Will Bulldozer eventually transcend Intel CPU's?

[teletran8]

Try and have an open mind.

Bulldozer's single core IPC is currently @ a 28-33 percent deficit to SB single core IPC depending on the SB clock-speed. According to this benchmark done by Anandtech in CB 11.5.

BD also appears to be currently slower in single threading (less than 10 percent) when it comes to a comparison to its older architecture of the later higher end Phenom II generation.

And in Multi-Threading SB shares a 12-13Percent speed advantage currently with the 2600k.

I don't believe these are as huge as deficits as people on these forums complain on and on about. In fact I think they are minor and will be solved shortly. Now getting to my points. SB came out in February, BD came out in Oct this year. They are both the leading cutting edge technology from each respective company. They will both have a 22nm die shrink when the time comes in Ivy Bridge and BD3 Steamroller.

When I look at Intels gameplan of i7's always being Quad Core at best with Quad Hyperthreading Cores to equal 8 Cores, and AMD now using 4 Compute Units/Modules to equal it's 8 Cores. I still think AMD's architecture has far more room to grow where Intel can't really innovate until they change their architecture (Hyperthreading seems a waste of an architecture design compared to AMD's Compute Units/Modules) and their 3D transistors basically provide better power consumption.

Intel is on their 2nd gen i7's currently, but their cores and Hyperthreading remain the same. Seems to me they are relying on clock-speeds instead of any architecture improvements. People think the 3D transistor will change the game, maybe it will and maybe it will solely lower power consumption, which seems to be the case it will be sensitive to power states and shut off more unused transistors than in the past.

Where as AMD's Bulldozer can make advancements in so many areas, with their L2 cache (latency/balancing/fetching) Picks an invalid way before picking a valid way alleged problems?, MSR/APC registers optimizing, OS support, Interrupt Delivery/Handling, etc. This is a big part of AMD's future (fine tuning this CPU) I think the yields were low and now that AMD has gotten them up and running out the door they will continue to fine tune these chips, and be experimenting alot more. I see them exploiting this chips design architecture in amazing ways in the future. I see Intel shrinking and saving power lol! In fact I see Intel stealing AMD's design shortly.

Does anyone else think we might ever see a Windows OS that is optimized for AMD CPU's?

But the main question I have for you fellow techies is which CPU will transcend further in that time Intels or AMD's and why you think so? (I know the future is unwritten, but this is fun and I think the CPU wars are finally starting)

I personally see BD being optimized in this 1st/2nd generation and then blowing up into a mega beast CPU by 22nm if not before this point in time. I even see the BD3 Steamrolller being just that, it is when AMD will pick up Steam and blow Intel CPU's away for good

Let's try and keep this thread civil and intelligent guys.

 

[mzs_biteme]

...I even see the BD3 Steamrolller being just that, it is when AMD will pick up Steam and blow Intel CPU's away for good....


If this quote was at the beginning of your post, I would have stopped reading there... Really, I'm no Intel "fanboiiii", but it stands to reason that Intel WILL NOT be caught with their pants down, like they had been in P4 days... They have proven this quite well so far.....

As to BD itself, I do recognize the new architecture being very flexible and "mold-able" in the future re-releases. Currently, the biggest beef I have with BD, is ridicules power envelope... That's one of the reasons I never touched P4 Presshot, and rode it out with Northwood until Core arch came out...

As to software optimizations, that remains to be seen. It's doable, and AMD should pursue OS and App creators to fully utilize BD's capacity, but they should really be concentrating on the CPU design itself.

One thing that I'd love to see (if at all possible) is some kind of advanced CPU "controller" app, where through software you can assign what/how many core(s) you want to assign to any given app, i.e. game, video editing app, rendering software, etc... That's would be the $#~iT....

 

[Mat3]

AMD has had a long time to fix problems in BD with Piledriver, but all they expect is a 3-5% increase in IPC by increasing tables structure sizes and the like. They obviously have no intention to fix the cache problems (or whatever it is hampering performance) any time soon, probably never.

Considering the following:
The ex-AMD's comments that a lot of the talented, skilled engineers left several years ago.
Charlie at Semiaccurate saying talented people have been leaving recently.
Evidence of the above with their K8 architecture which stagnated for five years (only minor, simple improvements to the core).
Other high profile people leaving for whatever reason (Myer, Bergman)
BD the way it is now, what we're hearing about Piledriver.

I think it's true; AMD CPU engineers are not good enough to provide anything better than a 'decent' CPU at best. All they know how to do is throw more cores at the problem, and not even in an efficient way anymore considering the die size of BD.

