Would AMD be better off if they had not developed Bulldozer?

[Red Hawk]

Developing a whole new architecture like Bulldozer must have taken a whole lot of money for research and development. Seeing as there is little performance improvement over even AMD's own previous architecture Stars (or K10...whatever you want to call it), would it have been better for AMD to focus on doing whatever tweaks they could to Stars and pouring the rest of their research money into Fusion?

 

[BLaber]

Developing a whole new architecture like Bulldozer must have taken a whole lot of money for research and development. Seeing as there is little performance improvement over even AMD's own previous architecture Stars (or K10...whatever you want to call it), would it have been better for AMD to focus on doing whatever tweaks they could to Stars and pouring the rest of their research money into Fusion?

Would it have been better if Intel had not Produced P4 , think how it could have effected current Intel architecture if P4 was not developed ?

 

[Ancalagon44]

Would it have been better if Intel had not Produced P4 , think how it could have effected current Intel architecture if P4 was not developed ?

Yes! They had to abandon it completely because it was an evolutionary dead end! Big waste of money.

Same with Bulldozer - the things they need to do to make it faster will pull it away from what is originally designed to be.

However, it may be too early to tell. As far as I know, it has been a flop on the desktop (duh) but done okay on the server market. Not wonderful, just okay. If Piledriver can significantly improve on it, and maybe if GlobalFoundries can sort out their manufacturing issues, it wont be so bad.

 

[Arzachel]

Yes! They had to abandon it completely because it was an evolutionary dead end! Big waste of money.

Same with Bulldozer - the things they need to do to make it faster will pull it away from what is originally designed to be.

However, it may be too early to tell. As far as I know, it has been a flop on the desktop (duh) but done okay on the server market. Not wonderful, just okay. If Piledriver can significantly improve on it, and maybe if GlobalFoundries can sort out their manufacturing issues, it wont be so bad.

Bulldozer is everything but a dead end. You have to take into account that the longterm goal for AMD is a total CPU and GPU hybrid, An efficient integer cruncher is just what they need. Even if the current implementation is less than stellar.

 

[Steelski]

No, I don't think it would have been better for them not to develop Bulldozer,
reasons why,
Taking into account,
leaky 32nm process,
not enough cach,
does very well in multithreaded software already,
scales way better than Sandy bridge,
little software compiled for it yet.

Think of the future,
improved IPC,
increased speed with better optimization 32nm and 22nm process,
software support,
more cores,
better thermal values,
linked with GCN FPU performance, on a 14nm process you can probably expect a fully functional 4 module (8 core) chip with quad channel DDR4, GCN FPU unit all running at 4GHZ with a graphics equivalent of a 6970 performance.
Or how about a 24 core chip with a serious FPU the size of 4 modules, CPU running at 5 GHZ with a near 35% IPC improvement on every core. With serious software optimizations now commonplace. And a lower power envelope.
thats all I'm asking for, nothing more.
All at the price of a $300 APU.

 

[bronxzv]

Developing a whole new architecture like Bulldozer must have taken a whole lot of money for research and development. Seeing as there is little performance improvement over even AMD's own previous architecture Stars (or K10...whatever you want to call it), would it have been better for AMD to focus on doing whatever tweaks they could to Stars and pouring the rest of their research money into Fusion?

to continue to pretend full x86 compatibility it wasn't possible to escape AVX support including forthcoming FMA3 in Piledriver, it wasn't possible to ensure this support without deeply changing the core

 

[Ancalagon44]

Bulldozer is everything but a dead end. You have to take into account that the longterm goal for AMD is a total CPU and GPU hybrid, An efficient integer cruncher is just what they need. Even if the current implementation is less than stellar.

Its not a dead end, but I guess the question is, would AMD have been better off using another architecture as the base for going forward?

I mean.. look at Intel and Core 2 Duo. In some ways it still is the basis of the modern Sandy Bridge CPUs, and it is awesome.

A total CPU and GPU hybrid is still a long ways off, and it also needs much better software support. Intel tried an x86 GPU remember, and they didnt too well on it.

I think AMD could have made a better CPU to use as the base going forward, than Bulldozer. I dont think it has enough oomph per core to be usable. It will have to be significantly changed to be competitive.

 

[Soulkeeper]

I have no doubt that bulldozer can be fixed.

Many of us are just frustrated with how slow AMD moves, how uncompetitive they seem compared to the past.
I doubt they spend nearly as high of a % of their income on R&D as they did with jerry sanders at the helm. They seem to be mostly treading water, trying to survive, and paying off debts the last few years.
Hopefully things will get better and we'll get the old AMD back

 

[nenforcer]

I have no doubt that bulldozer can be fixed.
Hopefully things will get better and we'll get the old AMD back

http://www.dailytech.com/AMD+Plunges+Back+in+the+Red+Posts+Gloomy+Outlook/article23861.htm

 

[Soulkeeper]

http://www.dailytech.com/AMD+Plunges+Back+in+the+Red+Posts+Gloomy+Outlook/article23861.htm

yeah, I read that. Hard to be impressed
but there is always hope right ?

