What happened to AMD's C3 stepping?

[jvroig]

I thought the NDA about the C3 stepping lifts at Nov. 4... I've been looking for reviews after the NDA, but I've come up empty. All the pre-Nov 4 stuff I can find seem to be not as spectacular as Intel's D0 stepping. Will there be a review about this at anandtech? (I'd send an email to Anand, but given that there's a contest right now, the email will probably get lost amongst the thousand emails coming from the contest)

Anybody here know more about the new stepping?

Thanks.

 

[Idontcare]

Phenom II X4 965 Black Edition ver. 2.0: AMD Introduces C3 Processor Stepping

 

[nyker96]

quite a few reviews are up now, here's one:
http://www.xbitlabs.com/articles/cpu/display/phenom-ii-x4-965-c3.html

bumer idk beat me to it by seconds.

 

[jvroig]

Thanks.

No wonder AMD didn't seem to make too big a deal about it, it's not really that spectacular. Hehe, I guess it's my fault for hoping the C3 stepping will make the Phenom II as overclockable as its Intel counterparts.

But maybe the effect of this C3 stepping is not so great for the Phenom II architecture, but perhaps it will be much better on future architectures, like Thuban? Will Thuban be overclockable like crazy? Ah... who knows, really, but at least what is believable at this point? (And when is Thuban due, Q3/H2 of 2010?)

bumer idk beat me to it by seconds.

How come I regularly see "Idontcare" referred as "idk' instead of 'idc'? Just curious if there's an awesome backstory about this

Thanks guys.

 

[Idontcare]

Thanks.

No wonder AMD didn't seem to make too big a deal about it, it's not really that spectacular. Hehe, I guess it's my fault for hoping the C3 stepping will make the Phenom II as overclockable as its Intel counterparts.

Exactly, AMD didn't promise anything with this stepping...the hype was all "home-grown" by the enthusiast community as it were. (you are shocked, I know)

That said, it didn't help that AMD slapped an NDA on the stepping, that was the atypical part that only helped to fan the flames of rumor and speculation regarding the prospects of what AMD was trying to hide.

But maybe the effect of this C3 stepping is not so great for the Phenom II architecture, but perhaps it will be much better on future architectures, like Thuban? Will Thuban be overclockable like crazy? Ah... who knows, really, but at least what is believable at this point? (And when is Thuban due, Q3/H2 of 2010?)

For the record understand that the architecture of PhII is really very much the same thing as the architecture in Istanbul and Thuban as well as Shanghai and Athlon II, etc.

All AMD 45nm CPU's share the same basic Stars G45 K10.5 core architecture, the differences between the products comes down to differing configurations of cache hiearchy and different ways to deal with cache coherency between seperate die connected in the same system (either multi-socket or MCM).

So we really wouldn't consider Thuban to be a different architecture than PhII X4 in the way we like to use the term to describe materially significant changes that do exist between say the 65nm Athlon X2 (K8 architecture) and 65nm Phenom (K10 architecture).

How come I regularly see "Idontcare" referred as "idk' instead of 'idc'? Just curious if there's an awesome backstory about this

Sorry to dissapoint but there is no awesome backstory to this as far as I know, now it could be some kind of cruel inside joke that I've just never caught onto but I think folks just make it synonomous IDC=IDK (idontcare=idontknow). It don't bother me a bit in any event, it beats being referred to as dipshit or numbnuts

 

[Pederv]

Looks like reviewers are getting lazy.
Give them a chip with an unlocked CPU multiplier and they don't bother pushing the overclock on the data path.
One of the fixes was the downclocking of the memory when 4 sticks were used, did anybody see any verification on this?

 

[jvroig]

For the record understand that the architecture of PhII is really very much the same thing as the architecture in Istanbul and Thuban as well as Shanghai and Athlon II, etc.

Ouch, then I misunderstood. I thought Thuban would be a different architecture. So that means the only new AMD archi would really be only for Bulldozer at 2011?

At the risk of being guilty yet again of home-grown hype, I thought Thuban would outperform Deneb significantly (especially over-clocking potential) instead of just being a "6-core Deneb + 'Dragon in Arabic' name". So I guess Thuban really will be just a 6-core Deneb? IDC, shoot me the bad news now so I can stop expecting Thuban to be significantly better than it otherwise really would be Or is it really too early to tell?

Thanks.

 

[brybir]

Ouch, then I misunderstood. I thought Thuban would be a different architecture. So that means the only new AMD archi would really be only for Bulldozer at 2011?

At the risk of being guilty yet again of home-grown hype, I thought Thuban would outperform Deneb significantly (especially over-clocking potential) instead of just being a "6-core Deneb + 'Dragon in Arabic' name". So I guess Thuban really will be just a 6-core Deneb? IDC, shoot me the bad news now so I can stop expecting Thuban to be significantly better than it otherwise really would be Or is it really too early to tell?

