Haswell: News, Rumors & Reviews

[FALC0N]

I'm not sure why malmental posted those links. (I only read the second one) It doesn't show what he is claiming. While I normally ignore synthetic benchies they can play a part in this case. Check out the Cinebench R10 benchmarks. The first one is single threaded performance. This would/should ignore number of cores and can show which arch is better. But even though the x4 is clocked 100MHz higher, the x6 beats it. Because the arch's are so close to each other and the x4 is clocked higher, it should have the higher number. To claim that the x6 is slower per core doesn't reflect in that benchmark. AMD must have made improvements to the arch to allow it to be better.

Moving on to the real benchmarks the x6 isn't always as a % as high as the extra cores should make it, but it usually does well enough. Again, I'm not sure I see what Malmental does.

I just posted the same comparison malmental made, but used an intel 4 core vs an intel 6 core at the same clock. It showed the same basic performance pattern as deneb/thuban.

He is confusing scaling with core performance.

So I agree with you 100 percent.

 

[4745454b]

Yikes. That's scary....

 

[fazers_on_stun]

That would be nice for SB/IB users if they could do that, but I also wonder if the Ram will be only for the IGP or if Intel will make it dedicated to both CPU and GPU. 1GB of Ram would be overkill for any iGPU for quite a few years, but loading an entire program in the Ram for the CPU would be a benefit.

I sorta think Intel might go with another memory stack for a sort of "L4" cache in the package - seem to recall somewhere that Intel is concerned with cache thrashing and they took some steps with IB to minimize it. But yeah 16 enhanced EU's and 1GB dedicated GPU memory would be a, uh - what's that term BM is fond of?? Oh yeah - MONSTER!

Exactly what I think on that subject. Besides the tooting part. Though I wonder how much the extra cache bloated the cpu compared to old designs.

Well I went through some calculations over on the BD thread (probably got deleted however), starting with the 4-cour PII at 258mm^2 at 45nm, 6MB of L3. Going to 32nm should reduce the PII to about half that size, say 130mm^2. Going from 4 full cores to 8 CMT cores should add either 5% or 12% extra die space, depending on who you last talk to . Assuming that's per CMT core, thats either 20% or 50% more area, which brings us up to 195mm^2 using the 12% figure. Then accounting for the extra 2 MB of L3 should add maybe 10% more (a guess from the die shot over on S/A), the extra HT links and such, and I come up with around 250mm^2 as a rough estimate. Dunno what the Interlagos pricing is going to be, but I'd bet it ain't cheap - it is going to cost AMD a lot to have GF fab it if indeed the yields are low.

Since AMD seems to be having yield troubles, probably not many Interlagos parts at all. 70% yields(if they're even that good) at 315mm^2X2 is not a pretty sight.. AMD could lose quite a few sales, because they just can't put out enough Interlagos processors.
The recession also doesn't seem to be hurting Intel much at all. People have become so reliant on processors that it is almost a necessity these days. Heck, I basically only use my computer to watch tv/movies now, so it makes it an all-in-one entertainment center for me.

It's funny how some of the fanbois over on the BD thread mention that just because AMD has apparently lowered the price of the 8-core by nearly $100 even before the launch, it's not because of performance issues, it's because AMD wants to undercut Intel and take away market share. Well Hector Ruiz tried that tactic some years ago and got surprised when C2D came out at a relatively low price compared to K8, and tons more performance. AMD promptly started hemorrhaging cash, while Intel still made a hefty profit, probably due in no small part to more efficent and cost-effective 65nm process. If AMD were to try and reverse this with BD, I think they'd just get themselves back into the red as Intel can afford to cut prices far more than AMD, again in part due to superior process (yields & maturity on 32nm anyway).

Anyway, I guess we'll see BD benchies some time next month so from those plus AMD's quarterly report, we should be able to figure out some facts and dispell some rumors and hype on both sides ..

 

[sonoran]

It's funny how some of the fanbois over on the BD thread mention that just because AMD has apparently lowered the price of the 8-core by nearly $100 even before the launch, it's not because of performance issues, it's because AMD wants to undercut Intel and take away market share. Well Hector Ruiz tried that tactic some years ago and got surprised when C2D came out at a relatively low price compared to K8, and tons more performance. AMD promptly started hemorrhaging cash, while Intel still made a hefty profit, probably due in no small part to more efficent and cost-effective 65nm process. If AMD were to try and reverse this with BD, I think they'd just get themselves back into the red as Intel can afford to cut prices far more than AMD, again in part due to superior process (yields & maturity on 32nm anyway).

