Rumour: Bulldozer 50% Faster than Core i7 and Phenom II.

[chihlidog]

I gotta say, it's really great to see so much anticipation for an AMD launch after them being second best for so many years. That's probably exactly what they want. I hope they deliver.

 

[OCGuy]

Seen a lot of speculation on hardware sites about how much performance competition BD will give Sandybridge. I think that's just enthusiasts getting caught up in their personal interests. If AMD can provide an 8 integer core equivalent of even the i7-760 somewhere in the 200-300 range it will be my next main PC. If they can't do that then I'll probably OC a 2500K.

I'm pretty confident this isn't going to be another Phenom 1 launch for AMD. Although I'm not very optimistic on the OC headroom of this first round of 32nm product considering the delays. Hope they surprise me.

I think you are much closer than the hopeful thinking that is going on by some people. Somehow they expect AMD to go from < C2D ---> SB, skipping more than 2 years, out of the blue.

I think the most ironic part about it is that because of the high expectations the AMD faithful are putting on BD, anything less than equal or better than SB will be considered a failure.

I almost feel sorry for AMD at this point...

 

[Cerb]

I'm finding myself wanting to QFT OCGuy quite a bit, lately. If they can match Nehalem (+10&#37;, say), offer good efficiency, and truly compete in threaded scenarios (recall that even Nehalem could lose to STARS here and there), I'll be quite content, whether I choose AMD, Intel, or the waiting game.

If they can match SB on the desktop, I'll be struck dumb for a week.

AMD doesn't need to best Intel; BD needs to be good enough to sell enough chips at high enough prices for their CPU division to actually make money. On our end, that success will translate into midrange price competition, and possibly some server CPU price competition.

 

[podspi]

Ya, I think you're right.

Makes sense. They only have a small window to be competitive in terms of performance. I think the idea is that they will start out balls to the wall making 8-core parts to try and get as many high end parts as possible, and then crippling the rest as far as they need to be crippled to get them to work properly. I bet the initial run of quad cores will actually be crippled eights. Once they get the yields up and (are forced) to compete in the budget market, they will start manufacturing 4-cores and have good enough yields to sell them whole.

I am hoping those "quads" will be unlockable and affordable .

I need a quad-core CPU ~ 6 months, but that need could be filled with my Thuban if I get lucky...

I think you are much closer than the hopeful thinking that is going on by some people. Somehow they expect AMD to go from < C2D ---> SB, skipping more than 2 years, out of the blue.

I think the most ironic part about it is that because of the high expectations the AMD faithful are putting on BD, anything less than equal or better than SB will be considered a failure.

I almost feel sorry for AMD at this point...

I think you have to keep in mind that AMD doesn't have to start from where they left off. That's like saying nVidia can't produce a high-performance ARM core because they haven't made the equivalent of an 8086 yet...

I'm sure AMD subscribes to all the latest EE and semi journals Not to mention that I bet their internal research is far ahead of what they're shipping. IMHO the reason K10 is still a product AMD is shipping is because of execution issues. Sad, but true. Heck, if 32nm hadn't been delayed K10's last hurrah(Llano) would have been out how many months ago?

Anyway, my point is that while I wouldn't necessarily expect AMD to entirely leapfrog the industry (though they really have had their home-runs in the past) it isn't a stretch for them to leapfrog themselves.

Remember, everything is easy when you know how it is done and have a pretty comprehensive cross-licensing agreement :sneaky:

 

[Cogman]

I am hoping those "quads" will be unlockable and affordable .

I need a quad-core CPU ~ 6 months, but that need could be filled with my Thuban if I get lucky...



I think you have to keep in mind that AMD doesn't have to start from where they left off. That's like saying nVidia can't produce a high-performance ARM core because they haven't made the equivalent of an 8086 yet...

I'm sure AMD subscribes to all the latest EE and semi journals Not to mention that I bet their internal research is far ahead of what they're shipping. IMHO the reason K10 is still a product AMD is shipping is because of execution issues. Sad, but true. Heck, if 32nm hadn't been delayed K10's last hurrah(Llano) would have been out how many months ago?

