The reviews have focussed on the performance which is very disappointing but if you look at the power efficiency it’s truly appalling: see this link - http://www.pcper.com/reviews/Processors/AMD-FX-Processor-Review-Can-Bulldozer-Unearth-AMD-Victory/Power-Consumption-and-P.

The best remedy for bad news is often humour and I found this video on Hitler's response to the Bulldozer benchmarks really funny - http://www.youtube.com/watch?v=SArxcnpXStE

Actual CPU efficiency is even worse than reviews lokking at system power consumption would have you believe.

If you want to see how efficient Bulldozers can be, you'll have to try underclocking them. It seems hopeless given the superiority of Intel's tech but I have yet to see such a test.

The Xbitlabs review made that very clear:

“The “pure” power consumption of the eight-core FX-8150 is about twice as high as that of Sandy Bridge processors. Since all of them are manufactured using the same production process and have similar core voltage, it becomes extremely interesting what exactly AMD meant by the energy-efficiency of their Bulldozer microarchitecture.”

Ouch - http://www.xbitlabs.com/articles/cpu/di ... html#sect0

The power consumption of each core is actually pretty good. No doubt this is what AMD meant. But if you need 2 cores to keep up with 1 Intel core...

Never mind all this did you watch the video? Anyone that doesn’t find it funny is definitely a fanboy.





So BD with 1 thread loaded = 115W and SB with all cores loaded = 115W so I fail to see how BD has good power consumption per core?
Ultimately I focus on power efficiency at the platform level as that is what people actually buy.

http://www.xbitlabs.com/articles/cpu/di ... html#sect0

FX8150 in action... and less efficient than a Core 2 Duo E6700?

"...at this point it is hard to recommend any FX processor over the Phenom II, especially the X6 “Thuban” which, after all is one hell of a CPU. But then , Intel has better processors at this point, and they are smaller, faster, cheaper and more energy-efficient. That’s the really tough part for AMD...
...snipped out...
...Let’s take a look at the worst offenders in the current design:

# L1D cache size – too small and too slow. Especially at 16 kB size there should not be a reason to need 4 cycles access latency.
# L2 cache latency: 27 cycles. This is almost twice the access latency of the L2 cache in Phenom II and while the L2 cache here is substantially larger, the combination of the insufficient L1 size with the extremely slow L2 cache is a recipe for disaster. I dare say that by reducing the L2 latency to 12-15 cycles, Zambezi would most likely see a 20-30% performance increase. Of course, this is pure speculation because I have not run any simulations.
# AVX: I believe we have shown enough data and discussed the theory of contention and misalignment to validate the point. The question is what to do about it and that’s going to be a tough one. On the other hand, seriously, we can live without AVX, it’s been hyped and pushed and there is really not that much to it that could not be accomplished otherwise. Even if it opens up a specialty market for Intel.
# HyperThreading: if eight physical cores cannot keep up with four and eight logical cores establish absolute supremacy, then maybe it is time to really think about what decisions were made by whom and what were the agendas behind them. Corporate decisions are not dictated by rational thinking but by egos, which is why things get done but which is also why things sometimes go wrong.

..."

It looks like another big splash for AMD: we're going more and more towards a market-driven monopoly...

That factor is what turned me away from upgrading as soon as they hit the market. Performance was fine with me considering what I need, but in combination with the power consumption and to a smaller extent the price it was not right for a planned long term investment. Although an undervolting test would have been interesting to see, I have a feeling there's just way too much cache on these chips for end user needs.

I am rather disappointed with just how bad the single threaded performance, or rather the IPC (as it is the actual culprit) is.

Originally the Bulldozer's IPC was expected to be a few percent better than the Phenom 2's IPC, the expectations were dropped months ago when they said that it was "now" expected to be the same as the Phenom 2's, but clearly this is not the case, some of this is going to be improved with Windows 8.

Windows 8 will shut down pairs of cores that are not being used, which will mean that data wont have to run through the L3 cache so much between "modules" that are trying to process 1 or 2 threads and different cores in different modules will pick up and run with that thread (as currently on W7), instead with Windows 8, it will (far more likely) be processed by a single "module" thus gaining that obvious benefit as well as better use (and more frequent use) of "Turbo Core".

