Conroe Benchmarks

I only see marks for PCMark05 and a beta version of POV-Ray. I don't think we can make sweeping generalizations about the architechture with just that. The new conroe cores do sound great though. Can't wait for some more testing.

Well, you're still comparing an unreleased product to a released and available one. That's hardly fair.
We don't know:
- What kind of availability Conroe will have. For all we know, they might have only a dozen chips available at launch.
- Whether they'll perform as well in the general case. We've only seen a few benchmarks, mostly cherrypicked by Intel.
- Whether they'll all overclock as well as the press samples

IIRC, they just announced that the official Conroe chips would be clocked lower than the press samples, in fact. I think they both turned the FSB and core speed down a few notches.

And finally, the 4x4 thing is a bit more than that. Allowing *you*, the user, to choose to add a second cpu is a pretty nice option to have. And it's something Intel can't easily do because of their different architecture that's still using "good old" FSB, and off-die memory controller.

And clocking chips really high is not a bad thing. It's how CPU performance has managed to improve for the last 30 years. There's nothing shameful in that, and it's not "pulling an intel". If you've got a chip that can run at a faster clock speed, you'd be dumb not to use it.

Also, keep in mind this isn't all AMD has to counter the Conroe. Opening sockets to 3rd party processors sounds brilliant, and the K8L they're launching later should be able to close the performance gap as well. And their upcoming 65nm transition should help a lot too.

But yeah, AMD will *probably* be unable to claim the performance crown for a couple of months. Then again, they've had it for the last 3 years, so I think they'll survive. [wink]
And they've still got the most flexible and scalable architecture. In the higher end server space, the Opteron still reigns supreme. If you try to stick more than ~4 Conroes on a chip, they'll be memory-starved, just like the P4 was.

But nice to see Intel catching up. The Conroe looks great (although I'm more interested in its mobile counterpart)

Okay, maybe it isn't as bad as I thought. [smile]

But you're right, it's Intel's turn to be on top a little while. [smile]

I've seen several benchmarks of Conroe over the past few months. Most notably, a 5ghz overclocked Conroe that had a 10s Super Pi 1million time. Overall, Conroe seems to be destroying the competition.

Yeah, I feel like you are an AMD fan SpoonBender, and have a slight bias toward them. Everything I have seen has indicated that the new Intels will blow away the AMD chips. Not "probably hold the performance crown for 3 months." It is AMD's turn to step up to the plate, they are behind. Only time will tell if they can step up or not.

Quote:

Original post by Spoonbender Well, you're still comparing an unreleased product to a released and available one. That's hardly fair. We don't know: - What kind of availability Conroe will have. For all we know, they might have only a dozen chips available at launch. - Whether they'll perform as well in the general case. We've only seen a few benchmarks, mostly cherrypicked by Intel. - Whether they'll all overclock as well as the press samples

- There are already enought samples out there. Don't worry, the yields will be high enough. Intel doesn't disappoint in this category (at least I don't remember anything major).
- We have seen much more than a few benchmark. Conroe, as well as the rest of Intel's roadmap is very promising.
- Judging by past experience, most of the CPUs will overclock at least as well.

Quote:

Original post by vNistelrooy - There are already enought samples out there. Don't worry, the yields will be high enough. Intel doesn't disappoint in this category (at least I don't remember anything major).

Over the past few years intel has invested billions of dollars in their fab facilities. No corporation on earth has the fab capability that intel has now. If intel were to suddenly go out of business tomorrow, chip prices would skyrocket, because no one can make as many as intel can right now (including AMD)

That's one of the reasons why Apple switched to intel, instead of a more promising PPC provider, or even AMD - Apple has dealt with chip supply shortages for years and they want to go with a guaranteed supplier.

Quote:

Original post by vNistelrooy - Judging by past experience, most of the CPUs will overclock at least as well.

Overclock? How? By increasing FSB...

Conroe FSB 1066 MHz. MB max FSB 1066 MHz. There is a VERY little room for increase. (Of course you might hope Intel would unlock multiplier... You could hope...)



...

