Performance figures for K10 from theINQ

[LOE]

here it is

http://theinq.com/default.aspx?article=40749

For starters, SPECint_rate2006 and SPECfp_rate2006 speak the voice of doom for the 366 MHz faster clocked Xeon - and these are the reasons why Intel was downplaying performance of FP and praising the INT.
These figures were from a few months ago, so things may well have changed since then. But in the Integer test, a Barcelona 2.3GHz yields 21% higher score than Clovertown 2.66 GHz, but Floating Point test leaves a staggering 50% performance deficit for Clovertown, and this is not something 45 nanometre Penryn can solve overnight. Unless, of course the clock deficit for AMD is such that Intel speeds past.

[Piotrsama]

For starters, SPECint_rate2006 and SPECfp_rate2006 speak the voice of doom for the 366 MHz faster clocked Xeon - and these are the reasons why Intel was downplaying performance of FP and praising the INT.
These figures were from a few months ago, so things may well have changed since then. But in the Integer test, a Barcelona 2.3GHz yields 21% higher score than Clovertown 2.66 GHz, but Floating Point test leaves a staggering 50% performance deficit for Clovertown, and this is not something 45 nanometre Penryn can solve overnight. Unless, of course the clock deficit for AMD is such that Intel speeds past.

The inq..... :rofl:

[red]

What's the point of the Barcelona thread if people are going to make new ones :rolleyes:

But in the Integer test, a Barcelona 2.3GHz yields 21% higher score than Clovertown 2.66 GHz, but Floating Point test leaves a staggering 50% performance deficit for Clovertown

It's 2.6GHz, clear as day on the footnotes http://multicore.amd.com/GetFile.aspx?aliaspath=%2fImages%2fbenchmarks%2fBa rcelona_SPECintrate_040207v2

and this is not something 45 nanometre Penryn can solve overnight

http://www.spec.org/cpu2006/results/res2007q2/cpu2006-20070528-01175.html

106 vs 102, 106 wins. On 65nm.

The other numbers aren't impressive either and those are for 2.3GHz not the supposed 2GHz launch.

[BrowncoatGR]

Red I'm sure you notice those are estimated numbers... AMD has not released official numbers yet. 21% Int and 50% FP are not impressive? They look scary to me. That probably means a 2.0ghz can match the 2.66 Kentsfield in 2P int and be faster in FP! Its the Inquirer so take this with a grain of salt but they do occasionally post valid stuff(since they post anything they get their hands on) Since this is not from Charlie and the tone is quite somber for the Inq my guess is this could be real.

[Revv23]

i hope barcelona clocks well... i dont see it competeing if the high end product can only run 3ghz, even if it is faster clock for clock, my e6400 does 4ghz.

[accord99]

Red I'm sure you notice those are estimated numbers... AMD has not released official numbers yet. 21% Int and 50% FP are not impressive? They look scary to me. That probably means a 2.0ghz can match the 2.66 Kentsfield in 2P int and be faster in FP! Its the Inquirer so take this with a grain of salt but they do occasionally post valid stuff(since they post anything they get their hands on) Since this is not from Charlie and the tone is quite somber for the Inq my guess is this could be real.
But as been mentioned before, SPEC rate has always run well on the Opteron platform. There are more meaningful results, such as TPC-C, SAP-SD and SPECjbb2005 which suggest Barcleona 2.3GHz will have trouble with the 2.66 X5355.

[GripS]

yeah. Estimated. I'll believe it when i see 'actual' performance numbers.

Don't get me wrong. I'm ALL for AMD pulling out a win. I always used AMD up until about 1 month ago when i assembled this rig.

[freeloader]

yeah. Estimated. I'll believe it when i see 'actual' performance numbers.

Don't get me wrong. I'm ALL for AMD pulling out a win. I always used AMD up until about 1 month ago when i assembled this rig.

The figures aren't even "estimated". They're simulated.