 

[Dark Shroud]

Bulldozer is actually selling well enough, I still plan to buy one. The price vs performance makes them just fine for mid range and good deals for the low end.

People also seem to forget that MS has yet to re-release their scheduling patch for Win7 to properly use Bulldozer's new module design. MS had to do patch Windows for Intel's Hyper Threading with XP and it wasn't that noticeable until Vista was released.

 

[sm625]

Yeah it was a huge waste of time. They did not even attempt to place one or more of their GPU SIMDs into the FPU. Even if it would not be fully functional or useful today, they have to start the long process of getting DirectX code optimized to run graphics code within the cpu, and begin removing the overhead associated with having a separate gpu on a separate bus with long latencies inbetween. They have accomplished nothing to that end.

They also needed to improve integer IPC. Again they have not accomplished anything toward this end either.

Those are their two primary goals and they failed at both. I am quite shocked at how badly they have failed.

 

[Sweepr]

Bulldozer is actually selling well enough, I still plan to buy one. The price vs performance makes them just fine for mid range and good deals for the low end.

People also seem to forget that MS has yet to re-release their scheduling patch for Win7 to properly use Bulldozer's new module design. MS had to do patch Windows for Intel's Hyper Threading with XP and it wasn't that noticeable until Vista was released.

Here's some results with the patches: http://www.tomshardware.com/reviews/windows-7-hotfix-bulldozer-performance,3119.html

Minimal (if any) gains.

 

[Joseph F]

No, I don't think it would have been better for them not to develop Bulldozer,
reasons why,
Taking into account,
leaky 32nm process,
not enough cach,
does very well in multithreaded software already,
scales way better than Sandy bridge,
little software compiled for it yet.

Think of the future,
improved IPC,
increased speed with better optimization 32nm and 22nm process,
software support,
more cores,
better thermal values,
linked with GCN FPU performance, on a 14nm process you can probably expect a fully functional 4 module (8 core) chip with quad channel DDR4, GCN FPU unit all running at 4GHZ with a graphics equivalent of a 6970 performance.
Or how about a 24 core chip with a serious FPU the size of 4 modules, CPU running at 5 GHZ with a near 35% IPC improvement on every core. With serious software optimizations now commonplace. And a lower power envelope.
thats all I'm asking for, nothing more.
All at the price of a $300 APU.

Why do you think that there's not enough cache? There's 8MB of L2 cache alone on the FX-8150. I think what they need is faster, lower latency cache.

 

[know of fence]

The bulldozer architecture is nothing like P4. I suspect that higher clock rates actually will more than others benefit from future die-shrinks and so will technologies like turbo-core.

Thanks to clock scaling CPUs go up and down the clock range all day, naturally it makes sense to increase the higher clock cap. This is where P4 comments are completely missing the point. Also good luck trying to test the resulting responsiveness and scalability with a benchmark.

Having a higher range of clocks, makes the cpu much more able to directly scale with tasks, something that multi-threading is struggling to achieve every time the number of cores is increased. With the FX-cpus single core performance is increased (em, maintained) and parallelism is increased, despite less IPC.

People also tend to forget that the goal of more cores, is not to add a bullet point to the box but to actually bring GPU and CPU computing closer together. Never mind this requires software changes. Never mind that Win8 will bring further improvement still.
http://www.anandtech.com/show/5448/the-bulldozer-scheduling-patch-tested/1

 

[BrightCandle]

No. They needed to develop a 28nm CPU. It didn't turn out as performant as they had planned or hoped for but they clearly meant to put out something competitive. I am quite certain that in many ways its more advanced than Piledriver, even if it doesn't show it in the benchmark charts.

They seem to be selling everything they make so its not all bad. The future of chips is looking very different to the past 20 years, could be that things get switched up quite a bit.

 

[Wuzup101]

It is currently pretty lackluster in the desktop segment. However, given the pricing, and the total package price (including a solid motherboard), it really does compete decently in the mid range area. I would personally still go 2500k most likely, but might take a look at the lowest end BD for lighter applications (that being said, I already had purchased all of my 2500k based systems before BD came out so AMD didn't even have the option of selling those 3 chips to me).

I'm excited for what BD could become if (hopefully when) AMD gets their act together. There is some serious potential there, they just need to start tweaking. Also, I hope that further windows optimization helps BD gain some ground.... if for nothing else than competition. We definitely need MORE CPU/GPU manufactures... not less!

 

[Rvenger]

No. They needed to develop a 28nm CPU. It didn't turn out as performant as they had planned or hoped for but they clearly meant to put out something competitive. I am quite certain that in many ways its more advanced than Piledriver, even if it doesn't show it in the benchmark charts.

They seem to be selling everything they make so its not all bad. The future of chips is looking very different to the past 20 years, could be that things get switched up quite a bit.

They don't need to develop a 28nm CPU now. They need to work on maturing their current process before they move on to smaller processes.

I mean.. look at Intel and Core 2 Duo. In some ways it still is the basis of the modern Sandy Bridge CPUs, and it is awesome.