Thanks.

That is basically correct as I understand it. 4 ---> 6 cores, same basic architecture with what will likely be some tweaks here and there to optimize it for 6 core operation instead of 4. I don't think anyone is expecting anything major beyond an expected performance increase in applications that can take advantage of the additional cores.

 

[richierich1212]

Yep. But Intel's got HT working very nicely. Even with only 4 physical cores their processors still prove that more cores does not equal better performance. But we shall see.

 

[Idontcare]

Ouch, then I misunderstood. I thought Thuban would be a different architecture. So that means the only new AMD archi would really be only for Bulldozer at 2011?

At the risk of being guilty yet again of home-grown hype, I thought Thuban would outperform Deneb significantly (especially over-clocking potential) instead of just being a "6-core Deneb + 'Dragon in Arabic' name". So I guess Thuban really will be just a 6-core Deneb? IDC, shoot me the bad news now so I can stop expecting Thuban to be significantly better than it otherwise really would be Or is it really too early to tell?

Thanks.

Its too early to tell what Thuban will have in store for us in terms of power-consumption and non-suicide clockspeeds on air and water.

But the architecture determines the IPC, and yeah we can't really expect much to change there for single-threaded apps, nor can we really expect much more than linear scaling of performance from the change in core-count alone between a 4core deneb to 6core Thuban.

(in fact if Istanbul is anything to go by for indication of what to expect of Thuban's cache hiearchy then we should assume core count increases but the shared L3$ size does not, so if anything multithreading performance will scale all the less than the core-counts would suggest because cache/thread starvation opportunities will be higher in thuban than on deneb)

My expectations of Thuban are same peak clocks and same power-consumption at those clocks as the original pre-C3 stepping Denebs...which is a phenomenal thing to accomplish if you think about it. IPC will remain unchanged and the absolute best case scenario is that we see a 50% increase in the aggregate IPC that 6cores offers over 4cores on those corner-case apps which require little cache and invoke little interprocessor communications during execution (pov-ray for example, cinebench is another).

 

[jvroig]

Same peak clocks? At 6 cores? And at the same 140W TDP? If they can make a 6-core 140W TDP CPU reach 3.2Ghz (stock - just assuming this), then why can't they make a quad-core @ 4Ghz stock at the same TDP? Or, release a quad-core at the same TDP but is amazingly easily overclockable to beyond 4GHz without LN2?

 

[Idontcare]

Now you're using the old noodle to contemplate what AMD's options are for 2010...and you are headed exactly where I am thinking AMD is headed.

By end of 2010 we will see 4GHz stock deneb SKU IMO. Mind you it had to compete with Sandy Bridge near the end of its life before 32nm BD is rolled out and 32nm gulftown/clarkdale for the early part of the year, so its not like the coming higher-clocked higher-core SKU's are going to be a panacea of cash flow for AMD.

 

[GaiaHunter]

A few more reviews:

http://www.tomshardware.com/reviews/phenom-x4-965,2468.html

http://www.firingsquad.com/hardware/amd_phenom_ii_x4_965_125w/

http://www.guru3d.com/article/phenom-ii-x4-965-be-revision-c3-review-test/

.
One of the fixes was the downclocking of the memory when 4 sticks were used, did anybody see any verification on this?

From firing squad review:

As you can see, the new C3 revision incorporates two enhancements beyond lower TDP. We typically disable power saving features like C1E in order to produce optimal CPU performance, then re-enable it when running power consumption. The new C3 chips can switch power states much quicker than older CPUs based on AMD’s C2 revision, reducing the performance hit you can sometimes encounter in some apps when power management features are enabled in BIOS.

The other significant tweak AMD has incorporated into their new C3 revision is support for up to 4 DIMMs at DDR3 1333MHz speeds. Previously AMD was limited to supporting just 2 DIMMs at DDR3-1333.

Guru3D says something similar. On the other hand both seem to have used 2x2GB sticks though.

On the other hand, they have reached over 4GHz on 64 bits OS.

I guess in the end the buyer get a bit more for the same money and/or less power consumption. Additionally is there still a reason to buy a Core2Quad?

But yeah, until buldozer, AMD is second place. Then we will see what happens.

EDIT: Guess the thread advanced a bit while I was posting this...

 

[jvroig]

Thanks, GaiaHunter.

