Yonah/Sossaman benchmarked

[Carfax]

I'm surprised this hasn't been posted in this forum. ComputerBase benched both a Sossaman and Yonah chip using the Cinebench 2003 benchmark:



Keep in mind that the score for the 2ghz Yonah is a projection based on the score attained by the 1.5ghz version; assuming it scaled linearly.

With that in mind, it looks disappointing. But, they were using the Lindenhurst chipset for a platform and that uses DDR2-400 only, so that would reduce the scores considerably.

Also, my German is non-existent but if you look at the CPU-Z screenshot, it shows the chip was running at 600FSB with a multiplier of 2.5 which is obviously jacked up.

I think the chip is really running at 150 FSB (150x4=600FSB due to quad pump) with a multiplier of 10.

The retail Yonah is supposed to run at 166 FSB, and use DDR2-667. With these enhancements, I'd expect Yonah to be equal clock for clock to a dual core X2 with 2MB of L2 cache in Cinebench.

Source

[alexio]

http://www.xtremesystems.org/forums/...ad.php?t=72737

[Carfax]

Ahh, it was in the News section. Thats why I never saw it, because I rarely ever go in there. Oh well, I guess this is old news

[freecableguy]

Not looking good.

[Carfax]

Not looking good.

Remember that the chip is castrated by running a slower FSB, an old chipset and slower DDR2-400 memory (with probably horrible timings)

I still think Yonah will be a really good chip, particularly for gaming.

Should make a cheap, but potent gaming platform.

[NiCKE^]

Let it have the same clock as the x2 4800+ and you will see some really nice

[PcCI2iminal]

the 780 just need a "real" fsb

[alexio]

the 780 just need a "real" fsb

Indeed, Dothan and Yonah are both laptop cpu's, the desktop counterparts with a higher bus will crush Todays X2's.

[PcCI2iminal]

yeah ,Dothan n Yonah r officially 100% laptop's chips and i can't wait to c Yonah doin things like that

[Carfax]

Indeed, Dothan and Yonah are both laptop cpu's, the desktop counterparts with a higher bus will crush Todays X2's.

Desktop counterparts? So it's official that Intel is going to put Yonah on the desktop? Last time I heard, they were thinking about it but they hadn't come to a decision.

If they do, it will be great. Now, if only they could throw EM64T on there aswell, and it would be spectacular

[Highland3r]

Desktop counterparts? So it's official that Intel is going to put Yonah on the desktop? Last time I heard, they were thinking about it but they hadn't come to a decision.

If they do, it will be great. Now, if only they could throw EM64T on there aswell, and it would be spectacular

Is the sossaman not the desktop version, with EMT64?

[Carfax]

Is the sossaman not the desktop version, with EMT64?

Sossaman looks to be just a server version of Yonah, with no other distinguishing features.

I thought that Sossaman would atleast include EM64T, and I was hoping that it would because I was going to upgrade to a Yonah platform.

It doesn't however

[Highland3r]

yes, sorry you're correct.... Got confused with another chip / next gen....

[Ugly n Grey]

the roadmap is on the THG article about the IDC...

[Carfax]

yes, sorry you're correct.... Got confused with another chip / next gen....

Actually, I just checked that IDF summary UG was talking about, and it did say that Sossaman will have EM64T.

However, the problem is that it looks like it will be using the 604 pin socket and the Lindenhurst chipset. The chipset is for servers, and thus would not perform well for desktop users not to mention the fact that it's expensive and uses registered DDR2-400 memory.

So, unless a company like Asus makes an adapter to enable the Sossaman to use LGA775 motherboards, then we won't have EM64T enabled Yonahs.

A damn shame really, because a 64-bit enabled dual core Yonah on a Asus P5WD2 would be a wicked gaming/multimedia platform.

I think that Asus would make an adapter for the regular notebook Yonah which doesn't have EM64T to use 955x motherboards, but I doubt they'll do it for Sossaman.

[kl0012]

Keep in mind that the score for the 2ghz Yonah is a projection based on the score attained by the 1.5ghz version; assuming it scaled linearly.