Yes Conroe effectivity per cycle is better than AMD, however not by too much, on the other hand AMD lost its fast memory access advantage, and Conroe is colder as well. MB for Conroe are more expensive however, because of pins on MB, and memory controller integrated on the MB.

Prices and clocks are known. Effectivity per clock is known as well, so you can calculate how it would looks like next few months. After Intel would remove Presslers from its manufacturing line, Conroe prices would drop even more. It's true that quad cores will not appear until 45 nm manufacturing proces would happen. If AMD will not skip 65 nm in favor of 45 nm it could have quad core on 65 nm.

Quote:

Original post by Raghar Quote: Original post by vNistelrooy - Judging by past experience, most of the CPUs will overclock at least as well. Overclock? How? By increasing FSB... Conroe FSB 1066 MHz. MB max FSB 1066 MHz. There is a VERY little room for increase. (Of course you might hope Intel would unlock multiplier... You could hope...) ... Yes Conroe effectivity per cycle is better than AMD, however not by too much, on the other hand AMD lost its fast memory access advantage, and Conroe is colder as well. MB for Conroe are more expensive however, because of pins on MB, and memory controller integrated on the MB. Prices and clocks are known. Effectivity per clock is known as well, so you can calculate how it would looks like next few months. After Intel would remove Presslers from its manufacturing line, Conroe prices would drop even more. It's true that quad cores will not appear until 45 nm manufacturing proces would happen. If AMD will not skip 65 nm in favor of 45 nm it could have quad core on 65 nm.

Actually, the FSB max will be 1300mhz. The samples some people have have been overclocked as high as 5ghz, so I wouldn't worry about that.

Quote:

Yeah, I feel like you are an AMD fan SpoonBender, and have a slight bias toward them. Everything I have seen has indicated that the new Intels will blow away the AMD chips. Not "probably hold the performance crown for 3 months." It is AMD's turn to step up to the plate, they are behind. Only time will tell if they can step up or not.

Nah, not really. I'm just anti hype and doomsday scenarios.
Conroe hasn't been released yet. So far, it looks awesome, but it still haven't been released yet. For all we know, Intel could change their minds tomorrow and decide only to release 500mhz versions of the chip. [lol]
Or they might be unable to get decent yields on it, making it unavailable in larger numbers. Or any of a dozen other things might happen.

Quote:

Original post by vNistelrooy - There are already enought samples out there. Don't worry, the yields will be high enough. Intel doesn't disappoint in this category (at least I don't remember anything major).

You don't? You only have to look back a few months. Their Core chips have only just recently been really available. Even Apple couldn't get more than a tiny number of them.
The samples they've released to the press are just that. Samples they chose to give to the press. They *might* be representative of the final chip, or they might be cherrypicked as the best of the best, easily overclocking to 4ghz while all the "normal" dies will have difficulty reaching 2ghz. We don't know yet.

Quote:

- We have seen much more than a few benchmark. Conroe, as well as the rest of Intel's roadmap is very promising.

We have seen a number of benchmarks, yes, but so far, not a single independent, neutral and guaranteed unbiased one. There are still NDA's in place, and in the existing tests, Intel has still dictated a lot of things which means we can't get an 100% accurate picture yet.

Quote:

- Judging by past experience, most of the CPUs will overclock at least as well.

Really? How much experience do you have with early press samples of a chip that has been desperately needed for 3 years, and which Intel is relying so heavily on?

I think the Conroe will take the performance crown for a while, at least. But I don't neccesarily think it'll get the 20-25% advantage some benchmarks have shown. It might, or it might not. I'll make up my mind when they drop the NDA. :)

Until then, I'll stick to two facts:
- The architecture looks kickass.
- Intel desperately needs this to be a success, meaning it's quite possible they do what they can to make the chip sound even better than it is. For example by giving above-average chips to the press to play with.

Oh, and a third one:
- Intel announced that the chips at release will be clocked lower than expected, and use only 1ghz fsb. That could mean any of a dozen things. Maybe they feel so superior that they dont need to pump it up to higher speeds, or maybe it's just that they can't get good yields at higher speeds.