It is very dificulte to be more optimistic than that.

 

[jeremyshaw]

...Let's try and keep this thread civil and intelligent guys. ...

good luck. If it's civil and intelligent... than this thread is dead Nobody stays for good discourse anymore.

however, saying HT is a waste of Intel's Core arch is about as useful as saying the i7 2600k isn't faster than the i5 2500k. It's there because Intel's CPU is fast enough that merely reordering execution instructions (out of order execution) isn't merely enough, and branch prediction isn't enough. So easy way? Execute an entire seperate thread on the CPU's downtime between waiting for RAM latency. Finding a balancing act is the tough part, but I think Intel has got it, now.

I have no clue what is wrong with bulldozer's module arch, as it should be faster, but some indications are pointing at the L1 cache size/latency, in addition to L3 being exclusive cache, though the L2 seems to be prefect where it's at (shared between two cores).

I think BD can shine. I do think, despite the longer pipeline, AMD can catch up to Intel and present something similar to Thuban vs Nehalem/Lynnfield. But for AMD to be faster outright with a longer pipeline and low clocks... tough call.


IMO, of course.

 

[TheBlueChanell]

I think piledriver may (hopefully) bring more improvements then just 10-15%. I feel maybe they are just being a bit modest so they don't have another BD-type fiasco on their hands.

Expect less and give more and you make people happy.

Expect more and give less and you make people angry.

 

[sirmo]

I don't believe these are as huge as deficits as people on these forums complain on and on about.

You provided two best case scenarios with no regard to power consumption.

It took AMD twice as many transistors to deliver a product that has less IPC than their previous product and is behind Intel in everything. The longer pipeline doesn't even offer much in terms of overclocking headroom because of such an inefficient design.

As a result you have a design that doesn't scale. Intel's architecture has so much more room to grow.

Difference between AMD's Bulldozer and Intel's Sandy Bridge is like a difference between Ford's 400m v8 engine from the 70s and nissan's V6 gtr engine.

I don't even know if it's possible for AMD to trim the fat enough to even compete at the same fab process, let alone being stuck a fab process behind once Ivy B. is out.

 

[defaultluser]

Try and have an open mind.

Bulldozer's single core IPC is currently @ a 28-33 percent deficit to SB single core IPC depending on the SB clock-speed.

Absolutely NOT TRUE. Your entire thread is based on an incorrect calculation. Allow me to do the math for you.

Let's compare the i5 2400 and the FX-8150 single-threaded performance:

i5 2400 max turbo = 3.4 GHz
FX-8150 max turbo = 4.2 GHz.

We can safely assume that in a single-threaded bench, these are the speeds of both processors. So, scale the performance of the 8150 from 4.2 GHZ down to 3.4 GHz:

2400 score: 1.37
FX-8150 score: 1.02

FX-8150 score scaled to 3.4 GHZ = (3.4/4.2) * 1.02 = 0.78. Could be slightly higher, but this is a good way to roughly scale things.

So, at approximately the same clock speeds (3.4 GHz):
2400 score: 1.37
FX-8150 score: 0.78

This means the relative IPC advantage of Sandy Bridge is 1.37/0.78 * 100 =

%175 the performance of Bulldozer.

Or if you want to look at it the other way, that's a %45 performance DEFICIT for Bulldozer.

Let's try and keep this thread civil and intelligent guys.

Okay, I added the intelligence. Now you can keep it civil and accept that you made a mistake.

 

[mzs_biteme]

This "cements" my POV with regards to BD power issues... The day they can make a BD anywhere close to the efficiency of SB, they'll have my attention (read $$$upport..)

 

[Dan_D]

I think piledriver may (hopefully) bring more improvements then just 10-15%. I feel maybe they are just being a bit modest so they don't have another BD-type fiasco on their hands.

Expect less and give more and you make people happy.

Expect more and give less and you make people angry.

You won't get increases like that or greater than that with incremental adjustments to the architecture. At best we'll see the same type of improvement we saw going from Phenom to Phenom II. Expecting more than that is really just wishful thinking.

 

[fps4ever]

You won't get increases like that or greater than that with incremental adjustments to the architecture. At best we'll see the same type of improvement we saw going from Phenom to Phenom II. Expecting more than that is really just wishful thinking.

If they can get a modest 10% IPC increase AND get the power usage down to more manageable levels then they should be OK....that's A BIG IF though?

 

[drescherjm]

If they can get a modest 10% IPC increase AND get the power usage down to more manageable levels then they should be OK....that's A BIG IF though?

