makes for easy PR work too) and not much TDP advantage over what the basic node offers

Intel's 45nm offers both : high frequency and low power

http://www.techreport.com/articles.x/13470/15

What should be noted is : process improvements offer lower power consumption at the same freq or higher freq at the same power.

Metal gates drastically reduce the biggest factor in leakage once you go under 65nm.

and AMD AFAIK chose low-k dielectric along with IBM, which is not made for high clocks at all but low power usage. SOI and other straining techniques combined can do better than having a metal gate BTW.

That's utter BS.SOI and metal gates address different factors of leakage , SOI is becoming less useful once you get to 65nm and lower.

The fact that SOI doesn't adress gate leakage , the biggest problem at 65/45/32nm means that AMD had to use thicker gates which equal slower transistors.

Then again I read AMD chose to go for high-k dielectric metal gate at 32nm, so I'm not sure what they're doing for 45nm or 32nm anymore. You can read more accurate info here: http://www.semiconductor.net/article/CA6402509.html

So basically , you say something and then deny it ?

Around 1.5V is about the max possible a CPU MFG wants to retail at, and usually never above 1.4V nowadays. I wonder what the TDP is. Intel breaks the 130W barrier after 3GHz on Penryn at 1.216-1.248V, so what would 1.536V 3GHz quad get? Must be high but they're planning to retail 2.8GHz this year, so we'll see what they have to offer then.

Huh ? Where do you get this info ?

1.5V is huge for a 65nm process ; 90nm used 1.35-1.4V.

[R101]

Can someone show load graph on the cores while timedemoing Crysis?

That way we could know for sure if it uses less than 4 cores.

[BeardyMan]

Can someone show load graph on the cores while timedemoing Crysis?

That way we could know for sure if it uses less than 4 cores.

core1:[------------------------------------------------------------90% ]
core2:[---5% ]
core3:[--3% ]
core4:[----5% ]


1 core only

[Shintai]

Be careful with the dual vs quad and so on in the time demo. Since there is no AI and such to be used. We need a real world run instead.

[Dainas]

Be careful with the dual vs quad and so on in the time demo. Since there is no AI and such to be used. We need a real world run instead.

Yep, specially considering the game ran smoothly DX10 16x12 maxed here until there was action. Thats where the margins will be a real kick if they are there to be found.

[Jacky]

OK, so Phenom performance is obviously sub par, but what about Penryn?

yep, no surprise here. while the penryn's improvements are kinda discouraging..

And QC? Hardly no gains over C2D at all! Is it true that the Crysis SP demo is not multithreaded? It better be, otherwise all this talk of 'QC is needed for Crysis' is just total BS!

I think it's the demo, otherwise we'd have to assume that the devs straight out lied to the public, saying that QC is important... I don't think so.

I guess the 'good' news is that Phenom is capable of overlclocking to 3GHz on the B2 stepping... but it doesn't do much good when C2D/C2Q is superior clock for clock and clocking to 4GHz+ (in the case of Penryn) or 3.5GHz+ with current 65nm Conroes/Kentsfields.

Neither good nor bad news. AMD already showed off their 3ghz parts. Now we have proof: at least they didn't lie. However 3ghz is just an overclock, like Intel showing off 3.6-4.0ghz penryn.

[Nedjo]

Be careful with the dual vs quad and so on in the time demo. Since there is no AI and such to be used. We need a real world run instead.

so basically this only useful for GPU testing?

[knightwolf654]

phenom x4 works period.
i myself am not trying to break records
thats all i need.

[Jacky]

phenom x4 works period.
i myself am not trying to break records
thats all i need.

great, it works! but isn't that what you'd expect from a multi-billion dollar company? or is there something special about amd

[Borgod]

Doesnt even tell us what resolution they were run at

Im guessing low res

[Gambit_2K]

Isn't Crysis highly optimized for Intel and nVidia? Taking in the fact that intel and nvidia had engineers helping out Crytek with the game. So is it a surprise for anyone that intel and nVidia are doing better in this game? And why are we even discussing results from a demo that is singlethreaded?

[AbelJemka]

Why everybody says AMD is owned with Crysis's benchmark showing less than 8% advantage for final Intel platform versus to beta AMD platform .
Moreover we don't know the price of AMD platform.
For me the only thing I think watching Crysis benchmarks is Penryn is same as conroe in term of performance in game.

