New Barcelona Server benchmarks...

[manveruppd]

Yes, but they will be faster given the potential if seen here. The K10 in benchmarking is about 35% faster than a similar C2D. seen with my own eyes.

Eh? Didn't you even see the benchmarks I linked to? The Barkie they tested was about 10-20% faster than a similarly-clocked A64, which is about the same or slower than a similarly-clocked Conroe in most tests.

 

[knowlesy]

didnt it get like 30k in 3dmark06 ?

 

[Emlyn_Dewar]

But the whole point is that it's on a less than amazing motherboard with slow memory, as said above.
I'd rather wait until I see the benchmarks of the desktop chips on the desktop platforms...

 

[Gashman]

Seen it where i live, Engeneering sample in Thailand where i live. Shop promised me that i could buy it in a week or 2. So keep you posted. Will be happy to get that intel out of my pc though. put that together with a crossfire mainboard.

right on, ill watch this space with 'envious' eyes, it'll be ok as i have AM2 motherboard and it'll be great to know there new desktop chips will be worth the wait

 

[Gashman]

also is K10 still latency dependant like K8 was, cause it was also stupid comparison to C2D when one is handicapped slightly by high memory latency, plus don't forget conroe isn't that much faster than K8, marginally in some test, hugely in others but on average, its not leagues ahead like some claim it is, so all K10 needs is like 20% speed increase over K8 to equal core architecture, good scalability to challange in overclocking and good prices, if its faster than 20% ill be pleasantly surprised but its certainly gonna close the gap to intel by a fair chunk, lets all be happy AMD are back in the fight

 

[manveruppd]

also is K10 still latency dependant like K8 was, cause it was also stupid comparison to C2D when one is handicapped slightly by high memory latency, plus don't forget conroe isn't that much faster than K8, marginally in some test, hugely in others but on average, its not leagues ahead like some claim it is, so all K10 needs is like 20% speed increase over K8 to equal core architecture, good scalability to challange in overclocking and good prices, if its faster than 20% ill be pleasantly surprised but its certainly gonna close the gap to intel by a fair chunk, lets all be happy AMD are back in the fight

Agreed, unregistered RAM like real Phenoms will help them perform better than what that Opteron did in the "preview". And I do know that Conroe isn't that much faster than K8 clock-for-clock (except in SuperPi where it just rapes it, but I guess Barcelona's dual FPUs should even the score there). But how much difference do you reckon unregistered RAM will make? Cause the results in Anandtech's preview, while not bad, indicate that it won't be able to compete with Conroe equally unless it overclocks as well as it.

And I don't think AMD were ever out of the fight, their chips are still selling fater than they can make them. Their financial woes are partly due to the price war that Intel started, and I'm not sure Phenom can save them from that because Intel have much deeper pockets: even if K10 is significantly faster than Conroe, Intel can afford to cut prices to the point to the point where they'll still be competitive, and they can afford to sell at a loss for much longer than AMD can...
It's a pity that antitrust bodies in most countries move so slowly that AMD may already have gone under before they'll react.

 

[PhilGQ]

Anandtech has soem benchmarks as well with a 8800gtx overall about a 10-15% over the K8 clock for clock , for AMD's sake i hope there cheap

They kinda cheep if this is report is correct.

http://www.crn.com/hardware/201805351

$209 translates to just over a hundred quid + VAT and duty for the entry level and these are for a dual socket system. So hopefully when the Phenoms arrive they will be even cheaper.

Also bodes well for the Quad FX systems with 8 core over two cpu's.

 

[mrthingyx]

also is K10 still latency dependant like K8 was, cause it was also stupid comparison to C2D when one is handicapped slightly by high memory latency, plus don't forget conroe isn't that much faster than K8, marginally in some test, hugely in others but on average, its not leagues ahead like some claim it is, so all K10 needs is like 20% speed increase over K8 to equal core architecture, good scalability to challange in overclocking and good prices, if its faster than 20% ill be pleasantly surprised but its certainly gonna close the gap to intel by a fair chunk, lets all be happy AMD are back in the fight

Yes, they will still be largely latency-dependent, but AMD seemed to have compromised quite significantly in this area by adding the L3 cache.

This cache adds a massive (in comparison to that already in place with the AMD64s) 23-24ns latencfy penalty onto the chips which gives them a right royal shafting. I have no do doubt that adding some non-ECC DIMMs willl help matters, but it's not going to be the same.

However, the latency falls as the clock speeds increase...

 

[Corasik]

Hmm,

I thought Core2's were already averageing 20% faster than X2's clock for clock. And that's excluding the silly's like the synthetic benches where AMD score will due to their integrated memory controller, and also ignoreing SuperPI where Core2 rules due mostly to cache size.

Surely a 15% step up from X2 isnt going to cut it, especially as Penryn core is stepping up another couple of % on Core2.

All the talk of the new AMD being 20-35% faster than Core2, and it seems it will likely be about the same. I had believed that the enhanced FPU would be a strong point in its favour, but the results on the first benchmarks in this thread seem to indicated that its FPU is not really any better than X2.

Core 2's a year old now, and still going strong. Sure Phenom is a monolithic 'native' quad core processor. But most consumer and gaming apps bearly make use of dual core, let alone quad.

Not to mention Phenom's die is larger than X2's. That's going to cost AMD a lot of die space on each wafer, reducing profits. Penryn is considerably smaller than Conroe, even with its enlarged cache.

In a race to the performance crown driven by clock speeds on two cpu's with 'similar' IPC.. I would put my money on Intel. All the research they put into getting P4 running faster will give them a lot of potential to ramp up Core2.. As already show by core2's generally good overclocking records.