[MAS]

extracted from http://vd.verysell.ru/files/ie/252_10_DOCUMENT_Server_Section_PUBLIC.ppt , leaked amd presentation under NDA

[Jacky]

These performance numbers were based on systems using tge Opteron 2356, or known to the world of us regular folk as AMD QuadCore at 2.3GHz. The orice of this processor is currently set at $796 USD in quantities of 1000, so we'll see what can Clovertown at 2.66 GHz (Xeon 5355) do about it…
n1 spelling, that's why I love theinq :D
better not comment on the content at all (but really can you believe anything they write if they can't even check WHAT they write?)

[LOE]

not using the latest firefox I guess ;)

bout the simulation... as far as I know amd has a software that can simulate chips running at 1/4 of their actual speed, so... I think the simulation is pretty close to what the CPU will score

The main problem for amd is getting the 2.3 Ghz parts for now, and then... we all know K8 is performing great in theese tests. K10 is pretty much an updated and upgraded K8, and should score nice as well.

But the real question is real world performance. So far there was so much FUD about K10 performing great, about K10 beeing a weak performer... a bag of fishy cinebench and pov ray benches, and some info released by members of this forum as well.

I really hope K10 turns out to be a good performer, I have the feeling K10 will repeat what R600 did... at first everyone was woooing at R600, but as all of you see as time passes drivers get better, R600 is better than the GTS - the card it shares the same price range with... and even beats the GTX in some tests. And I think it will only get better.

AMD has a problem with R600 drivers and K10 process, both are problems that will solve in time.

Some of the speculations recently point out that K10 is equal or even slower then K8 clock for clock. I think that is impossible, amd are not that dumb to spend a fortune on making a chip that isn't better than K8. The questions here are how much better clock for clock, core for core, and what clockspeed will it be able to reach.

[Shintai]

Some of the speculations recently point out that K10 is equal or even slower then K8 clock for clock.

Nobody said that asfar as I know. I think the slower part came from people comparing top K8s with K10. As in yes, a dual, dualcore K8 server at 3Ghz is faster than a single 2 or 2.3Ghz K10 quad. (If numbers are right). Even a MCM design K8 could be faster due to sheer higher clocks.

So i think thats where it came from. Not clock for clock.

[xlink]

that's great a faster CPU.

now when will every single other bottleneck be eliminated?

[lapdog]

Great! Some good "news" for AMD. Now if they can just sell some of those 2.6GHz monsters very soon. :shrug:

[Carfax]

SpecINT_rate and SpecFP_rate are practically worthless benchmarks when it comes to trying to figure out how fast Barcelona will be for desktop apps.

If they would have done SpecFP and SpecINT then maybe we would have something to talk about.

[[XC] riptide]

I think there was another thread that put this all to the grave.... The figures for barca don't add up.

[Jimmer411]

i hope barcelona clocks well... i dont see it competeing if the high end product can only run 3ghz, even if it is faster clock for clock, my e6400 does 4ghz.


Lol with just about all 3ghz+ P4 chips clocking upwards of 4.5ghz in todays boards, why arent you running one of those instead of your 6400? :ROTF:



On a serious note I hope this is true. Would be good to see intel push back (yay for the consumer on either end)

[lapdog]

Lol with just about all 3ghz+ P4 chips clocking upwards of 4.5ghz in todays boards, why arent you running one of those instead of your 6400? :ROTF:

The thing is it looks as if AMD will not have the clocks to beat Intel, even if these predictions are true. Can they even beat a similar priced part, or similar part on power consumption? No real answers here. :down:

[Yukon Trooper]

What's the point of the Barcelona thread if people are going to make new ones
Can say that about 90% of the threads lol. People too lazy to use the search function.

[Brother Esau]

Personally call me a fanboy whatever I still think AMD is superior to Intel:D

[kl0012]

Yes the 106 wins on 65nm...
with 8 cores...;)

So what is your point? The Barselona system on the picture have 8 cores as well.

[nemrod]

Yes the 106 wins on 65nm...
with 8 cores...;)

And according to you, what does mean dual-cpu quadcore in the footnotes of the figure? ;)

[ernestlin]

Yes the 106 wins on 65nm...
with 8 cores...;)

Who tell u!!! rumormonger!!!

[iddqd]

Personally call me a fanboy whatever I still think AMD is superior to Intel:D

A belief that corporation A is in any way superior (or in any way with different paradigms) to corporation B, C, D... through Z is the very definition of fanboyism.