Anyway, I guess we'll see BD benchies some time next month so from those plus AMD's quarterly report, we should be able to figure out some facts and dispell some rumors and hype on both sides ..

I wouldn't put too much faith in those supposed price drops until we see something straight from the source (or at least something from a highly reputable source).

If poor yields at GF are the big problem (and who knows?), then someone grossly mismanaged the planning for this process upgrade cycle. Yields should have been solid months ago. And if they weren't, what does that imply about GF's commitment to producing leading edge process chips for AMD? It's one thing to have FAB R&D and production costs spread across multiple customers - that would save AMD money. But it would be a different story altogether if GF starts to view them as "that problem customer" that consumes too much of the FAB's money and time compared to other customers. Of course none of us know the inside story of what's going on over there, or in GF's corporate offices.

 

[Cygnus x-1]

Fazers, some of the AMD guys are so emphatic and emotionally tied to the product, it makes me question their sanity or if they are on the payroll.

 

[Cygnus x-1]

Falcon, I believe Malmental is correct about thuban. It's basically a denab with two extra cores. However, the extra cores become a bit of a bottle neck at the chips front end, hence the performance hit. It does overcome that problem in apps well suited to use all six cores, but in apps or games that it doesn't, it can be slower.

 

[Cygnus x-1]

LOL Mal, just trying to set things straight.

 

[4745454b]

It does overcome that problem in apps well suited to use all six cores, but in apps or games that it doesn't, it can be slower.

Not in the links provided so far. Someone want to open a new thread about this (or link old ones) and PM me? From what I've seen so far its not the case, but I don't want to crap on this thread anymore.

 

[fazers_on_stun]

I wouldn't put too much faith in those supposed price drops until we see something straight from the source (or at least something from a highly reputable source).

Well that's true of course . Guess I was giving too much credence to all the recent rumors, but then so was everybody else . But if true, then either AMD management is idiotic for underselling their flagship CPUs to gain marketshare (which I doubt), or performance is not too competitive. As sometimes mentioned here, Intel could forego a quarter or two without their billions in profits, but AMD would be stretched out on the meat wagon if they got into a serious price war.

If poor yields at GF are the big problem (and who knows?), then someone grossly mismanaged the planning for this process upgrade cycle. Yields should have been solid months ago. And if they weren't, what does that imply about GF's commitment to producing leading edge process chips for AMD? It's one thing to have FAB R&D and production costs spread across multiple customers - that would save AMD money. But it would be a different story altogether if GF starts to view them as "that problem customer" that consumes too much of the FAB's money and time compared to other customers. Of course none of us know the inside story of what's going on over there, or in GF's corporate offices.

Heh, that's what I've been saying for 2 years now, ever since the spinoff of the fabs. AFAIK AMD is GF's only 32nm SOI customer, so I don't see how GF could spread the fab expenses across multiple customers, unless they can run 28nm strained silicon through the same fab lines as the 32nm SOI wafers. But then the advantages of tweaking & tuning each step for performance & yields would be lost, having to adjust from one set of criteria to another..

Also, it should be obvious that once an organization splits into two camps, each with different goals and priorities and leadership, their interests begin to diverge pretty quickly and communication becomes more problematic. The change to only paying for good chips provision in the contract with GF is evidence that not all is well in the AMD/GF camp.

 

[Cygnus x-1]

It does overcome that problem in apps well suited to use all six cores, but in apps or games that it doesn't, it can be slower.

Not in the links provided so far. Someone want to open a new thread about this (or link old ones) and PM me? From what I've seen so far its not the case, but I don't want to crap on this thread anymore.

I don't have a link sorry. I'm going on what I have read here and on other sites. I remember seeing that info a number of times, including in the BD thread.

 

[fazers_on_stun]

Fazers, some of the AMD guys are so emphatic and emotionally tied to the product, it makes me question their sanity or if they are on the payroll.

And yet those same AMD fanbois claim to be rational, sane, etc etc. - we've all seen that over the last 5-6 years now .. There was an interesting article about fanboyism and how people's pride and feelings all get tied up in the choices they make - forgot the link already but it made sense.

I probably come across as favoring Intel too much, but really it's just a reaction to the green Koolaid kiddies more than anything else. As I've said for the last year or so, if BD is the better choice for gaming & the other stuff I do, according to my criteria (where performance is more heavily weighted than cost), then that's what I'll go with. BD is an interesting design and I'll give full innovative credit to the AMD engineers if it turns out as good as they say it will. But I also think the first gen chip might not live up to its potential, despite the "MONSTER" design ..