On another note about performance, yes C2D was a game changing architecture but AMD could very well come out with something that makes Sandybridge look like P4 all over again. Not likely at all, but you never know

 

[exar333]

No, I don't think it would have been better for them not to develop Bulldozer,
reasons why,
Taking into account,
leaky 32nm process,
not enough cach,
does very well in multithreaded software already,
scales way better than Sandy bridge,
little software compiled for it yet.

Think of the future,
improved IPC,
increased speed with better optimization 32nm and 22nm process,
software support,
more cores,
better thermal values,
linked with GCN FPU performance, on a 14nm process you can probably expect a fully functional 4 module (8 core) chip with quad channel DDR4, GCN FPU unit all running at 4GHZ with a graphics equivalent of a 6970 performance.
Or how about a 24 core chip with a serious FPU the size of 4 modules, CPU running at 5 GHZ with a near 35% IPC improvement on every core. With serious software optimizations now commonplace. And a lower power envelope.
thats all I'm asking for, nothing more.
All at the price of a $300 APU.

BD scales worse than SB actually. Not a positive there...

 

[exar333]

I think it's trully too early to say if it was a somplete mistake or not. There may be some positives that come from BD, who knows.

Intel did get a few good things from the P4; they added HT and also learned to be as efficient with leakage and high power consumption as possible.

Intel was more fortunate with the P4 'mistake' because of a couple reasons:

(1) For most of the P4 life, everyone was still using single cores, so the difference in IPC could be mitigated more easily by raising the clock freq to compensate.

(2) The performance difference between P4 and AXP was pretty close, and only when A64 (and especially X2) was the performance delta more significant.

AMD is sitting with a HUGE IPC penalty right now, poorer power consumption, and lower clocks. There is nothing in BD that 'sets it apart' in any real metrics at all. There are definite strengths, but nothing that really makes you need it compared to 2500k/2600k. When you compare it to a SB-E, it doesn't even hold a candle.

 

[frostedflakes]

Here's some results with the patches: http://www.tomshardware.com/reviews/windows-7-hotfix-bulldozer-performance,3119.html

Minimal (if any) gains.

Tom's saw much bigger gains in WoW under Win 8, 8-12% improvement in FPS depending on resolution. Maybe Win 8 includes additional tweaks and performance optimizations that weren't backported in the Win 7 scheduler hotfix. More CPU limited games like Skyrim and StarCraft II should have seen more solid gains.

http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-23.html

 

[Sweepr]

Tom's saw much bigger gains in WoW under Win 8, 8-12% improvement in FPS depending on resolution. Maybe Win 8 includes additional tweaks and performance optimizations that weren't backported in the Win 7 scheduler hotfix. More CPU limited games like Skyrim and StarCraft II should have seen more solid gains.

http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-23.html

10% over BD would still lag behing a 3-year old Phenom II, and by the way, it is up to ~10% (not an average 10% improvement).

AMD marketing slide from BD launch: http://img23.imageshack.us/img23/6118/bdwin8.png

For the most part Windows 8 saw a 1%-5% improvement with the AMD FX-8150 processor, though benchmarks like Sandra's multi-media float x4 test net a healthy 26% improvement in Windows 8.

Unfortunately, Sysmark refused to run in Windows 8 and PCMark Vantage gave up the ghost for half of its sub-tests. Where PCMark Vantage did work correctly, the results saw an average 4% - 6% improvement in Windows 8 vs. Windows 7. Computational benchmarks like Bibble 5.0, Sciencemark and SuperPi were too close to call.

http://www.pcstats.com/articleview.cfm?articleID=2622

Another link: http://www.anandtech.com/show/5251/microsoft-releases-hotfix-to-improve-bulldozer-performance

Any improvement is nice, but few % doesnt change the overall picture.

 

[Joseph F]

Tom's saw much bigger gains in WoW under Win 8, 8-12% improvement in FPS depending on resolution. Maybe Win 8 includes additional tweaks and performance optimizations that weren't backported in the Win 7 scheduler hotfix. More CPU limited games like Skyrim and StarCraft II should have seen more solid gains.

http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-23.html

I don't think that Skyrim is going to be CPU-limited after the v1.4 patch is released. It's been shown to improve performance up to 70% in certain cases from what I've heard.

 

[grkM3]

AMD should not of bought ATI.From there on they had limited money for r&d and thought the opteron was going to be enough to compete with intel.They were shoked when intel came out swinging with the core arch and eversince then AMD is just been trying to catch up.

I have noticed from the last 7-8 years or so from amd that Im always waiting for there next chip,even after there new chip comes out Im always waiting for there next one to compete.

They have not had a decent cpu from the athlon days.Forget about the low/mid range.I want the AMD that had the top end back.

BD came out and a week later everyone is talking about pildriver,same story on every arch from them for the last decade.New chip comes out and we wait again for there updated version of it.

When intel showed its early benchmarks of sandy and the ceo of amd stepped down should of been a sign of whats to come with BD.

He knew they spent most of there money and wasted 5 years on a flop.

 

[blckgrffn]

On the distributed computing side, I am interested in BD and definitely PD if new CPU instructions (avx) find their way into more projects. It seems to make a big difference and if they can make PD even wider than BD (more modules) while maintaining TDP and clock speeds, well, that would be exciting.