If that happens I believe this would be very competitive to Intels lga1155 SB and even IB offerings. Although IB 1155 will certainly use less power (again unless Intel lied about its 22nm process).

 

[TheBlueChanell]

You won't get increases like that or greater than that with incremental adjustments to the architecture. At best we'll see the same type of improvement we saw going from Phenom to Phenom II. Expecting more than that is really just wishful thinking.

Generally I'd agree but this is a brand new arch on a brand new process so who knows whats in store. Maybe there's something that is crippling performance and if they manage ti find it and squash it we'll see some nice gains.

This could all be mathmetically impossible, I wouldn't know as I'm not micro-processor engineer.


We'll see what happens, and tbh I don't really care what happens as long AMD stays at least somewhat competitive. I'll go wherever I get the most bang for my dollar.

 

[Dan_D]

Generally I'd agree but this is a brand new arch on a brand new process so who knows whats in store. Maybe there's something that is crippling performance and if they manage ti find it and squash it we'll see some nice gains.

This could all be mathmetically impossible, I wouldn't know as I'm not micro-processor engineer.


We'll see what happens, and tbh I don't really care what happens as long AMD stays at least somewhat competitive. I'll go wherever I get the most bang for my dollar.

If they were able to get that kind of performance increase in a single incremental upgrade / improvement of an existing architecture, it will be the first time we've seen that in the history of desktop processors. These types of improvements have only ever occurred with the introduction of a new architecture. Even then such increases are rare. Incremental updates to an architecture typically produce a 3-7% improvement in IPC or even any type of general performance at best. Now we could see that type of improvement over the total lifespan of the architecture assuming it's refreshed more than once or twice. By the time the next scheduled refresh comes out? Doubtful. Again it's never happened in the last 20 years. Incremental updates are just that. You get small increases in efficiency and performance. It takes a revolutionary change to the architecture, or an entirely new architecture to see < 15% gains in performance.

 

[fps4ever]

Dan_D your killing my buzz..

 

[Dan_D]

Dan_D your killing my buzz..

People expect huge things out of the refresh and end up disappointed. Be realistic about what we are likely to see with the next iteration of Bulldozer now and avoid a second round of disappointment.

Look back at history. We saw no hope of Northwood catching up to the Athlon 64. Intel fan boys even talked about Intel switching Northwood to an integrated memory controller and keeping socket 478. (What the hell were they smoking?) They had high hopes of a 64bit Prescott core evening the odds with AMD. Prescott came out with EM64T disabled in hardware and it was actually worse with regard to IPC than Northwood. People hoped that Intel's x86-64 / EM64T would make them out perform AMD or at least close the gap and yet their performance was identical with 32bit and 64bit code. They had high hopes for Smithfield, and again, it didn't happen. AMD fan boys raved about what Phenom II would bring to the table over Phenom I and that it's architecture was really long legged and would scale better as time went on. That architecture had one mildly decent refresh and it damned sure wasn't a 10-15% performance improvement over the original. In the last 20 years we never saw massive increases in performance so long as the architecture remained largely unchanged. Incremental updates have never brought that much to the table. Only new architectures can do that. Bulldozer had it's shot and failed to deliver. Just like Phenom and Phenom II before it. The next iteration of Bulldozer will be more compelling but again, keep your expectations reasonable or you'll eat a second round of crow and suffer even more disappointment.

 

[MrMike]

What process is this refresh supposed to be on? Does it even stand a chance of getting close to Llano on power usage? A 10% decrease in power usage would barely be equal to a 45nm Thuban.

As was already mentioned by Dan, refreshes don't result in a large increase in IPC. As such 22nm Ivy isn't likely to be bringing much if any improvement to IPC, but a decent drop in power usage and a higher max overclock is likely. I don't think it's absurd to suggest that a quad Ivy could use literally half the power of an 8-core Bulldozer, cost the same price and be faster.

 

[xBanzai89]

32nm.

 

[mesyn191]

I think piledriver may (hopefully) bring more improvements then just 10-15%. I feel maybe they are just being a bit modest so they don't have another BD-type fiasco on their hands.

You're not going to get more than 10% improvements in performance with very minor incremental updates to an existing arch. Even AMD's own slides on future CPU's suggest modest gains for at least the next 2 revisions well off into the future.

 

[fps4ever]

I heard a lot of talk about the clocks didn't reach as high as expected,(not mentioning the power envelope on purpose) but the 8150 reached 4.2GHZ on Turbo? Exactly how much higher can you expect those processors to go for performance on 32nm and still be sold to consumers with standard cooling?