[Brother Esau]

All right that works for me! Just personal prefference mostly due to where the memory controller resides but XYZ to you to and thanks for asking!:D

[iddqd]

What does a memory controller design have to do with AMD?

Actually, integrating the memory controller was the easy way for AMD to squeeze some extra performance out of their architecture without having the same strength of branch prediction as Intel. Which of course required much more work to implement. Of course there's a price to having an integrated memory controller; every time you make changes to the controller you have to change the socket layout - saves people the trouble of figuring out why the CPU won't work in a board it should work in; it won't fit in the first place.

Intel may still do it, if it proves that they need it. It's not a difficult feature to design (relatively speaking of course), Intel already has a lot of experience with memory controllers.

But it's only one feature of one product that AMD sells. Product features change with product lineups. But at the end of the day, AMD and Intel both are only trying to sell you pieces of silicon.

[saaya]

you dont have to change the socket every time you make changes to the mem controller :P

[iddqd]

you dont have to change the socket every time you make changes to the mem controller :P

Every time a major change is made.

[xlink]

Every time a major change is made.

you don't have ton change sockets per se, but you'd definitely need to change the board.

[alayashu]

Every time a major change is made.

K10 has an improved mem controller and it works in AM2 boards. If this isn't a 'major change', then i doubt there are any 'major changes' :P

[iddqd]

K10 has an improved mem controller and it works in AM2 boards. If this isn't a 'major change', then i doubt there are any 'major changes' :P

Then why is s939 gone? Expired. Followed 754 into the land of no return!

[Shintai]

The single biggest issue with IMCs is memory development. When Intel changes to IMC over their entire line too. Then we gonna hit a stagnation in memory development. Or atleast the famous chicken and the egg issue.

The only alternative (And crappy) route will be an endless pin, singal and so on compability forever. Unless they want to add some extra 300-400 pins to the CPUs for double memory support.

Currently AMD got the cheap way out of it, by letting Intel push memory standards.

But imagine a new CPU today that only supported DDR3.

[xlink]

The single biggest issue with IMCs is memory development. When Intel changes to IMC over their entire line too. Then we gonna hit a stagnation in memory development. Or atleast the famous chicken and the egg issue.

The only alternative (And crappy) route will be an endless pin, singal and so on compability forever. Unless they want to add some extra 300-400 pins to the CPUs for double memory support.

Currently AMD got the cheap way out of it, by letting Intel push memory standards.

But imagine a new CPU today that only supported DDR3.

you could just maintain signaling compatibility between standards. probably would take more than 300 pins total.

maybe make a signaling standard which all dRAM tech should follow for 5 years and improve every 5 years or so.

[Shintai]

you could just maintain signaling compatibility between standards. probably would take more than 300 pins total.

maybe make a signaling standard which all dRAM tech should follow for 5 years and improve every 5 years or so.

Problem is, as you also soemwhat describe is the "lock" in terms of improvement. You cant come out and say, now we want serial instead of a parallel design.

And even after those 5 years, you sit with the egg vs the chicken.

[BrowncoatGR]

The single biggest issue with IMCs is memory development. When Intel changes to IMC over their entire line too. Then we gonna hit a stagnation in memory development. Or atleast the famous chicken and the egg issue.

The only alternative (And crappy) route will be an endless pin, singal and so on compability forever. Unless they want to add some extra 300-400 pins to the CPUs for double memory support.

Currently AMD got the cheap way out of it, by letting Intel push memory standards.

But imagine a new CPU today that only supported DDR3.
You do realize that AMD intends to release CPUs capable of supporting both DDR2 and DDR3. What you are describing is a non-issue

[Shintai]

You do realize that AMD intends to release CPUs capable of supporting both DDR2 and DDR3. What you are describing is a non-issue

You do know how close DDR2 and DDR3 is? Try one that supports DDR1 and DDR2. Or SDRAM and DDR1. Or look abit forward to DDR4. Unlike the otehrs DDR2 and DDR3 is pincompatible.

