The Fastest gets faster - Intel Penryn Benched

[Pitch]

The newest revision of the Intel Core microarchitecture, Codenamed Penryn, has been benchmarked.

Penryn is released in two flavour, Wolfdale, a 45nM Dual Core CPU, and Yorkfield, a 45nM Quad Core CPU. These new parts will replace Conroe and Kentfield respectivly.

Apart from the die shrink to 45nM, Intel have increased the L2 Cache to 6MB per dual core.

They were benched with a 8800GTX and to me, the most interesting number is that Yorkfield is 37.3% faster than Kentsfield in the Lost Coast, Half Life 2 Techdemo.

The full article is here.

[bonesaw]

nice but why was the conroe faster in the last few test from the wolfdale?

[Samuknow]

That thing is crazy fast. For anyone running some major multitasking this will be the thing to have.

[keithwalton]

Originally Posted by bonesaw

nice but why was the conroe faster in the last few test from the wolfdale?

Which test are you meaning as I cant see it.

on the second page of the tests the Core 2 Extreme QX6800 (2.93GHz)
is a Quad core Kentsfield part which is why the wolfdale comes last as its 2x3.33GHz (6.66GHz) vs 4x 2.93GHz (11.72GHz) with the Yorkfield coming in at 13.33GHz combined processing power!

The half arsed SSE4 implementation would appear to give incredible boost in performance in multimedia like things which is really been the only weakness of the 'core' design when compared to the old netburst p4.

I just hope folding moves to make use of sse4 if it can. i doubt we'll see the same boost again as we did with conroe (128bit sse vs prev 64bit sse meant that most sse instructions could be done in half the clock cycles as previous)

for sure though this chip will own on smp folding esp with 2x 6megs of cache.

[Gizmo]

On page 3, Intel's benches show the Conroe outperforming the Wolfdale in some benches, and by a sizable margin.

My guess is that they got some numbers swapped in that chart.

Thing is, TBH, I'm not terribly impressed with the Penryn numbers. Ignore the hype for a second and take a look at the real clock-for-clock performance improvement. With the exception of the SSE4 extensions (which you really can't count, because those are new instructions), Penryn only gains 8-10% over Conroe.

They are clocking upwards of 3.3 GHz, which is good, but bear in mind that these are hand-picked engineering samples. Also bear in mind that the existing Conroes are hitting those speeds and better now in the hands of overclockers.

I've got ask a couple of questions:

1) Why isn't Intel releasing faster Conroes?
2) Why is the initial Penryn offering only 3.3 GHz, if the 45 nM process is all that?

For the time being, I'm going to reserve judgement on Penryn until we get some more detailed info.

One thing's for sure; even if Penryn doesn't hit but 3.6 GHz or so in production, AMD's got their work cut out for Barcelona. I really don't expect Barcelona to do much more than get AMD back in the race. Still, as long as it gets them back in the race, that is a good thing for competition.

[Got EpOx]

Competition is ofcourse one of the many reasons why companies regularly make better products, if there were no competition then companies could take aslong as they like producing a product. Go AMD!

[cadaveca]

Originally Posted by gizmo

I've got ask a couple of questions:

1) Why isn't Intel releasing faster Conroes?

Ebcasue they cannot guarantee thta the chipsets can consistenly keep up under standard cooling methods, and not kill other components wehn doing so. Chipset volts are still too high for the cpu, as something must "switch" these voltages to something the cpu can use, adn right now, this produces far too much heat. 333mhz is fairly easy...but over 356 causes issues with memory control as we ahve many differing voltages in chipset that must be dealt with accordingly. P35 is the savior...and why there will be both DDR2 adn DDR3 versions.

Originally Posted by gizmo

2) Why is the initial Penryn offering only 3.3 GHz, if the 45 nM process is all that?

BEcause it really isn't all that. Shipments of this cpu are not expected to top over .05% of all cpu's shipped up to an including 1st-Q '08.