[R101]

core1:[------------------------------------------------------------90% ]
core2:[---5% ]
core3:[--3% ]
core4:[----5% ]


1 core only

Waiting for the full game, then.

[Jacky]

Why everybody says AMD is owned with Crysis's benchmark showing less than 8% advantage for final Intel platform versus to beta AMD platform .
Moreover we don't know the price of AMD platform.
For me the only thing I think watching Crysis benchmarks is Penryn is same as conroe in term of performance in game.

Intel's expected to launch most of it's line up after AMD. Barcelona is supposed to be here, and we are *supposed to see* a nice range of Phenoms in Q4 BUT only the XE from Intel (they don't have the 45nm capacity for high volume atm).
Intel's platform is also in "beta stage", and Intel can do a lot through memory,fsb, chipset tweaking while AMD can't (IMC...). True, AMD's arch is more of a redisgn than Intel's shrink, but both companies are going to sell their platforms in the same timeframe - so AMD should be in the same stage as Intel.
CPUs aren't like GPUs they don't improve much after release.

[NapalmV5]

sorry, 333fsb vs 200 ? nice try by anand

higher fsb on amd still makes a difference

also, dont forget, demo @ 1 thread

[Hornet331]

sorry, 333fsb vs 200 ? nice try by anand

higher fsb on amd still makes a difference

also, dont forget, demo @ 1 thread

even more memory bandwidth... we sure need that.

well i guess some people like level to load faster.

[Epsilon84]

Why everybody says AMD is owned with Crysis's benchmark showing less than 8% advantage for final Intel platform versus to beta AMD platform .
Moreover we don't know the price of AMD platform.
For me the only thing I think watching Crysis benchmarks is Penryn is same as conroe in term of performance in game.

Well, how about the fact that a fully overclocked Phenom @ 3GHz is still 7.5% slower than a STOCK 3GHz C2D/C2Q that can be overclocked to 3.5GHz+ for 65nm and 4GHz+ for 45nm?

I'd say AMD has had its ass handed to it on a silver platter.

[NapalmV5]

even more memory bandwidth... we sure need that.

well i guess some people like level to load faster.

yeh?

well, try to keep up with my X4/areca1680/4x15Ksas or 4x ssd

lol

[Shadowmage]

Phenom B2 have max HT multiplier of 8... What he did to achive 3GHz was by setting the multiplier up to 15... HT Multiplier still at 8

so HT Link : 200MHz x 8 = 1600MHz
CPU speed : 200MHz x 15 = 3000MHz

NB speed should be arround 1.6GHz on that setup as well.....
Dual Channel with ganged mode .....

If what you say is correct, then that would explain the horrendous performance. 1.6GHz NB would slow down performance by a lot.

[Hornet331]

why is everyone so obsessed with the fsb... if it brings more than 1-2% performance increase i would be surprised. (notice how much core2 gained from 266 to 333...)

[PhilDoc]

Wow, people must really need AMD to fail to see that from this insignificant little test. Even if it stays the same with other benchmarks, AMD will still sell these chips and provided the ATI sector sells their Graphic cards AMD will start making a profit. They don't have to regain the performance crown to stay in business and thank god for all of us that, if this test is an indication, they will.

Does it matter that it took over 1.5V to reach 3G what is probably a 2.6G sample. To us it seems to, but its probably more significant that they hit 3G in the first place. They start out with 2.4 and 2.6G Phenoms. If this test is an indication, it looks like a 2.6G Phenom will outperform a 2.4G Yorkie and if they sell it lower, they'll sell a bunch of them. Sure, the Yorkie can be overclocked to probably over 4G with air, but beyond sites like these, no one cares.

If this trend continues, will I build my next computer around a Yorkie, certainly. But, I represent less then 1% of people who buy computers. Honesly, I'm encouraged by the results. It looks like AMD will stay in business and I'll get to buy cpus at a reasonable cost. Let hope they don't lose more important benchmarks by 20%. Then, they'd be in trouble and so would the rest of us. Unless of course you have lots of money or you get free cpus.

[D444niel]

Not alot in it between dual and quad core; would suggest that maybe the game is only taking advantage of 2 cores?