amd fightback 2007??

str · 20 Jun 2006 at 19:45

james.miller said:
i've already posted the link in a previous post oblivion is also around 50% faster on average with a conroe @ 1600x1200

It seems a bit ridiculous that Conroe can be so inconsistent at showing it's performance advantage (HL2 vs fear) which leads me to suspect the testing might have been flawed. Not that I disbelieve that guys results I'd rather wait for more results to back him up and if it's proven to be real then yeah that's an incredible performance advantage.

str · 20 Jun 2006 at 19:47

NathanE said:
Yup scalability is the key. Even if AMD launched something to match Conroe... It would take Intel about a month to react by increasing the clocks and/or FSB. However Conroe scalability on the multi-core front is still very very very poor. All of Intel's upcoming quad core chips require a dual FSB. Ludicrous. AMD's Hypertransport literally takes an engineer about 5 mouse clicks to add another core or interconnect to another CPU socket. With Intel's current FSB design it takes months of work by a whole team of engineers.

Personally I feel Conroe is _still_ a bit of a stop-gap solution (though obviously no where near on the scale that Smithfield was) for Intel. You get the odd person saying how they admire the Conroe architecture so much. But really it is nothing special. All they've done is taken the Pentium-M, put it onto a next-gen fabrication process, added a ton of new execution units, increased cache sizes, improved prefetching, improved ILP (macro/micro-ops fusion), and a few other logic tweaks. Nothing special there really. It's basically what they use to do back in the good old Pentium I to III days.

I don't believe Intel will be fully back in the game until they have a Conroe-derived chip running on a CSI interconnect. None of this ancient FSB crap.

Yeah, that's an informative post mate.

NathanE · 20 Jun 2006 at 19:55

str said:
It seems a bit ridiculous that Conroe can be so inconsistent at showing it's performance advantage (HL2 vs fear) which leads me to suspect the testing might have been flawed. Not that I disbelieve that guys results I'd rather wait for more results to back him up and if it's proven to be real then yeah that's an incredible performance advantage.

It's down to the software and that some of the games are hitting the GPU limit rather than the CPU limit. Software with plenty of SSE1/2/3 receive the largest boost from Conroe because it has 3 vector logic units whereas all other x86 chips to date have only had 1. Basically a Conroe has 3 Oxbridge mathematicians inside it, working in parallel, whereas an A64 or a NetBurst chip only have 1. Big difference.

str · 20 Jun 2006 at 20:02

Yeah I know what you mean and the testing probably should have been much less about the graphics cards and more about the CPU. It might actually be his SLI setup wasn't working properly on the Opteron resulting in those huge differences which would better explain the almost unbelievable difference IMHO but sure if it's actually Conroe is that good at running code then that's a cool thing but I'm gonna stay on the fence for now regarding those results.

Cyber-Mav · 20 Jun 2006 at 20:23

amd is not going to win this time.

GAMEfreak · 20 Jun 2006 at 21:17

None of this ancient FSB crap.

well we can see if its going to make much of a difference by overclocking the FSB and dropping down a multiplier to compare the chip at the same core speed but with a higher FSB. maybe i'm wrong but i doubt a conroe chip will require anything over the 1066fsb, until 3.2ghz or more. thats just my opinion. we shall see though...

str · 20 Jun 2006 at 21:40

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2771&p=5

The above link shows fear results in Anand's X6800 (2.93 GHz) comparison with the FX-62 (2.8 GHz) as well as results from Quake 4 which is very heavily SSE optimised and BF2.

So yeah Conroe is much more impressive now I've seen more of the preview benchmark results.

NathanE · 20 Jun 2006 at 21:53

GAMEfreak said:
well we can see if its going to make much of a difference by overclocking the FSB and dropping down a multiplier to compare the chip at the same core speed but with a higher FSB. maybe i'm wrong but i doubt a conroe chip will require anything over the 1066fsb, until 3.2ghz or more. thats just my opinion. we shall see though...

That's not the point I was getting at. FSB is inherently limited by the number of cores it can support... HT/CSI is not.

FrankJH · 20 Jun 2006 at 22:38

No point speculating about Conroe (IMO) until real cpus are on the street - as apposed to the hand picked parts that are floating around for reviews etc

Gashman · 20 Jun 2006 at 23:00

gotta ask ourselves the question about the next set of K8s though, sure K8 is getting a wee bit old now, so it doesn't stand a chance against conroe in its current incarnation, but to be fair it never did. AMD might however in there move to 65nm take the time to correct a few issues in the K8 core (such as the prefetch problems i heard a bit about) and add a few performance tweeks, SSE4 (if they don't already use it in AM2 chips). but i think the main thing K8 is gonna gain from the 65nm switch is clock speed, getting athlons upto higher speed will obviously bring down the gap to conroe, keeping them competative, they'll still have to have a price advantage over intel

ajgoodfellow · 21 Jun 2006 at 01:35

This certainly sounds promising but if it's a workstation/server chip as speculated then Intel could still have the desktop lead with Conroe