Originally Posted by gizmo

For the time being, I'm going to reserve judgement on Penryn until we get some more detailed info.

One thing's for sure; even if Penryn doesn't hit but 3.6 GHz or so in production, AMD's got their work cut out for Barcelona. I really don't expect Barcelona to do much more than get AMD back in the race. Still, as long as it gets them back in the race, that is a good thing for competition.

Well, that's the last bit...Intel is playing safe at the moment. if they give AMD too muc hof an idea as to what they will release, they take the chance of being overstepped by AMD in market share, and will lose this standing in the market that they have been enjoying lately. Also keep in mind altohugh in the entusiast circle...Conroe Wins, but when it comes into the mainstream segment, Intel is doing OK, but not as good as in enthusiast market. AMD still has issues keeping up with demand, in part, so this really portrays the scenario in a light that noone seems to acknowledge, but me. Enthusiasts are but 15% of the market.

[Gizmo]

Originally Posted by cadaveca

AMD still has issues keeping up with demand, in part, so this really portrays the scenario in a light that noone seems to acknowledge, but me. Enthusiasts are but 15% of the market.

My Dad used to say "If you can't make enough for the market demand, it doesn't matter how good your product is".

That's kinda the spot AMD is in, now isn't it? They are giving away market-share because they can't keep up with demand. It's good to have demand exceed supply for a month. It's maybe even ok for 1 quarter, but it's bad for 2 quarters, which is what AMD has right now, and it's really bad for 3 quarters, which is what they are forecasting. Any way you slice it, that's a marketing disaster.

Granted, the enthusiasts don't directly spend but about %15 of the market (actually, I thought it was less than that). But don't forget that many enthusiasts are also in the position of advising the purchase of hardware, so their impact on the market is disproportionate to their direct purchasing power. Intel has always been an easier sell than AMD, and if Intel is offering superior performance at equal pricing, then who isn't going to buy Intel?

[cadaveca]

The enthusiasts that buy 65nm AMD parts that can clock almost as well as Intel parts, for much cheaper. Due to current pricing, AMD has a much better portfolio than Intel currently. We are seeing these details now from INtel as they are trying to retain customers, not just get new ones. 1.8ghz Brisbane core for $70, or intel 1.6ghz for $179...and pretty much comparable performance, nevermind that AMD boards are cheaper.

I agree with most of your points, but Intel cannot keep up with demand either. Most people are not buying cpu's at this point...not the enthusiasts that make the market, anyway(thier price share has increased as pricing in the market overall increased). They are awaiting decent quadcores. A single quadcore priced over 1k, and one just under is not enough to keep intel in the performance lime-light. This news makes it seem that way, but here's new stuff not to be released yet, against the still old 90nm offerings of AMD.

New Core2Duo's do not overclock like earlier samples...I'm seeing steadily declining clockspeeds in overclocks as Intel tightens the binning on these cpus. This brings AMD even closer to Intel's performance mark...we just await a new chipset to take advantage of what current AMD offerings have to provide.


We are starting to see AMD get back what it had before...It cannot happen overnight...and although financials for AMD look a bit bleak...and compnay that dumped 4.5B on an aquisition is gonna have some tough times as businesses are integrated, and they reign in the new company. AMD/ATI does not really deserve to shine until next year...when all products that they have are fully AMD-designed, and not just collaborations. Don't forget that those 300 jobs or so they cut back can mean $600K plus in the black too, in just one quarter! Where do those funds go?