Anyway, my point is that while I wouldn't necessarily expect AMD to entirely leapfrog the industry (though they really have had their home-runs in the past) it isn't a stretch for them to leapfrog themselves.

Remember, everything is easy when you know how it is done and have a pretty comprehensive cross-licensing agreement :sneaky:

The last time I remember AMD getting this much hype was the original Phenom... While it was a decent boost from the previous generation, it was nothing like it was expected to be.

AMD is in a crappy position, they simply don't have the R&D team like intel does. Intel may be able to do a generational Jump (P4->C2D) However, I can't see AMD doing the same.

Leap frogs are very rare things and they generally aren't cheap. AMD is changing the design quite a bit, so who knows, but I'm not holding my breath.

 

[drizek]

The original Phenom was terribly executed by AMD (although Intel managed to execute it pretty well).

Phenom II on the other hand was a great product. Sure, it couldn't keep up with Nehalem, but it was at least competitive with Core 2 Duo, and could scale to 6 and 12 cores, keeping AMD at least somewhat competitive in certain niche markets.

Bulldozer design and Phenom II execution could mean that AMD comes away from this doing pretty well. If not, well at least I can come away from this with a really nice octalcore CPU for around $150.

 

[Cogman]

The original Phenom was terribly executed by AMD (although Intel managed to execute it pretty well).

Phenom II on the other hand was a great product. Sure, it couldn't keep up with Nehalem, but it was at least competitive with Core 2 Duo, and could scale to 6 and 12 cores, keeping AMD at least somewhat competitive in certain niche markets.

Bulldozer design and Phenom II execution could mean that AMD comes away from this doing pretty well. If not, well at least I can come away from this with a really nice octalcore CPU for around $150.

The original Phenom suffered from being rushed through the gates. AMD was ahead at the time and had no expectations for Intel to pull the C2D out of its hat.

Everything pointed to them saying "Crap... we need to get something out there, and fast." Hence the infamous TLB bugs and general slowness. It just wasn't a polished product. Phenom II really seemed to be what AMD wanted Phenom to be. Though, it was too late. By the time the Phenom IIs came out, Phenom had left a pretty bad taste in many peoples mouths.

AMD grew too big, that is why intel came back and bit them hard.

 

[smartpatrol]

I think the most ironic part about it is that because of the high expectations the AMD faithful are putting on BD, anything less than equal or better than SB will be considered a failure.

I almost feel sorry for AMD at this point...

AMD fans certainly have a lot of faith. I still remember when Core 2 Duo came out around the same time as AM2. . . AMD fans kept insisting that it was an unfair comparison and we had to wait until the mythical "K8L" came out. (Remember, the native quad-core that did "reverse hyperthreading" . . . and never actually existed.)

It's like they are still waiting today. I would love to see a return to the days when AMD was kicking Intel's ass, but realistically, it's just not going to happen.

 

[AtenRa]

You don&#8217;t have to be a fan to understand CPU architecture and translate all the information we have until now in to performance.

Bulldozer (4 module 8-Core) will be faster than Intel SB (4-core 8-threads) in multithread/multitask simple as that.

 

[darkswordsman17]

I think you are much closer than the hopeful thinking that is going on by some people. Somehow they expect AMD to go from < C2D ---> SB, skipping more than 2 years, out of the blue.

I think the most ironic part about it is that because of the high expectations the AMD faithful are putting on BD, anything less than equal or better than SB will be considered a failure.

I almost feel sorry for AMD at this point...

I guess you're forgetting Intel basically did just what you're describing when they went from Prescott and Pentium D to Core 2.

So glad you decided to add nothing to the thread though, other than calling people morons and AMD fanboys for expecting that an 8 core chip might be at least competitive with a 4 core/8thread chip.

Well, really, if Bulldozer doesn't match Sandy Bridge then yeah, I'd say its a failure, as that would mean they couldn't improve things with a drastic redesign, doubling up on actual cores, and on a smaller process.

 

[HW2050Plus]

You don’t have to be a fan to understand CPU architecture and translate all the information we have until now in to performance.

Bulldozer (4 module 8-Core) will be faster than Intel SB (4-core 8-threads) in multithread/multitask simple as that.