 

[fazers_on_stun]

But it would be a different story altogether if GF starts to view them as "that problem customer" that consumes too much of the FAB's money and time compared to other customers. Of course none of us know the inside story of what's going on over there, or in GF's corporate offices.

Update - check out AMD Fires Rick Bergman Under Cloud of Mystery :

Rick Bergman, Senior Vice President and General Manager at AMD has been fired according to HardOCP sources. AMD has given no reason as to why the well seasoned Bergman was leaving his position. It was told to us that Bergman had interviewed for the just-filled CEO position months back at AMD and had been turned down. Since then we have been lead to understand that AMD has lowered the boom on Bergman for not managing the GlobalFoundries relationship properly. Sources tell us that GlobalFoundries is simply not up to the task of supplying AMD its needed parts and Bergman is first in line when it comes to the responsibility of making sure AMD is sourced properly. We understand that Bergman will be the first of the dominoes to fall.

If true, then that would be some pretty hard evidence of the growing rift between AMD and GF..

 

[Cygnus x-1]

So what's the deal with that intel water cooler? Is that just a new product like their ssd's, or are the new 2011 socket chips gonna run that hot?

 

[Cygnus x-1]

Your gonna buy it without knowing how well it cools? I would feel safer with a h80 or just a good heatsink/fan.

 

[Cygnus x-1]

kinda messed up if SB-E really needs this thing to stay cool. Would seem like a step backwards to some degree. No pun intended.

I'm interested in seeing the benches as well, I can't see it doing better than a h80 or what ever is close to the same price, but who knows.

 

[fazers_on_stun]

kinda messed up if SB-E really needs this thing to stay cool. Would seem like a step backwards to some degree. No pun intended.

I'm interested in seeing the benches as well, I can't see it doing better than a h80 or what ever is close to the same price, but who knows.

Hmm, I had heard Intel wasn't going to bundle any kind of cooler with SB-E. At any rate, if they do bundle a water unit with it, I wouldn't worry about it as it would still have the warranty if the cooler leaked. So it would behoove Intel to make sure it was decent quality and last as long as the warranty period. Same with BD if AMD goes that route too.

 

[Cygnus x-1]

This says optional, for enthusiast and overclockers, so not sure what comes with it. http://www.frostytech.com/permalink.cfm?NewsID=92895 If I had to guess, maybe somewhere around the h50 or h70 based on the looks of it.

 

[Haserath]

I sorta think Intel might go with another memory stack for a sort of "L4" cache in the package - seem to recall somewhere that Intel is concerned with cache thrashing and they took some steps with IB to minimize it. But yeah 16 enhanced EU's and 1GB dedicated GPU memory would be a, uh - what's that term BM is fond of?? Oh yeah - MONSTER!

That would be quite the monster, but with how good the predictors are already, I don't know if it would be that great of a benefit unless they really drop latency to barely above L3 cache levels.

Well I went through some calculations over on the BD thread (probably got deleted however), starting with the 4-cour PII at 258mm^2 at 45nm, 6MB of L3. Going to 32nm should reduce the PII to about half that size, say 130mm^2. Going from 4 full cores to 8 CMT cores should add either 5% or 12% extra die space, depending on who you last talk to . Assuming that's per CMT core, thats either 20% or 50% more area, which brings us up to 195mm^2 using the 12% figure. Then accounting for the extra 2 MB of L3 should add maybe 10% more (a guess from the die shot over on S/A), the extra HT links and such, and I come up with around 250mm^2 as a rough estimate. Dunno what the Interlagos pricing is going to be, but I'd bet it ain't cheap - it is going to cost AMD a lot to have GF fab it if indeed the yields are low.

But they've increased the pipeline length and increased the L2 cache to double the amount as well.

It's funny how some of the fanbois over on the BD thread mention that just because AMD has apparently lowered the price of the 8-core by nearly $100 even before the launch, it's not because of performance issues, it's because AMD wants to undercut Intel and take away market share. Well Hector Ruiz tried that tactic some years ago and got surprised when C2D came out at a relatively low price compared to K8, and tons more performance. AMD promptly started hemorrhaging cash, while Intel still made a hefty profit, probably due in no small part to more efficent and cost-effective 65nm process. If AMD were to try and reverse this with BD, I think they'd just get themselves back into the red as Intel can afford to cut prices far more than AMD, again in part due to superior process (yields & maturity on 32nm anyway).