If the Bulldozer was actually capable of having the same IPC as the Phenom 2 as we were expecting months (years) ago, then this would be a really good CPU, at a very nice price. Alas the IPC is pants, and the clock speed even with "Turbo Core" means that it sometimes looses to the Phenom 2 that its supposed to be replacing, AMD must improve the IPC with the next refresh of the Bulldozer architecture, the clock speeds are going to improve anyway, but that is only going to help a bit, and even at 4.6GHz doesn't beat the i5-2500K in all tests.

AMD, sort your shit out, you are looking silly now, you can only win on price (again).

On the plus side, the "FX-4100" looks good at £90 and is exactly half an "FX-8150" at less than half of the price, and should overclock just as well, and most importantly it looks good price/performance wise against existing AMD and Intel CPU's at that price point - although I have only read one review that covers the "FX-4100" (guru3D) it looked pretty good all-round - bar of course the IPC.


Andy

If you don't care about power consumption, that is. I guess the i3-2100 would beat it, even at multithreaded performance if you don't overclock. And power consumption is bad enough without overclocking.

I do, but not as much as other people.



Maybe, I have not seen a head to head review of the "FX-4100" that covers (many) other CPU's at that price point, at which point a great deal of work would need to be done comparing reviews that have one or more common CPU in them with reviews that also cover the "FX-4100". Although to be honest I am really not that bothered, my motherboard is Bulldozer ready (with a BIOS update) but I have no plans to put a Bulldozer into my machine, so all contemplation is deferred.



Yes it is, I don't disagree with you there at all, but I feel that it is wise to point out that each of the reviews I have seen there have been large differences within the testing equipment, the software being tested and the results - that includes the Power Consumption.

http://www.anandtech.com/show/4955/the- ... 0-tested/9

Anand measured the load power consumption to be massive, but the idle was better than previous AMD CPU's, but still not down to the levels of the Intel competitors.

http://www.tomshardware.co.uk/fx-8150-z ... 95-22.html

Tom's came up with quite different results vs other AMD CPU's for the idle and load, but this did not correlate very well to the Intel CPU's with the obvious point that Intel is lower and AMD is higher.

http://www.hardocp.com/article/2011/10/ ... e_review/9

HardOCP again had different results, but most importantly each and every review had different results for the performance as well.

-----

Please don't believe that I have suddenly turned into an AMD fanboy, that is not the case, I want competition, and AMD has failed to give true competition at the high end with the exception of a few specific tasks, they are not competing well on the power/heat front either, and whether the lower end models such as the "FX-4100" are really competitive at that price point is to be decided when reviewers do a balanced review with other CPU's at x price point - until then I will be keeping an open mind, and I am looking forward to the next batch of Bulldozers to be released - which (finger crossed) will have better power consumption, higher clocks and cost less. Nothing miraculous is going to happen until the "Piledriver" core turns up next year, and even then that doesn't look to be fantastic.


Andy

FX-8150 v i7-2600K – idle / load

76W / 209W v 64W / 144W (TechReport - Cinebench)
85W / 229W v 78W / 155W (Anandtech – x264 HD)
60W / 174W v 44W / 115W (Xbitlabs - LinX)
130W / 271W v 91W / 172W (Hothardware – ?)

Viewed as the absolute difference between the two platforms: idle / load
12W / 65W (TR)
7W / 74W (Anandtech)
16W / 59W (Xbitlabs)
39W / 99W (Hothardware)

Viewed as the relative difference between the two platforms as in AMD consumes x% more than Intel: idle / load
18.8% / 45.1% (TR)
9% / 47.7% (Anandtech)
36.9% / 51% (Xbitlabs)
42.9% / 57.6% (Hothardware)

The average percentages across the 4 reviews are: idle / load
26.9% / 50.4%

The idle figures vary a lot which isn’t surprising but the load figures are much more closely spread.