I'll be upgrading in December/January time and I'll look at workstation/server chips as well as desktop chips, depending on price of course! If they do indeed retail for £1.5k then that's just out of the question! I'm currently running a Socket 940 Opteron 146 with ECC Registered memory bought over 18 months ago (and not top of the range at the time) and it can still hold its own against the newer chips

Conroe was a long time coming - before then the P4 was competing against a far newer, technically superior architecture and it was suffering. Intel have needed a new design for a few years now

This is the beauty of competition though. If it wasn't for AMD, Conroe might not have arrived (or at least taken longer) and we'd be stuck with the netburst architecture (which seemed to hit a wall at 3.8GHz). It is impressive how scalable the Conroe architecture seems to be

If AMD can ramp their clock speeds up this could buy them more time. It could be the case that we compare a 3.6GHz AMD vs a 3GHz Conroe (at similar prices). Not long ago the 1.8GHz A64s were comparing very favourably against the 3GHz P4s

As said, Conroe still has its problems, mainly the reliance on the FSB and the lack of an on board memory controller or Hyper Transport function

drunkenmaster · 21 Jun 2006 at 02:22

FrankJH said:
No point speculating about Conroe (IMO) until real cpus are on the street - as apposed to the hand picked parts that are floating around for reviews etc

retail cpu's have in the past very much tended to clock better than older ES samples, this has tended to be the case for both intel and amd es chips. i had a 939 3600+(2.2Ghz 1mb cache) before 939 was officially available and the week on the chip was just about before 754 launched. they don't perform differently clock for clock(just to cover myself i guess there can be crippled cache early built cores, or possibly even very very early ones with slight design differences but very unlikely).

games will never show big differences on cpu's, not for the last 3 years, not for the next couple. when you are maxed out on resolution due to gpu you will not be at the fps limit, the difference between a fx62 and a 3000+ will be under 5%. this is why games benchmarks are run at extremely low resolutions with TOP END gfx cards, to purely show the difference between cpu max fps, you won't ever see this difference when playing. so for games conroe won't bring anything new, only if you really use the cpu in other areas will it be helping out. or possibly with sli'd 1900xt's, 7900's, or most likely 7950's, stupid resolutions and then the higher clocking conroe's might help out.

as for K8L not being in desktop, and £1500 prices, really, honest to god some of the stupidest stuff i've ever heard. AMD just can't afford to make vastly different products yet for server and desktop. they have new fabs, but their current ones are being upgraded, they need production at max to pay for the multi billion projects they have running. they won't be having dedicated server/desktop/mobile chips for a good few years at least. its funny, as not one amd guy, not one rumour, not one industry insider, not one wall street analyst has ever suggested it as a possibility.

at the moment they are pretty much dropping 1mb cache ath 64's(x2's aswell i'm not so sure about) as they need more cpu's for servers so will be pushing the 1mb to servers, pushing more 512kb cache waifers through at a lower cost due to die size as they need to bring down prices to compete now that they don't have the faster cpu. they simply can't compete selling slower chips that are more expensive. this will be the only difference we will see in the K8L server vs desktop markets, different cache sizes(the fx line will retain the cache, possible one or two other cpu's aswell).

FrankJH · 21 Jun 2006 at 08:06

drunkenmaster said:
retail cpu's have in the past very much tended to clock better than older ES samples, this has tended to be the case for both intel and amd es chips. i had a 939 3600+(2.2Ghz 1mb cache) before 939 was officially available and the week on the chip was just about before 754 launched. they don't perform differently clock for clock(just to cover myself i guess there can be crippled cache early built cores, or possibly even very very early ones with slight design differences but very unlikely).

Eventually retail chips do clock better you are right - but not initially. There are always hand picked ones that are given to reviews that really perform well and the first couple of weeks of a brand new chip dont usually work that well ( or chipset for that matter - where conroe has to have additions to the present intel chipset to be fully functional)

Also I think you are getting confused as when S754 was first released S939 hadnt even been thought of yet let alone designed and fabricated. THere may well have been S940 chips around at this time I am not wholly sure but definitely NOT S939

drunkenmaster said:
games will never show big differences on cpu's, not for the last 3 years, not for the next couple. when you are maxed out on resolution due to gpu you will not be at the fps limit, the difference between a fx62 and a 3000+ will be under 5%. this is why games benchmarks are run at extremely low resolutions with TOP END gfx cards, to purely show the difference between cpu max fps, you won't ever see this difference when playing. so for games conroe won't bring anything new, only if you really use the cpu in other areas will it be helping out. or possibly with sli'd 1900xt's, 7900's, or most likely 7950's, stupid resolutions and then the higher clocking conroe's might help out.