[iddqd]

There are good reasons to keep the memory controller separate from the CPU, as many as good reasons to integrate it. Intel seems to be increasing cache sizes considerably lately, so it looks like their CPUs will be less and less dependent on (relatively) high latency dram. Of course increasing cache size has is also a double-edged blade.

Also, I'm willing to bet anything that AMD has another socket that's coming out to replace AM2 within 2008.

[BrowncoatGR]

Also, I'm willing to bet anything that AMD has another socket that's coming out to replace AM2 within 2008.

They'll release AM2+ in 2H07 and AM3 in 2H08. Both AM2+ and AM3 cpus are compatible with AM2 motherboards as long as the manufacturer updates the bios. AM2/AM2+ CPUs are not compatible with AM3 though.

[b0bd0le]

Welcome to 2007: "Simulated Benchmarks" (http://www.dailytech.com/Welcome+to+2007+Simulated+Benchmarks/article7927.htm)

In this day and age we don't even need silicon to benchmark our processors -- some of us anyway

Last April, benchmarks of AMD’s Barcelona running at 2.6 GHz made rounds across Internet tech sites. These benchmarks, according to AMD, provided a 21% performance gain over quad-core Xeon 5355 in SPECint_rate2006 and a 50% gain in performance over Xeon 5355 in SPECfp_rate2006.

This looks pretty phenomenal until you read the footnote courtesy of AMD: "Estimated performance @ 2.6GHz based on internal AMD simulations."

Much to my surprise, these benchmarks surfaced again over the last few days. This time, however, the author did not tell you the benchmarks were simulated and at least four months old. He did not tell you that the benchmarks ran were presented to him on a PowerPoint document, and he did not tell you the numbers featured in his "review" were of a 2.6 GHz simulated chip instead of a 2.3 GHz simulated chip.

In fact, AMD publically denied the Barcelona would top out at anything over 2.0 GHz at launch when it does launch on August 27.

So what's going on here? Shoddy journalism and even the author wouldn't deny that. Any author willing to pad his byline with such omissions of fact would almost certainly have an alternate agenda brewing.

I didn't give these benchmarks much thought when I first saw them in February, and given the confirmed top-out frequency from AMD, I certainly don't give these benchmarks traction now either. I suggest those with interests in Barcelona wait until someone tests actual DVT or Retail silicon.

The funny thing is I cannot recall the term "simulated benchmarks" in my pre-Barcelona lexicon. Since when did benchmarking a 1.6 GHz processor give the "simulated" performance of a 2.3 and 2.6 GHz chip? Does anyone find it odd that Googling for "simulated benchmarks" yields only 574 entries -- all from AMD, Microsoft and Sanda. AMD holds the number one spot.

Maybe simulated benchmarks will go down with some of the other great marketing terms of the last decade: FUD, paper launches and ship dates.

[Brother Esau]

I don't know bro, all I can say is that I preffer AMD over Intel its not being a fanboy its called personal prefference!

[RAMMAN]

same thing.

[kl0012]

Ziff man calls DAAMIT on Barcelona numbers (http://www.theinquirer.net/default.aspx?article=40792)
A SENIOR ANALYST for the Spliff Davis empire has called DAAMIT's benchmarking practices "blatantly deceptive" and has blasted the firm for having " no intention of behaving honestly or ethically."
George Ou is the Technical Director for ZDNet, which puts him at the top of the techie tree in the old Bill Ziff empire. He makes his claims after checking out AMD-presented benchmarks for the forthcoming Barcelona launch. In those benchmarks, AMD claims a 20% clock-for-clock advantage over Intel's Xeons.

However, the benchmarks clearly leave out faster-performing Intel parts and base their numbers on 2.6GHz Barcelona chips, a performance grade that won't even be released upon initial launch - the fastest Barcelona will launch at 2.0GHz in September, Ou says, meaning that "AMD is deliberately leaving out Intel’s best scores, leaving out Intel’s best products that shipped months ago, and putting in theoretical Barcelona scores for products that don’t even have a ship date."

[LOE]

he is right on most of the points, and still analysts are stupid

amd not behaving honestly or ethically... I wonder what that fool has to say about intel's business practices :down:

[b0bd0le]

amd not behaving honestly or ethically...

why would they have to lie?

amd's new architecture really must be in some trouble