[cadaveca]

more fuel to MY personal fire:

AMD: Blood In The Streets, Gross Margin At 31%

AMD (AMD) rallied like a house afire, closing the regular session at $14.28. up 2.7% on over 50 million shares, about double its normal volume.
AMD reported revenue of $1.233 billion and a GAAP loss of $504 million. In the quarter a year ago, revenue was $1.33 billion and the company had a profit of $259 million.
Gross margins fell to an unbelieveable 31%. In the same quarter a year ago, they were 59%. In Q4 06, 40%.
The company expects the next quarter to be flat.
The company's cash dropped to $1.167 billion from $1.541 billion at the end of last year. Receiveables dropped from $1.14 billion to $667 million. Never a good sign.
Advanced Micro Devices was expected to report a loss of 48 cents per share for the first quarter. Analysts on average, expect AMD to report a 5 percent sales decline to $1.26 billion
The company this month forecast sales of $1.22 billion after last month saying it was unlikely to meet its previous target for sales of $1.6 billion to $1.7 billion.

http://www.247wallst.com/2007/04/amd_blood_in_th.html

[bonesaw]

whos buying one of these the first few days its out?

[keithwalton]

Originally Posted by gizmo

On page 3, Intel's benches show the Conroe outperforming the Wolfdale in some benches, and by a sizable margin.

I have looked again and stick with my original post, there are NO conroes on page 3, only Wolfdales, Yorkfields and Kentsfield's

As i said before the QX6800 is not a conroe it is a Quad core kenty'

The conroe is already a very efficent core clock for clock. its got the A64 at least matched and mostly beaten in nigh on all usefull cases. Finding 10% is no mean feat, if a 3GHz penryn can keep up with a 3.3GHz conroe then thats good.

This is still very early silicon, at this stage of development conroe was only running at 2GHz. Its all about Quad Core clockspeed really, most kentsfield chips even in the hands of hardcore overclockers cant get much past 3.5GHz mark, if intel are hitting 3.33GHz on these early cores with a managable TDP, over clockers are going to hit 4GHz + with ease

The reason why they're not building faster chips now is so that they can sell faster chips later. amd are currently putting no presure on intel in the performance front and in terms of value amd are currently unsustainable. As loosing $600 million in a Quarter aint good for business (and is also bad practice, if demand exceeds supply then they should most defo be making money) stock holders will loose confidence especially if intel start the slow leaking of benchmarks trick again suggesting that they have more than a match waiting for anything amd can do in the cpu front.

Amd need the R600 to take off, sell well and make them a fortune else Q3 aint going to look good either. (it will be to late for Q2) and if they've just lost nearly 1/3rd capitol they cant keep that up

[Gizmo]

Originally Posted by keithwalton

I have looked again and stick with my original post, there are NO conroes on page 3, only Wolfdales, Yorkfields and Kentsfield's

As i said before the QX6800 is not a conroe it is a Quad core kenty'

By golly, you're right. I plumb skimmed over the fact that it was a quad-core chip at 2.93 GHz.

However, a Kentsfield IS a Conroe at its heart; you can stack all the multichip modules you want on there and give them any name you like, but that doesn't change the fact that you've taken multiple Conroe cores and pasted them together with some glue logic. When you look at the Yorkfield's numbers compared to the Kentsfield, and then take into account the difference in clock and bus speed (and allow for the fact that Yorkfield is true quad-core vs. Kentsfield 2+2 approach) the performance difference isn't that stellar.

So, it really looks like the only real advantage Penryn will have is clock scaling, and that remains to be seen.

[keithwalton]

Actually yorkfield is just two penryn (wolfdale) cores on one substrate its still a 2+2 design. They are both really just a die shrink with a few tweeks to the core.
The shrink is allowing them to up the cache from 4 megs to 6 megs per core and raise the clockspeed for the same TDP.

I am impressed with it clocking this fast this soon, its probably 2nd and at most 3rd generation silicon by now which would mean they started making it BEFORE the first samples came out of the fab (lead time is about 3 months i believe)

[Arctucas]

My question is; has anyone heard if the Yorkfield and Wolfdale procs will work with nVidia 680i mobos?

[Pitch]

That's firmly in nVidias ball court. But as of yet, I don't think there's any -concrete- evidence on whether this chips will run on existing chipsets.