Quote:

on the other hand AMD lost its fast memory access advantage

No they haven't. They still have lower latency due to the on-die memory controller. Conroe is better than Netburst at memory latency, yes, but it's not as low as A64.

Quote:

and Conroe is colder as well

Is it? What do you base that on?
What I've seen is pretty similar TDP figures for the two architectures, but then you have to keep two things in mind:
- AMD includes a memory controller. With Conroe, that goes on the motherboard, which isn't counted in the TDP, which means the Conroe-based system will generate a bit more heat than the CPU TDP claims.
- They measure TDP differently. AMD uses the absolute max heat it can dissipate (and rounds up for entire processor families, for example almost all current single-core A64's have a TDP of 67W, despite the low-end ones obviously producing less than that). Intel uses some kind of "the most we expect under normal use" figure, which is typically 10-15% under the actual max.

I also haven't noticed anything from Conroe that can compete with AMD's upcoming 35W line.

Quote:

Actually, the FSB max will be 1300mhz. The samples some people have have been overclocked as high as 5ghz, so I wouldn't worry about that.

The keyword here is *sample* [wink]

Personally, I'm looking forward to Conroe. Looks like a great chip. I'm planning on buying a merom-based notebook when they're released.
But let's not assume it's the holy grail and that it will reign supreme ever after. Or even that it's all that Intel claims. It's an unreleased product still.

Quote:

Yes Conroe effectivity per cycle is better than AMD, however not by too much

Depends on what you measure. For SSE, it's a beast, vastly outperforming AMD. [smile]

Quote:

Personally, I'm looking forward to Conroe. Looks like a great chip. I'm planning on buying a merom-based notebook when they're released. But let's not assume it's the holy grail and that it will reign supreme ever after. Or even that it's all that Intel claims. It's an unreleased product still.

Do you honestly think that the samples will be worse than the actual product? The amount of money it costs to produce those things, I doubt there will be any significant changes between the current ES out there and the final product.

Quote:

Original post by SticksandStones Do you honestly think that the samples will be worse than the actual product?

No, I'm saying they might be better. If you were Intel, what'd you do? You've been getting slaughtered in performance benchmarks for the past 3 years. Finally you have a competetive chip. Wouldn't it be tempting to pick the best of them to send to the press, just for maximum effect?

Quote:

The amount of money it costs to produce those things, I doubt there will be any significant changes between the current ES out there and the final product.

Oh, I don't expect any manufacturing differences. But some chips turn out better than others. Some are just plain defective, some work if you disable half the cache, or only work at relatively low speeds. And some can be overclocked like mad.

Quote:

Original post by Spoonbender Quote: on the other hand AMD lost its fast memory access advantage No they haven't. They still have lower latency due to the on-die memory controller. Conroe is better than Netburst at memory latency, yes, but it's not as low as A64.

Mostly true, but debatable. AMD has been stuck for quite a while with regular DDR (precisely because of the on-die memory controller) and only recently has begun with DDR2, so the jury is still out. Chances are, the newest A64's with DDR2 will have lower latency, but most A64's still have regular DDR and conroe can pull ahead there.

Quote:

Quote: and Conroe is colder as well Is it? What do you base that on? What I've seen is pretty similar TDP figures for the two architectures, but then you have to keep two things in mind: - AMD includes a memory controller. With Conroe, that goes on the motherboard, which isn't counted in the TDP, which means the Conroe-based system will generate a bit more heat than the CPU TDP claims. - They measure TDP differently. AMD uses the absolute max heat it can dissipate (and rounds up for entire processor families, for example almost all current single-core A64's have a TDP of 67W, despite the low-end ones obviously producing less than that). Intel uses some kind of "the most we expect under normal use" figure, which is typically 10-15% under the actual max. I also haven't noticed anything from Conroe that can compete with AMD's upcoming 35W line.

Also debatable. Power is a tricky thing to measure. Honestly, even for super power gaming users, CPU's sit idle a lot of the time while the bus takes a beating. In which case, it depends on which chip company implemented the best "average use" power consumption, which we won't really find out until release and people measure real-life temps and battery life.

Quote:

They *might* be representative of the final chip, or they might be cherrypicked as the best of the best, easily overclocking to 4ghz while all the "normal" dies will have difficulty reaching 2ghz. We don't know yet.