 

[mesyn191]

What process is this refresh supposed to be on?

32nm, GF's 22nm process isn't ready yet for x86 and as far as anyone knows still has major yield issues which is why AMD's 7xxx and nV's new GPU's are going to be late.

Does it even stand a chance of getting close to Llano on power usage?

Possible but unlikely. BD's heat problems are inherit to it being a "speed demon" arch. on a crappy process while also being huge. Minor refreshes almost never fix any of that.

As such 22nm Ivy isn't likely to be bringing much if any improvement to IPC,

Given the leaks IB will be around 10% faster than SB but will draw lots less power, OC better, and have a much better on die GPU. So even if PD improves performance 10% or so, which is what even AMD has suggested it will, comparatively PD will still be a terrible buy.

I don't think it's absurd to suggest that a quad Ivy could use literally half the power of an 8-core Bulldozer, cost the same price and be faster.

Yea. AMD is screwed. SB is already too good, IB will improve on that even more.

 

[mesyn191]

I heard a lot of talk about the clocks didn't reach as high as expected,(not mentioning the power envelope on purpose) but the 8150 reached 4.2GHZ on Turbo? Exactly how much higher can you expect those processors to go for performance on 32nm and still be sold to consumers with standard cooling?

They were hoping for 4Ghz non turbo on air reportedly, 5Ghz on air was supposed to be doable too. They must've known for at least the better part of a year that wasn't going to happen though when they got the first working samples back from fab around Feb. or March.

 

[mesyn191]

If they can get a modest 10% IPC increase AND get the power usage down to more manageable levels then they should be OK....that's A BIG IF though?

No they're still screwed. 10% more performance for BD still puts it behind Nehalem Core i5's and i7's which are quite old now.

 

[mrbigshot]

When I look at Intels gameplan of i7's always being Quad Core at best with Quad Hyperthreading Cores to equal 8 Cores, and AMD now using 4 Compute Units/Modules to equal it's 8 Cores.

current i 7's are 6 core's and the sb-e's are 8 cores with 2 disabled to pit in the 130w tdp limit of current desktop boards. you can buy 8 core xeon's if you really want. were not seeing 8 cores on desktops because there is very little need for it.

I still think AMD's architecture has far more room to grow where Intel can't really innovate until they change their architecture (Hyperthreading seems a waste of an architecture design compared to AMD's Compute Units/Modules) and their 3D transistors basically provide better power consumption.

amds design can be reifined but cannot be changed without a complete overhall of the core design. the next overhaul for intel will be haswell in 2013.

Intel is on their 2nd gen i7's currently, but their cores and Hyperthreading remain the same. Seems to me they are relying on clock-speeds instead of any architecture improvements.

sandy bridge was far more than a speedbump. it was a refresh that brought 15% more performance using the same basic core design of the previus i7's.

Where as AMD's Bulldozer can make advancements in so many areas, with their L2 cache (latency/balancing/fetching) Picks an invalid way before picking a valid way alleged problems?

this has already been corrected with windows 8 and it has been proven to not add much performance. the problem is the basic design. you have a single lane that must feed two cores.

Does anyone else think we might ever see a Windows OS that is optimized for AMD CPU's?

we already have windows 8 that does it correctly

But the main question I have for you fellow techies is which CPU will transcend further in that time Intels or AMD's and why you think so? (I know the future is unwritten, but this is fun and I think the CPU wars are finally starting)

hard to say, bulldozer still has a ways to go to equel the prior generations performance. very little can be changed the way it sits without a redesign. intel will be on a smaller more power efficient platform very shortly and they already can match physical core counts. ivy bridge isnt supposed to be much faster than sandy bridge, just cheaper to produce and use less power. haswell in 2013 remains to be seen.

I personally see BD being optimized in this 1st/2nd generation and then blowing up into a mega beast CPU by 22nm if not before this point in time. I even see the BD3 Steamrolller being just that, it is when AMD will pick up Steam and blow Intel CPU's away for good

by the time amd gets it right they will be competing with haswell, not there current sandy bridge or closer to nethleham like they are now. when the core 2 duo was launched it was the nail in the coffin. amd has been clawing at the lid for years now but i dont see them digging them self out for a long long time.

 

[drescherjm]

No they're still screwed. 10% more performance for BD still puts it behind Nehalem Core i5's and i7's which are quite old now.

The more important part of that was power usage under control. If that happens and allows 4GHz+ stock clocks BD will do fine even though it has lower IPC.