I'm excited about Bulldozer as the next guy... but considering Intel has about a bajillion dollars more in R&D it can funnel, I'm not holding my breath. My current PC is powered by an Athlon X2 5000+, and I'm planning to replace it soon with a Phenom II, so I like AMD (but before the Athlon X2, my 3 previous computers were all Intel powered: Pentium I, Pentium III, then a Pentium IV, before going AMD on the next upgrade), but given the current status of both companies, I'm not expecting AMD to hand Intel a beating like the Pentium vs Athlon days (but I'm still expecting an awesome processor, even if it won't exactly beat Intel's own)

 

[Fox5]

Yep. But Intel's got HT working very nicely. Even with only 4 physical cores their processors still prove that more cores does not equal better performance. But we shall see.

HT does work very nicely, and as Intel's cores are more powerful than AMD's, a 6 core Thuban may not be altogether better than a 4 core i7, even in heavily multithreaded apps. (oh sure the 6 core may win, but not by a lot)

BTW, what new is expected for bulldozer? I don't recall reading anything about it that even gives it a hope of matching i7 clock for clock, so unless high clock speeds are in the cards, I can't see it overtaking Intel.

 

[nyker96]

Thanks.



But maybe the effect of this C3 stepping is not so great for the Phenom II architecture, but perhaps it will be much better on future architectures, like Thuban? Will Thuban be overclockable like crazy? Ah... who knows, really, but at least what is believable at this point? (And when is Thuban due, Q3/H2 of 2010?)

Thanks guys.

I seriously doubt Thuban will be too Ocable. One reason is they will be produced on 45nm soi same as the current crop of PIIs. So with two extra core tagged on, it will be even harder to reach high clock speeds. besides I have some doubt that the current m.b. can even take a monster like Thuban on. I predict there will be some burnt out boards out there upon Thuban release like the cheap 1156 boards that couldn't handle 150W+ overclocks because they use low grade power circuitry. but that just a personal opinion. I still say 32nm is better for 6 to 8 cores. I will also be funny if AMD introduces X1 X2 ... X6 when Thurban comes for all the defective cores.

 

[Pederv]

Guru3D says something similar. On the other hand both seem to have used 2x2GB sticks though.

My point exactly. The review sites seem more interested in the overclockability of the C3 stepping than in the fix that has plauged the AMD IMC design since it was released.

My S939 4800 X2 took a big performance hit once I upgraded from 2 X 1MB to 4 X 1MB.

Since the 965BE that I just got still has the same problem, I figured that I would wait to increase the CPU speed until I got around to filling all 4 of my memory slots. An attempt to offset the performance hit of the down clocking the memory.

 

[deputc26]

Honestly if AMD can match Deneb's clockspeeds with Thuban at the same TDP I will be blown away. I think it's probably more realistic to expect clocks to be 10% lower, and even that would impress me.

 

[Idontcare]

Thanks, GaiaHunter.

I'm excited about Bulldozer as the next guy... but considering Intel has about a bajillion dollars more in R&D it can funnel, I'm not holding my breath.

Yeah who cares what Intel is fielding at the time, if BD is better than phenom then it is a win for us consumers. If Intel is fielding and even better CPU then that is yet another win for us consumers.

Its a great thing about this industry, all we consumers have to do is live another year and we'll have all the better performing products at our option to buy for the same or lower price.

I don't expect the moon from AMD, 1/4 to 1/6 the budget of Intel can't help but make an impact. But we don't expect things to get worse for AMD ala the P3->P4 transition for Intel.

BTW, what new is expected for bulldozer? I don't recall reading anything about it that even gives it a hope of matching i7 clock for clock, so unless high clock speeds are in the cards, I can't see it overtaking Intel.

Matthias' blog showcases what we can reasonably expect of BD based on AMD's patent application trajectory.

The CMT, clustered multi-threading, looks pretty exciting to me.

http://citavia.blog.de/

I seriously doubt Thuban will be too Ocable. One reason is they will be produced on 45nm soi same as the current crop of PIIs. So with two extra core tagged on, it will be even harder to reach high clock speeds.

I agree, however I just want to point out that unlike with the Intel chips you can set each individual core's clockspeed for AMD cpus (presumably will hold true for thuban as well) so while the chances of all 6 cores reaching higher clocks across the board is reduce we do have the opportunity for creating an OC profile which extracts the max performance possible from the CPU overall.

AMD really went all out with this innovation. I just wish they'd implement their own version of Intel's PCU.

My point exactly. The review sites seem more interested in the overclockability of the C3 stepping than in the fix that has plauged the AMD IMC design since it was released.

How much of this was their own decision versus the guidance given to them by their AMD sponsors steering them towards showcasing certain attributes over others?

Specifically regarding the IMC situation, my understanding was that the overwhelming majority of desktop apps are basically invariant to the memory bandwidth that today's dual-channel and triple-channel DDR2/DDR3 setups have to offer anyways. Will anyone actually find their AMD rig to be any faster with a new C3 stepping chip in it running DDR3-1333 instead of DDR3-1066?