On here, Conroe XE at 3.5GHz overclock need liquid cool. I can't imagine what kind of cooling solution needed for 4GHz and, that is on 65nm. Remember that AMD is still on 90nm.

Quote:

You don't? You only have to look back a few months. Their Core chips have only just recently been really available. Even Apple couldn't get more than a tiny number of them.

True. Even Dell having problem securing Core chip supply.

Quote:

Also debatable. Power is a tricky thing to measure. Honestly, even for super power gaming users, CPU's sit idle a lot of the time while the bus takes a beating. In which case, it depends on which chip company implemented the best "average use" power consumption, which we won't really find out until release and people measure real-life temps and battery life.

Things to consider are:
1. AMD CPU have built-in memory controller while Intel have none. So do we need to take account of FSB/north bridge chip power as well?
2. AMD uses max load for its TDP, while Intel uses 'typical' load for its TDP rating.
Yes it is hard to measure... I think it's easier to measure full system consumption.

Quote:

AMD has been stuck for quite a while with regular DDR (precisely because of the on-die memory controller) and only recently has begun with DDR2, so the jury is still out. Chances are, the newest A64's with DDR2 will have lower latency, but most A64's still have regular DDR and conroe can pull ahead there.

They are not stuck but, it is not commercially viable from moving to DDR2 back then - more expensive, and higher latency. Having more cores, higher frequency, and mature of the DDR2 market/manufacturing, the reason they move to DDR2 seems justified. I think AMD AM2 will be better off maxing it potential after its 3rd revision meanwhile Intel is trying on FB-DIMM to squeez performance.

Quote:

Original post by DerekSaw Quote: Also debatable. Power is a tricky thing to measure. Honestly, even for super power gaming users, CPU's sit idle a lot of the time while the bus takes a beating. In which case, it depends on which chip company implemented the best "average use" power consumption, which we won't really find out until release and people measure real-life temps and battery life. Things to consider are: 1. AMD CPU have built-in memory controller while Intel have none. So do we need to take account of FSB/north bridge chip power as well? 2. AMD uses max load for its TDP, while Intel uses 'typical' load for its TDP rating. Yes it is hard to measure... I think it's easier to measure full system consumption.

Due to some changes between DDR and DDR2 the amount of power used by the memory controller should be much less. So while AMD has some advantage power wise due to the memory controller being counted in TDP and intel isn't, that advantage will be less than in the past.

I, for one, will wait until representive laptops are out and then we can look at real-life battery as a judge - that's the only measure that really matters to most people in the end.

Does anyone get the feeling that we're reaching point where there is just too much technology in these things to be able to make accurate comparisons? It just feels like for every statement, there's a viable counter argument, and the only comparisons people can ever agree on are the benchmarks of software and games.

Quote:

Original post by etothex Quote: Original post by Spoonbender Quote: on the other hand AMD lost its fast memory access advantage No they haven't. They still have lower latency due to the on-die memory controller. Conroe is better than Netburst at memory latency, yes, but it's not as low as A64. Mostly true, but debatable. AMD has been stuck for quite a while with regular DDR (precisely because of the on-die memory controller) and only recently has begun with DDR2, so the jury is still out. Chances are, the newest A64's with DDR2 will have lower latency, but most A64's still have regular DDR and conroe can pull ahead there.

What do you mean? DDR2 doesn't have lower latencies than DDR. (Actually, it's the other way around, but the difference is rather small compared to the effect of the on-die memory controller)
Memory bandwidth != latency.

Quote:

Also debatable. Power is a tricky thing to measure. Honestly, even for super power gaming users, CPU's sit idle a lot of the time while the bus takes a beating. In which case, it depends on which chip company implemented the best "average use" power consumption, which we won't really find out until release and people measure real-life temps and battery life.

True. So far, AMD has been doing pretty well in that scenario, with Cool & Quiet doing a lot to lower idle temperatures. But Conroe might catch up. Hard to say.

Quote:

I, for one, will wait until representive laptops are out and then we can look at real-life battery as a judge - that's the only measure that really matters to most people in the end.