If you will look at a difference between Dothan 1.73GHz and Dothan 2.13GHz you will see that Cinebench is scaling perfectly linear (even a bit better than 1x)... but there is a little mistake in their calculation for Yonah 2GHz. In my opinion it is too early to do conclusions.

[Hans.Gruber]

Let it have the same clock as the x2 4800+ and you will see some really nice

Sossaman would need 2.8Ghz clockspeed to get even with Opteron 275, and single 275 is slower than X2 4800.



Also you have to remember that Yonah/Sossaman is 32bit only, and Opteron can run 64bit cinebench that improves performance up to 25% or more.
So running 64bit cinebench would give score of 1400-1500 to Opteron..

My estimate is that Sossaman would need 3.6Ghz clockspeed to get even (it wont scale perfectly).

[Carfax]

Sossaman would need 2.8Ghz clockspeed to get even with Opteron 275, and single 275 is slower than X2 4800.

The Sossaman can also be used in SMP configuration, so I don't know why you say it will need 2.8ghz. I agree though, that the Opteron 275 will definitely be in the lead due to better scaling and superior platform courtesy of hypertransport..

Opteron 275 is clocked at 2.2ghz, which is 200mhz more than the theoretical 2.0ghz Sossaman. On a clockspeed dependant benchmark like Cinebench, 200mhz counts for alot, not to mention the Opteron's better scaling.

Nothing from Intel is going to match the Opteron until Woodcrest I think. Woodcrest will have an onboard memory controller, and CSI (Intel's version of hypertransport)

Also you have to remember that Yonah/Sossaman is 32bit only, and Opteron can run 64bit cinebench that improves performance up to 25% or more.
So running 64bit cinebench would give score of 1400-1500 to Opteron..

Yonah will be 32-bit only, but Sossaman will have EM64T by all accounts.

My estimate is that Sossaman would need 3.6Ghz clockspeed to get even (it wont scale perfectly).

Man, where do you get these figures?

[Ugly n Grey]

pure speculation, who knows what the chips on either side will look like next summer....

[Duvie]

If you ppl read it and maybe did a bit of translation you will see after the cpu_z pic it acknowledges that the 1.30version is showing false readings....The german site admits it so we are really in the dark about some of the settings....


the point I found interesting was 2 things...

1) clock for clock yonah at 2.0ghz was slower then the 512kb cache 2.0gh 3800| X2....

2) the sossaman (2chiops) at 2.0ghz figured at linear scaling is almost 1/3rd behind the 2.2ghz opterons which admittedly are on a server board not know for faster memory timings as well....

I have seen stuff on cinebench and it does scale quite linear...

What p[pl need to remember is by the time we see the sossamon it isn't like AMD wont have migrated over to DD2 667 memory, 2nd HTT link, etc....maybe even more cause I bet these things wont be out until closer to the end of 2006...PPL always compare INtel chips from 6months to 12 months in the future to AMD now...like AMD uis going to move lateral with no movement...I mean cmon...

As it stands the Yonah chips have fixed some of their weaknesses but will not reverse the tide of performance crown...that and ofcourse they are laptop chips not marketed to desktop users....PPL always seem to forget that...means for some enthusaist it might be great but for most users they will never do this...

[Hans.Gruber]

Man, where do you get these figures?

3.6Ghz 32bit vs 2.2Ghz 64bit Opteron = equal performance in cinebench.

Very simple to calculate that.

1158 x 1.25 = 1447 for Opteron @64bit.

1447 / 663 = 2,18. 2.18 x 1,5Ghz = 3,27.

But since sossaman is using traditional FSB it scales miserably (like P4 and Xeon), so 3.6Ghz estimate is generous.
Also Yonah/Sossaman has 15 cycles L2 latency, so it wont be as fast per clock than dothan.

Like already said, this is pure speculation, but it's not that far from the truth.
Just take note that Intel is NOT talking about total performance, they are talking about perfromance per watt.
Intel is also comparing watt/perf ratio to worsest Intel CPU's in worsest situations.