As for power saving, The intel Xeons didnt look that much worse, as the 'high power' intel was considerably faster. Comparing the lower powered systems where performance was similar, so was power consumption. AMD seem to have a few watts better performance, but that could be anything, motherboard, ram, a servers power performance is not 'just' the cpu. I doubt many datacentres will be crying running on intel or amd.. both are pretty good these days, and both are getting better on virtually every refresh.

Pitty the graphics card's arnt following the CPUs power wise. I believe even with my "OLD" 7900GTX that 50% of my systems power usage is the graphics card. Still beats PS3

 

[Deleted User 234324]

I still think AMD's quad core will do very well in retail, specially with places like PC World, Dell etc where marketting jargon is all they use to sell products.

If mr & mrs average joe come in looking for a system which are they more likely to pick the core 2 quad or the "worlds first native quad core processor" the phenom/opteron.

 

[PhilGQ]

I still think AMD's quad core will do very well in retail, specially with places like PC World, Dell etc where marketting jargon is all they use to sell products.

If mr & mrs average joe come in looking for a system which are they more likely to pick the core 2 quad or the "worlds first native quad core processor" the phenom/opteron.

More people will buy Intel whether it is better or worse. AMD was ahead on performance for years but never managed to become the world leader.

People will by intel simply because it a bigger company.

This "Phenomenon" (no pun intended) is know in marketing as Double Jeopardy (has nothing to do with being punished for the same crime twice).

If you are a bigger company you will sell more in the future.

 

[Corasik]

I still think AMD's quad core will do very well in retail, specially with places like PC World, Dell etc where marketting jargon is all they use to sell products.

If mr & mrs average joe come in looking for a system which are they more likely to pick the core 2 quad or the "worlds first native quad core processor" the phenom/opteron.

Hah Last time I was in the purple squad high street store, I overheard the 'salesman' explaining that the P4 was considerably faster than AMD's Athlon64, based on clock speed. Those shops are a nightmare. I think in the 'componant' area they are still selling P4's rather than Core2 Duo.. Guess the local ones in this part of the country had a large 'backstock' of P4s .

Mr and Mrs Average are probably more likely to have seen Intel TV ads, and were still blindly buying P4's in preference to the superior AMD platform. I dont believe AMD have got the marketting pressure to get the 'Average' buyers away from Intel that easily.

The Majority of Dell systems still ship with Intel processors for example.

 

[Gashman]

its an almost impossible task for AMD, intel have everything, there everywhere and all there marketting rubbish, the second AMD dissappear intel will just take the mick with prices and evolution will stop, competition is the engine of all progress, without competition, it stops

 

[tamalero]

Yes, they will still be largely latency-dependent, but AMD seemed to have compromised quite significantly in this area by adding the L3 cache.

This cache adds a massive (in comparison to that already in place with the AMD64s) 23-24ns latencfy penalty onto the chips which gives them a right royal shafting. I have no do doubt that adding some non-ECC DIMMs willl help matters, but it's not going to be the same.

However, the latency falls as the clock speeds increase...

I remember reading in one of the reviews, that the northbridge of the barcie, is not splitplaned, and thus is locked to a certain speed, with a HTT3 mobo.. it should allow splitplaned power.. and thus it should increase the memory access ( and since L3 seems to run at the speed of the memory controller.. )

I wonder if it will do what you say ( IE, reducing latency as clock gets higher.. )

 

[Clairvoyant]

So are AMD`s new Quads going to be

A) better than Intels current Quads.
B) Worth waiting for.

 

[stickroad]

So are AMD`s new Quads going to be

A) better than Intels current Quads.
B) Worth waiting for.

AMD Quads will be better in the sense of their Architect, as they are Native Quad Cores. Intels current Quads (Kentsfields) are just basiaclly two Core2Duo stuck together on one die. (So to speak)

As for clock for clock against the Kentsfields im not to sure.

Are they worth waiting for?

I would say so yes.

 

[Biffa]

You should know the answers surely?

 

[Clairvoyant]

You should know the answers surely?

If I have heard it once I have heard it a million times.

 

[Corasik]

AMD Quads will be better in the sense of their Architect, as they are Native Quad Cores. Intels current Quads (Kentsfields) are just basiaclly two Core2Duo stuck together on one die. (So to speak)

As for clock for clock against the Kentsfields im not to sure.

Are they worth waiting for?

I would say so yes.

Core2's are designed with extremely good branch prediction, so they rarely stress the FSB. So even with Quads linked by FSB rather than 'monolithic' there really isnt a huge penalty. Its perfectly possible to get all cores runing flat out, without ending up with a huge FSB bottleneck.. Intel didnt even bother upping the FSB to 1333 on most of the Quads yet, and its not a problem.

Judging by the first server benchies, the Phenom may actually be a shade behind Core 2 Duo clock for clock, although it will obviously excell in synthetic apps which deliberatly stress intercore communication, and the memory controller.

If you run vista, then the difference will probably be even less, as Vista is able to tell which cores are linked, and which communicate by FSB, and split the threads in the most efficient manner. (IE 1 application with two threads would be assigned to linked cores, with a separate application could be thrown over to a core only linked by FSB.).

So unless a single program is using all 4 cores, vista will actually sort it out very nicely. Fact is intel use 2 x 2 rather than monolithic because A) its cheaper, and B) with conroe's architecture it has very little impact on performance.

Even Penryn is still a 2x2 design, wont be monolithic until Nehelem. Thats how confident intel are that their current cpu sandwich design will continue to shine.

 

[mrthingyx]

And shine it does... whilst Intel's design is comparatively unsosphisticated (that is tongue in cheek when talking about 65nm components, etc., so please bear that in mind), it's brutally effective.

There is no pressing need for Intel to modify their approach any sooner than they plan to because of this.