I would concur that its unlikely conroe will be a massive advantage to games players - unless they are using 24" w/s res or higher with top end cards. But then Intel chips havent really been gamer chips for quite a few years now, their major talent is in media encoding and design etc where the design of the CPU comes to the fore.

Obviously there will be inherent advantages due to it being a brand new design, but in general I would still think that this is where Conroe will be seen at its best

drunkenmaster said:
they won't be having dedicated server/desktop/mobile chips for a good few years at least.

They already have mobile A64 - surely this is a dedicated laptop / low power chip?

They also have multi (x4 & x8 core) chips available for servers - ok not a dedicated design like Itanium or such like, but different enough to not have in a workstation / general pc

Stelly · 21 Jun 2006 at 08:37

Cyber-Mav said:
amd is not going to win this time.

I agree with you there... especially now since Intel have just shown off the 45nm process

Stelly

drunkenmaster · 21 Jun 2006 at 11:17

FrankJH said:
snip

no, when 754 was out 939 was known about, was being finalised, and there were es chips available. i got the 3600+(pretty much what a current 3700+ diego is), a little before 939 was properly released but the dates for production on it were september-ish of when the 754 came out. most ES chips have way earlier production dates and don't infact clock better than most initial clocking chips.

infact i think cpu's very rarely get handpicked for reviews before release, most are ES and made ages ago and not all clock that well at all. for memory, gpu's and sometimes you get motherboards that in reviews clock great but don't when it comes to general release of the product. but most reviewers tend to be very very conservative on reviewing new types of cpu's, seeing how the clock with little extra voltage. i remember always thinking( i think a review site that ocuk has issues with so maybe can't be mentioned), that their reviews for overclocking were a bit pants as they'd not raise voltage. anandtech clock up cpu's now in motherboard reviews way after cpu's are released but generally not so much, and only on basic air without to much voltage increase. i had a 754 the first day you could in the uk, had a pre-production ES 3600+ before 939 was out, had a 3500+ newcastle when they came out all clocked great and beyond what any review told me.

the mobile parts at the moment are the same internally, just made to higher specs for lower wattage, IE they use higher quality silicon(possibly) or slightly tweak manufacturing process. but they are the same either way, same ipc, same performance at a given speed. itanium is intels very specific server product and costs a bomb, xeon and p4's are the same chips(even intel find it far far easier and cost effective with massive production ability to essentially make two versions of the same chip for mass production), and the p-m is a very different beast to either and is their specific mobile product.
amd's server chip is identical in every way to the desktop chips except it has all 3 ht connections enabled.

FrankJH · 21 Jun 2006 at 12:45

drunkenmaster said:
when 754 was out 939 was known about,

NO IT WASNT

AMD backtracked well after 754 debuted to create S939

drunkenmaster · 21 Jun 2006 at 17:28

FrankJH said:
NO IT WASNT

AMD backtracked well after 754 debuted to create S939

http://www.anandtech.com/showdoc.aspx?i=1937

2 months after 754 was properly available(thats just the first article i found, shows 939 chips. i knew, everyone apparently but you knew this. simple fact is not a single cpu, of any socket or any kind "suddenly" appears. there will 100% certainly be many many ES chips floating around that have earlier dates than the one i had. amd never backtracked once, i think it was around a month after 754 launched that they mentioned 939 would be out but it was a leaked fact that everyone had known for a while. 754 was no great secret in the first place. companies can't make a cpu and two weeks later have it on the shelves. mobo makers have everything ready to go months in advance of release, cpu's have small batches of future cpu's ready way way before release dates.

"As a result, almost since its launch, enthusiasts have been waiting for Socket 939 to bring dual channel memory to the Athlon 64 line." from http://www.anandtech.com/showdoc.aspx?i=2065&p=1

FrankJH · 21 Jun 2006 at 17:43

Apart from the fact you said BEFORE 754 was launched , and the article you clarify it with says POST launch, I never said that any CPU suddenly appears.

Apart from the fact that I have a degree in Electronics and being in the IT industry for nearly 2 decades (building PC's and networks etc) common sense tells you that things like this - in fact most products of any type - rarely just appear; design , developement and testing all take their time

Even you have stated "I think"

Beleive what you will....life is well too short as it is

james.miller · 21 Jun 2006 at 17:46

frankJH he said he had an es chip that was dated before the launch date of the 754's. That is entirely possible, maybe you cant remember but a64's were always heading towards dual channel like their opteron s940 counterparts

FrankJH · 21 Jun 2006 at 21:03

james.miller said:
frankJH he said he had an es chip that was dated before the launch date of the 754's. That is entirely possible, maybe you cant remember but a64's were always heading towards dual channel like their opteron s940 counterparts

Thats what I am saying - I cant believe any S939 chips where even fabricated prior to S754 release, just seems HIGHLY unlikekly to say the least (or is he part of AMD's high inner core?? Even then I find it dubious)