 

[Karaya]

1st post from a new poster but an very old reader.

What happened to AnandTech's review of the new C3 stepping?
All other major sites have made their reviews by now.

 

[nyker96]

I agree, however I just want to point out that unlike with the Intel chips you can set each individual core's clockspeed for AMD cpus (presumably will hold true for thuban as well) so while the chances of all 6 cores reaching higher clocks across the board is reduce we do have the opportunity for creating an OC profile which extracts the max performance possible from the CPU overall.

tweaking each core individually sounds cool but not sure how many people are doing it that way vs a simple all for one approach, but granted having this is definitely a plus.

I also have a question regarding the uArchitecture of BD, I'm not a chip designer so all this technical diagram doesn't speak to me. If you can, could you explain what type of advantages the new BD uA will have over the current k10.5 uA in layman's terms?

 

[jvroig]

I'm not a chip designer so all this technical diagram doesn't speak to me. If you can, could you explain what type of advantages the new BD uA will have over the current k10.5 uA in layman's terms?

I'll second that. How 'bout it, IDC?

 

[Idontcare]

I'm no chip designer either, we really need Hardball in here to help create a layman's Bulldozer vs. Westmere/Sandy list. But the two things that stand out to me are (1) decoders increase from 3 to 4, and (2) clustered integer units.

Regarding decoders, this is where core 2 architecture really started to one-up the old K8 architecture. http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2748&p=3

Core 2 (and Nehalem) can do 3 simple + 1 complex decode...netting a peak 5 decodes whereas K8 peaked at 3. Bulldozer looks to close this gap. (which would result in higher IPC for single-threaded apps and by extension multi-threaded ones as well)

Second is the clustering, as dresdenboy as envisioned it so far from the patent search results only the integer units are clustered but most desktop apps we use are more dependent on integer computations anyways. (for example one thing Phenom improved on over K8 was FPU execution and that was great for server and HPC stuff but in desktop markets we didn't recognize this enhancement because integer stuff wasn't improved as much)

I am no expert in architectures but the way I view clustered processing is that you bust up your execution units such that you then have the choice of processing one complex (large bit) instruction or two less complex (narrower bit) instructions at the same time...its not a doubling up of the execution units but rather better control over the way in which you load-up the existing one when you aren't using all the bits all the time.

This is no doubt a completely bassackwards way of viewing the proposed architecture improvements that BD will bring relative to K10.5 Stars G45 architecture but its the best I can conjure up until someone like Hardball comes in here and sets us all straight.

 

[lopri]

My point exactly. The review sites seem more interested in the overclockability of the C3 stepping than in the fix that has plauged the AMD IMC design since it was released.

My S939 4800 X2 took a big performance hit once I upgraded from 2 X 1MB to 4 X 1MB.

Since the 965BE that I just got still has the same problem, I figured that I would wait to increase the CPU speed until I got around to filling all 4 of my memory slots. An attempt to offset the performance hit of the down clocking the memory.

I'm in agreement on the point that reviews should have done with 8GB of RAM, but the rest of your post doesn't make much sense to me.

 

[jvroig]

Thanks, IDC. Perhaps you can do another of your "shiznit awesome" fanboy-parody posts about how kick-ass BD will be? That would certainly lighten up the AMD fan in me, given how pessimistic I find myself to be about AMD's battle against Intel (I still love AMD machines today, though, but mostly thanks to the i7's sticker price)

Now seriously, I'm just not seeing the bright side (yet). It's a David and Goliath fight, except there might be no God-who-will-bless-an-ordinary-slingshot in this case. The gap between the AMD and Intel flagship right now is just so wide (of course, so are the prices, which is the bright side of the story, thank God), that even if BD were to be better than Phenom II (awesome, of course), it might just end up as barely beating or even just equalling the i7, and that would mean another sound thrashing at the hands of Intel's new chip by then (not so awesome)

Given the chasm between Intel's and AMD's respective R&D budget (AMD being the company in the red and Intel being the company that can afford to lose a billion dollars and hardly feel it the next quarter), I just can't realistically see any other outcome right now. Add to that the fact that Intel's production capability is world-class. I'd even say "unmatched", except that I'm not sure that would be a fact.

Thanks to the price difference, my next machine will still be an AMD one (Phenom II, to replace my aging Athlon X2 5000+), and I hope that when I replace my soon-to-be Phenom II machine it would be a Bulldozer-powered one that will take its place. But I'm no fanboy, so if BD only happens to match Nehalem (is that even possible or am I being too pessimistic?), then I'll probably end up getting an Intel proc that time.

Lots of words so far, but the summary is just: I hope AMD really delivers on Bulldozer, and if anybody here knows anything concrete about Bulldozer that can possibly point to it being beyond awesome, I'd be glad to hear it.

Thanks.