That is out of question and not the problem. The problem arises from the amount of die space AMD Bulldozer needs to surpass the 4 Core Sandy Bridge. That is almost twice the die space (excluding graphics on Intel) and that causes the real problem, not in 2011 but in the upcoming years. So let's hope that Bulldozer II will make a huge step forward in performance vs. die size utilization.

 

[Dresdenboy]

That is out of question and not the problem. The problem arises from the amount of die space AMD Bulldozer needs to surpass the 4 Core Sandy Bridge. That is almost twice the die space (excluding graphics on Intel) and that causes the real problem, not in 2011 but in the upcoming years. So let's hope that Bulldozer II will make a huge step forward in performance vs. die size utilization.

If you look at the die you'll see a lot of area between components which might be reduced to optimize die size.
http://pc.watch.impress.co.jp/img/pcw/docs/430/044/html/1.jpg.html

 

[Lepton87]

The last time I remember AMD getting this much hype was the original Phenom... While it was a decent boost from the previous generation, it was nothing like it was expected to be.

AMD is in a crappy position, they simply don't have the R&D team like intel does. Intel may be able to do a generational Jump (P4->C2D) However, I can't see AMD doing the same.

Leap frogs are very rare things and they generally aren't cheap. AMD is changing the design quite a bit, so who knows, but I'm not holding my breath.

The generational jump you're talking about never existed. You just can't compare p4 with c2d as C2D architecturally wasn't a successor to P4, it was merely a beefed up Core Duo.

See for yourself:
http://www.anandtech.com/show/2056/4

 

[Idontcare]

The original Phenom was terribly executed by AMD (although Intel managed to execute it pretty well).

If you look at what happened with the original 65nm Phenom, take into consideration the relatively minor architectural improvements made to it that resulted in Phenom II, and look what happened to the 65nm Athlon (over that of their 90nm predecessors)...IMO I have come to view the achilles heel as being none other than the 65nm process tech itself having failed to enable a competitive product lineup more so than the architectures.

Had AMD been more aggressive with the 65nm process, enabled better clockspeeds and better sram density (two things that PhII had for it over PhI), then I think Phenom would have generated a very different legacy for itself.

65nm vs. 90nm: AMD's Athlon 64 X2 5000+

Sure the TLB bug didn't help Phenom but that was mere 3 month stepping correction, but something was generally amiss with that 65nm node shrink and everything that was produced from it showed this in one area or another.

It was almost like 65nm was to AMD what 90nm was to Intel. I think Phenom would have done just fine at the time had the underlying 65nm process tech really differentiated itself from the existing 90nm process tech at the time as Intel's managed to do (Cedar Mill vs. Prescott).

 

[Cogman]

The generational jump you're talking about never existed. You just can't compare p4 with c2d as C2D architecturally wasn't a successor to P4, it was merely a beefed up Core Duo.

See for yourself:
http://www.anandtech.com/show/2056/4

If you look at the timeline, the Core Duo (Yonah) architecture was released in January of 2006, C2d was released in august of 2006. The comparison of the Cd and the C2d isn't a fair comparison of where the leap came into play. Dothan is a better comparison for the jump.

8 months isn't enough time to develop a new arch. Yonah was the warning before the storm. (The difference between Yonah and Dothan was huge).

 

[smartpatrol]

The generational jump you're talking about never existed. You just can't compare p4 with c2d as C2D architecturally wasn't a successor to P4, it was merely a beefed up Core Duo.

See for yourself:
http://www.anandtech.com/show/2056/4

? C2D was definitely a successor to P4. Of course you can compare them. C2D replaced P4 on the desktop. The original Core Duo was a dual-core Pentium M, and was basically a stop-gap product for laptops until C2D came out.

 

[AtenRa]

That is out of question and not the problem. The problem arises from the amount of die space AMD Bulldozer needs to surpass the 4 Core Sandy Bridge. That is almost twice the die space (excluding graphics on Intel) and that causes the real problem, not in 2011 but in the upcoming years. So let's hope that Bulldozer II will make a huge step forward in performance vs. die size utilization.

we have to wait to see the real die size, could be smaller than 300-280mm2.