Anyway, I guess we'll see BD benchies some time next month so from those plus AMD's quarterly report, we should be able to figure out some facts and dispell some rumors and hype on both sides ..

SB is ~2/3 the size of BD and the yields are most likely in the 90% range at this point. Intel could probably price Sandy Bridge at half the price of AMD's BD and still make a profit on the desktop parts.

The question is whether BD is going to undercut SB with superior performance at a lower price. This would be a terrible move for AMD, because Intel definitely has the cash and yields to destroy AMD at any point unless Bulldozer is even better than C2D v. K8. That might not even make Intel cut prices; AMD wouldn't be able to produce enough chips to supply everyone, but AMD would lose tons of money for not pricing BD to where Intel's extreme processors are. The extreme market is small, but with yields the way they are, it wouldn't really matter since AMD could barely supply that market along with the server market.

 

[coyote2]

The question is whether BD is going to undercut SB with superior performance at a lower price. This would be a terrible move for AMD

If AMD could really do that, wouldn't it work out great for them to be out-competing Intel? And if the demand and profit were there (from this extremely hypothetical scenario) for AMD, couldn't and wouldn't they want to ramp up production accordingly?

, because Intel definitely has the cash and yields to destroy AMD at any point unless Bulldozer is even better than C2D v. K8. That might not even make Intel cut prices; AMD wouldn't be able to produce enough chips to supply everyone, but AMD would lose tons of money for not pricing BD to where Intel's extreme processors are. The extreme market is small, but with yields the way they are, it wouldn't really matter since AMD could barely supply that market along with the server market.

 

[Haserath]

If AMD could really do that, wouldn't it work out great for them to be out-competing Intel? And if the demand and profit were there (from this extremely hypothetical scenario) for AMD, couldn't and wouldn't they want to ramp up production accordingly?

It takes a long time and a ton of money(money that AMD doesn't have) to get a fab up and running to produce chips. AMD didn't know Bulldozer's true performance until B2 silicon which probably isn't even 100% of the performance they could get out of the silicon with rumors going around saying a new stepping will increase integer performance significantly.

AMD doesn't even have their own fabs anymore, so they have to rely solely on GloFo for their chips. AMD is not in a good position even if they did come out with an amazing design.

 

[FALC0N]

Update - check out AMD Fires Rick Bergman Under Cloud of Mystery :

Rick Bergman, Senior Vice President and General Manager at AMD has been fired according to HardOCP sources. AMD has given no reason as to why the well seasoned Bergman was leaving his position. It was told to us that Bergman had interviewed for the just-filled CEO position months back at AMD and had been turned down. Since then we have been lead to understand that AMD has lowered the boom on Bergman for not managing the GlobalFoundries relationship properly. Sources tell us that GlobalFoundries is simply not up to the task of supplying AMD its needed parts and Bergman is first in line when it comes to the responsibility of making sure AMD is sourced properly. We understand that Bergman will be the first of the dominoes to fall.

If true, then that would be some pretty hard evidence of the growing rift between AMD and GF..

Maybe, but this could be about mismanagement.

Perhaps he simply didn't diversify AMD's manufacturing outsourcing enough and thats what led to the problem. Part of the advantage of the companies separating was that AMD could use any fab they chose.

Just because Global Foundries was having issues meeting AMD needs, doesn't mean the relationship has deeper problems.

 

[FALC0N]

It takes a long time and a ton of money(money that AMD doesn't have) to get a fab up and running to produce chips. AMD didn't know Bulldozer's true performance until B2 silicon which probably isn't even 100% of the performance they could get out of the silicon with rumors going around saying a new stepping will increase integer performance significantly.

AMD doesn't even have their own fabs anymore, so they have to rely solely on GloFo for their chips. AMD is not in a good position even if they did come out with an amazing design.

They are not lock into GF. They can use any fab they wish to contract with.

 

[4745454b]

While they can in theory and ignoring any contracts they might have, how many chip makers do you know of who have a 28or 32nm SOL production facility? It's not like you can just fab it on something else.

 

[Haserath]

While they can in theory and ignoring any contracts they might have, how many chip makers do you know of who have a 28or 32nm SOL production facility? It's not like you can just fab it on something else.

Exactly, and GloFo won't want to share their secrets with any other fab. They're the only ones that can make 32nm SOI, so AMD is locked into their relationship with GF.

 

[ghnader hsmithot]

Forget about SB-E we have a new cherry picked cpu that is the i7 2700k
http://www.techspot.com/news/45590-intels-core-i7-2700k-arrives-next-month-for-330-350.html