 

[drescherjm]

They were hoping for 4Ghz non turbo on air reportedly, 5Ghz on air was supposed to be doable too. They must've known for at least the better part of a year that wasn't going to happen though when they got the first working samples back from fab around Feb. or March.

I think they expected a bit too much with GF's brand new 32nm process.

 

[fps4ever]

I think they expected a bit too much with GF's brand new 32nm process.

I don't think even Intel could get that on their 32nm process for non-turbo chips?

 

[MrMike]

A refresh on 32nm is irrelevant. This will be less than the improvement from 65nm Phenom > 45nm Phenom II, and they would need way more improvement than that.

Forget BD, make some 6-core Llanos with two other cores where the GPU would be and go back to working on the successor to Brazos.

 

[griff30]

Will Bulldozer eventually transcend Intel CPU's??

Not by a damn sight.

 

[fps4ever]

Will Bulldozer eventually transcend Intel CPU's??

Not by a damn sight.

Did Bulldozer piss you off personally or what?

 

[Weeth]

It's the dawning of the 23nd century. Hyperlightdrive starships are ferrying the first colonists to the Gliese planets. The first centennial of global peace is being celebrated. In other news, GloFo is still forecasting that their first 22nm silicon will be out next year...

 

[KillerFry]

It's the dawning of the 23nd century. Hyperlightdrive starships are ferrying the first colonists to the Gliese planets. The first centennial of global peace is being celebrated. In other news, GloFo is still forecasting that their first 22nm silicon will be out next year...

But will it blend?

 

[drescherjm]

I don't think even Intel could get that on their 32nm process for non-turbo chips?

Not for 8 core + HT. Possibly for lga1155 4 core + HT but that would compete too much with SB-E. For dual core + HT they have been selling a 4.4GHz stock xeon for months.

 

[Weeth]

But will it blend?

The 23rd century Blendtec features a fusion motor with dilithium blades which can blend diamonds, but even it cannot blend any sense into the GloFo fab execs.

Let's face it, GloFo makes RIM, Netflix, and HP all look like supremely well run companies. It's an unmitigated disaster of a fab that should be repurposed as a coffee grinding factory.

 

[mzs_biteme]

I don't think even Intel could get that on their 32nm process for non-turbo chips?

Meh..... 32nm - old school... This is where it's at!!!! 22nm, IB to have 77W TDP !!!!! Let's see GloFo try that in the next ohhhhh, 2~3yrs....

 

[-Dragon-]

One thing that I'd love to see (if at all possible) is some kind of advanced CPU "controller" app, where through software you can assign what/how many core(s) you want to assign to any given app, i.e. game, video editing app, rendering software, etc... That's would be the $#~iT....

It's called "Task Manager"

 

[KazeoHin]

It's called "Task Manager"

+1

Didn't Tom's have a special background ap that could automatically assign tasks affinity as per user choice? Its great for working with the Unreal Dev Kit, because that damn program hogs at least 25% of each core, causing other real-time programs to chug.

 

[soulsavior]

I think the scary thing for AMD (even moreso than architecture) is how far Intel is getting in front of them in fabrication abilities. Looking at the launch (and all its delays), it appears GF wasn't able to create any kind of supply at all. The 8150 were darn near non-existent even on day one. The 8120's sold out shortly thereafter. I haven't even seen a 4100 online. There were all those reports about problems with Llano yields. They don't seem to be able to keep up at all.

I suppose you could make the case that high demand caused the shortage, but that seems like a bit of a stretch. The fact that the 4100 basically doesn't exist yet tells me they're so short on the 8 cores they can't afford to spare any time or materials on 4100s yet.

Regardless of what they do on the architecture, they're always gonna be playing from behind if Intel keeps staying two steps ahead of them in fabrication.

 

[Forceman]

That's been AMD's problem for ages - even when they had the better processor they never had enough production capacity.

 

[mzs_biteme]

It's called "Task Manager"

Assigning priority/affinity in Task Manger is one thing, having a nice GUI app with a list of apps without chasing after process executables... Gimme that!

 

[mesyn191]

That's been AMD's problem for ages - even when they had the better processor they never had enough production capacity.

Yuuuuuuuuup.

Even when AMD was firing on all cylinders so to speak they could never get much more than 30% market share simply because they didn't have the fab capacity. That has been one of the biggest problems with the CPU biz trending towards multi core for AMD since that has eaten up any possible chance AMD ever had to get a real leg up on Intel. They simply couldn't bring up enough fab space fast enough to ever really be a strong competitor financially.