 

[Lepton87]

? C2D was definitely a successor to P4. Of course you can compare them. C2D replaced P4 on the desktop. The original Core Duo was a dual-core Pentium M, and was basically a stop-gap product for laptops until C2D came out.

Architecturally C2D was a continuation of laptop line of processors that trace their origins back to P6 arch, it has nothing to do with Netburst, that's why there was such a huge improvement in performance. C2D was a successor to P4 only in marketing/product placement etc. Internally they are completely different.

 

[smartpatrol]

Architecturally C2D was a continuation of laptop line of processors that trace their origins back to P6 arch, it has nothing to do with Netburst, that's why there was such a huge improvement in performance. C2D was a successor to P4 only in marketing/product placement etc. Internally they are completely different.

I don't get what you're arguing. So was the P4 not a successor to the P3 because P4 is based on the Netburst archictecture? C2D was a successor to P4 in that it was what came after and replaced P4. You know, the literal definition of a successor.

 

[Cerb]

I don't get what you're arguing. So was the P4 not a successor to the P3 because P4 is based on the Netburst archictecture?

Precisely. Netburst was not an evolution. P6->Pentium-M->Core->Core 2 was.

You just can't compare p4 with c2d as C2D architecturally wasn't a successor to P4

Emphasis added.

 

[jimbo75]

The original Phenom was terribly executed by AMD (although Intel managed to execute it pretty well).

Well TLB bug aside, it was more to do with the 65nm being really quite bad in my opinion.

 

[formulav8]

It was the crappy 65nm process that really held phenom back. Look at what happened to the Athlon x2's. They got a subpar reduction in die size going from 90nm to 65nm despite officially having the same number of transistors and a increase in cache latency. Thats alittle screwy. 45nm definitely changed all of that. AMD's 45nm is top notch despite not having the super-duper features Intel's 45nm process has. :thumbsup:

 

[OCGuy]

Bulldozer (4 module 8-Core) will be faster than Intel SB (4-core 8-threads) in multithread/multitask simple as that.

So in a game that can use 4 cores, your prediction will still be true? Or are we going to have to take the flagship 8-core chip and benchmark programs that can use 8 "cores" in order to show BD being superior to SB?

What about your mythical "Average joe" walking into Best Buy who isn't going to consider anything over a quad-core. What is BD going to do for him compared to a SB quad?

It isn't the high-end 8 core BD or the SB-E 2011 6 core chips that are going to make or break quarterly profits for either company.

 

[Pontius Dilate]

You don’t have to be a fan to understand CPU architecture and translate all the information we have until now in to performance.

Bulldozer (4 module 8-Core) will be faster than Intel SB (4-core 8-threads) in multithread/multitask simple as that.

This is wildly general statement and we don't have nearly enough information about Bulldozer performance to substantiate it. The best we can say is that for certain tasks BD may be faster than SB, assuming BD clocks high enough. That's truly all we can say. We don't know for sure how BD will perform at single-threaded tasks in comparison to SB and at what relative clocks. We don't know how high AMD will be able to clock BD or what its thermal performance will be in comparison to SB. We don't know for sure at what kind of multithreaded tasks BD will perform especially well. We know SB performance and have a few unconfirmed BD benchmarks and speculations on introductory clocks. It's as simple as that.

 

[Phynaz]

This is wildly general statement and we don't have nearly enough information about Bulldozer performance to substantiate it. The best we can say is that for certain tasks BD may be faster than SB, assuming BD clocks high enough. That's truly all we can say. We don't know for sure how BD will perform at single-threaded tasks in comparison to SB and at what relative clocks. We don't know how high AMD will be able to clock BD or what its thermal performance will be in comparison to SB. We don't know for sure at what kind of multithreaded tasks BD will perform especially well. We know SB performance and have a few unconfirmed BD benchmarks and speculations on introductory clocks. It's as simple as that.

C'mon, he's comparing a CPU with twice the number of cores in apps that are multi-threaded. Pretty much impossible for it to lose, unless AMD really botched it badly - which I don't believe they did.

Now an 8 core BD against a 6 core SB, or 4 core against 4 core, that will be interesting.