Except for one thing. We're not talking about laptop cpu's. Athlon 64 and Conroe are both desktop parts. [wink]

But of course, they're closely related to their mobile counterparts (Turion and Merom)

But yeah, as you said, we'll see at release, when we have actual products and benchmarks without NDA's to make the comparison. At any rate, it should be an interesting couple of months.

Quote:

Original post by evanofsky From ExtremeTech Summing up for the lazy: the Athlon FX-62 got pwnt. So, what do you think? Is it a hoax? Will you wait for third-party benchmarks before you believe it? Personally, I think AMD had it coming. AM2 isn't good enough, IMO. It's just a slight revision of the Athlon64 with DDR2 support and the usual Mhz increase. Whereas Conroe IIRC is pretty much all new. And now AMD plans on a "4x4" quad-core architecture? Sounds like they're pulling an Intel, to me. If you can't beat 'em, slap four of 'em on a mobo, clock it really high, or plop an insane amount of cache on it *cough*P4 Extreme Edition*cough* Comments?

The K8 core is now 4 years old.


To put it another way, if the Conroe DIDN'T beat the K8... Intel would be truly incompetent.



Furthermore, the first conroe's will be on Intel's 65nm process, and the K8's they're being compared with are still on AMD's 90nm process; they recently opened up FAB36 in Dresden, which has machinery to produce chips on the 65nm process as well.

When AMD gets the 65nm chips out, they'll give the Conroe a run for its' money.


Just in time for AMD to release the K10. Rumors are that it will support inward hyperthreading. The idea behind this is that the only time software gets a boost from dual/quad cores is when it's specifically optimized for threading. However, most software isn't, so the benefit is often negligible.

Inward hyperthreading will make a dual/quad core chip look like a single core chip to the OS, and automatically thread everything on the fly; therefore automatically multi-threading even single-threaded applications.

:)

Intel ought to enjoy the performance lead while they have it.

Quote:

Original post by Mithrandir When AMD gets the 65nm chips out, they'll give the Conroe a run for its' money.

Possibly. Although Conroe still has a few tricks up its sleeve that K8 can't match.

Quote:

Just in time for AMD to release the K10.

Last I heard, K10 isn't going to be out any time soon, if it hasn't been completely cancelled. They're making a K8L though, which looks very nice.

Quote:

Rumors are that it will support inward hyperthreading. The idea behind this is that the only time software gets a boost from dual/quad cores is when it's specifically optimized for threading. However, most software isn't, so the benefit is often negligible. Inward hyperthreading will make a dual/quad core chip look like a single core chip to the OS, and automatically thread everything on the fly; therefore automatically multi-threading even single-threaded applications.

*groans*
I still think this is only a rumour caused by a virulent AMD fanboy disease.
"Intel sux, AMD rocks. Intel has done hyperthreading, hence hyperthreading sucks. So obviously AMD must be doing the opposite, which'd rock. Let's call it reverse hyperthreading!"
Sprinkle that with a few zomg's and !!1's, and you've got a very likely theory.

While it might be possible to implement, it's just not going to be worthwhile.
What you've described is basically an ordinary OOO-architecture. Which is what is so silly about this rumour.
It's being done already, except that it's done on a single core, because that's more efficient, than making multiple cores that have to somehow share a single thread. It takes less die space, and it reduces latencies (which would really cripple singlethreaded performance)

What they *might* be doing (and which would be awesome) is to make a really wide core, (so wide that certain fanboys might think of it as two cores), and then implement hyperthreading. That way, you'd essentially get the same results:
A chip that can run two threads simultaneously (if we assume it's twice as wide as a K8 chip, it could even do so with no performance penalty in theory), or which can "pool" the resources to execute a single thread more efficiently.
It's still the same concept as hyperthreading, but approached from the opposite direction. Instead of making a single-thread chip, and allowing a second thread to use leftover resources, you'd make a chip that's designed to run two threads via "hyperthreading", and then, when a single thread requires a lot of resources, it's got the entire core at its disposal. That's just speculation based on what I'd like to see, and what this rumour *might* mean, if translated with a healthy dose of imagination.