[Carfax]

1) clock for clock yonah at 2.0ghz was slower then the 512kb cache 2.0gh 3800| X2....

Cache size doesn't seem to matter in this benchmark, as it is so dependant on pure clockspeed. The 3800+ X2 with 512kb of L2 cache would perform the same as a theoretical 4000+ X2 with 1MB of L2 cache at the same clockspeed.

Anyway, the 3800+ X2 was "barely" faster, and as you say, the Yonah was running a slower FSB and slower memory.

DDR2-400 is just plain yucky no matter what timings used, but I'm sure they used the worst timings regardless.

What p[pl need to remember is by the time we see the sossamon it isn't like AMD wont have migrated over to DD2 667 memory, 2nd HTT link, etc....maybe even more cause I bet these things wont be out until closer to the end of 2006...PPL always compare INtel chips from 6months to 12 months in the future to AMD now...like AMD uis going to move lateral with no movement...I mean cmon...

I agree. Those M2 socket dualies from AMD will rock, and Intel will not be able to compete with them until Conroe emerges I think.

Anyone who thinks Yonah/Sossaman will allow INtel to regain the performance crown is smoking some powerful .

[Duvie]

Opteron 275 is clocked at 2.2ghz, which is 200mhz more than the theoretical 2.0ghz Sossaman. On a clockspeed dependant benchmark like Cinebench, 200mhz counts for alot, not to mention the Opteron's better scaling.

Nothing from Intel is going to match the Opteron until Woodcrest I think. Woodcrest will have an onboard memory controller, and CSI (Intel's version of hypertransport)

First part....

1) 200mhz is 10% more then 2.0ghz so it will still be ahead by about 20-23%
2) opteron system likely using 400ddr ECC ram....AMD has started standardizing and validated that new chips are DDR500 capable....MOst mobo bioses have new bioses with those divider options as standad...no mod bioses but the real official bioses...

Second part...

And you dot think by the time woodcrest arrives that AMD will hav eadvanced anything??? Intel has a big deficit to make up already let alone kowing that AMD will likely increase its performance ability as well with tweaks and new platform chipset...

[Carfax]

3.6Ghz 32bit vs 2.2Ghz 64bit Opteron = equal performance in cinebench.

Very simple to calculate that.

1158 x 1.25 = 1447 for Opteron @64bit.

1447 / 663 = 2,18. 2.18 x 1,5Ghz = 3,27.

I doubt Sossaman's clockspeed will be 1.5ghz when it becomes available next year for sale. It will probably top out at 2.5ghz over time, but to start I'd expect atleast 2ghz models to be available from the get go.

Therefore, 884 (2ghz scores) x 1.25 = 1105 (Sossaman will have EM64T)

But since sossaman is using traditional FSB it scales miserably (like P4 and Xeon), so 3.6Ghz estimate is generous.
Also Yonah/Sossaman has 15 cycles L2 latency, so it wont be as fast per clock than dothan.

Where did you hear that the L2 cache latency had changed for Yonah? Thats the first I'm hearing this.

[Duvie]

Anyway, the 3800+ X2 was "barely" faster, and as you say, the Yonah was running a slower FSB and slower memory.

DDR2-400 is just plain yucky no matter what timings used, but I'm sure they used the worst timings regardless.

the bold part??? I didn't say that..I dont kow that...i just stated the cpu_z as admitted by the german site is erroreneous...

You are right about cache but I ran I have tested it and it did help the score by 1%...compared to games that cache can make up to 5-7% it is small but dont be fooled it is not all cpu...very much cpu dependent....It will increase with bandwidth has well as I tested with DDR500 (recognized and validated standard for A64 memory controllers now)...it can gain 2-3% there as well...


what you should do is look at how 1.5ghz versus 2.0ghz should be 33.333% faster if linear but they increased the score by 36.3%....BAD MATH...If linear with no increases in bus it should bee 33% or less...means IMO the 521 score should have been more like 511 which would put it 4.9% behind the A64 X2 at 2.0ghz....