I had high hopes for these processors but sadly looking at the numerous reviews out there FX is a direction that's the exact reverse of what AMD should have been going for (ie do more work per core not lots of cores doing less work)
Power consumption is a let down too again I had expected something pretty good but it seems that was not to be.

Heck even the 4 core Athlon II in one review had the FX 4 core gasping to keep up even putting power issues aside I can't possibly see how this range is going to appeal to users.

AMD is verging on irrelevancy.

It isn't just a matter of (bad) absolute figures, IMO.
Even if it's just one of the (possible) application/benchmarking, give a look at this:



"...It's been awhile since we have come across that kind of total energy used for a single render pass."

Maybe that "FX" name did not bring good to that CPU.

Agreed, which is why my first link in this topic was looking at performance per watt:
http://www.pcper.com/reviews/Processors ... tion-and-P.

Having had another look through the Guru3D review that also covers the 4 and 6 core Bulldozers I am very unimpressed - link below.

http://www.guru3d.com/article/amd-fx-81 ... ce-review/

I have a reasonably cheap CPU in the form of the Phenom II 840 (really an Athlon II as it doesn't have any L3 cache), which runs at 3.2GHz, it is barely any faster (100 MHz) than the Athlon II 645 in the review (3.1GHz, also with no L3 cache - same core design).

Note how the "Bulldozer" FX-4100 does against the AII 645 - not very well at all - (wins most tests, but not by a lot, and actually looses a few - notably 3D Mark), the AII 645 costs just £79 (the PII 840 costs £80) whereas the FX-4100 costs £90, which makes its closest AM3 competitor as the Phenom II x4 955 (a real Phenom with 6MB of L3 cache) that also runs at 3.2GHz like my x4 840 (no L3 cache).

The obvious winner on the AM3 platform at that price point is the Phenom II x 4 955, or if you want to save a bit of cash (and cache) the PII 840 is a good choice. the only problem with comparing the PII 955 with the FX-4100 is that the PII 955 is a 125W part (the 840 is 95W).

All in even the relatively cheap, relatively fast FX 4100 doesn't even look like a good deal compared with existing AM3 products. FYI I called the FX 4100 relatively fast because it runs at the same clocks as the FX 8150 which is why it beats its 6-core brother in some tests - that all important clock speed that translates directly into single threaded performance which is the Bulldozers bugbear multiplied by however many cores are present.

I can only see a few ways out of this fiasco for AMD.

1, Drop the price of the bulldozer CPU's by 20% - going to be costly for those large die's. This of course does nothing to improve the performance, but it will help the damned things to actually sell, and most importantly, to out-sell the AII's and PII's which they are supposed to be replacing.

2, Re-design each "module" to dramatically improve the IPC - going to be costly, and is going to take months before they hit the shelves.

3, Increase the clock speeds by about 1GHz all round so that they can compete - seems to be possible as these CPU's were always meant to clock high, and every review has overclocked them by 1GHz anyway - serious drawback on the power front though.

4, Start producing the Athlon II's and Phenom II's using the 32nm process, this would allow the clock speeds to be given a boost whilst reducing the die size, thus costing less to make - I don't see a reason why this shouldn't be done, and it is probably easier to do than some of the others - but will still take months before they hit the shelves.

5, Make the cache faster - AMD's on die cache has always been slow compared to Intel's, but this time around it looks like a very weak link in the chain.

If I were AMD I would do all of the above, number 4 as a safety net on the basis that number 2 could run way over a few months (Bulldozer was very very late already). Dropping the price would be an instant boon and would allow them to sell well right now because they would be price competitive instantly.

Rapidly ramp up the clock speeds over the coming months to stay competitive with Intel who will drop the price of their CPU's once AMD dropped the price of theirs, find some spare 32nm capacity to churn out some AII's (ideal for their small die space) to beef up their low end offerings for a while whilst re-designing the Bulldozer so that they can be competitive when they release the "Bulldozer 2" in the middle of next year.


Andy

I wouldn't go so far as to say AMD are almost dead that ignores the appeal they have for low cost systems..I've turned over a few unlocked Semprons to good effect lately and the Athlon II range is still quite well priced as are some of the PhII ones that have had price cuts.

Bottom line is the OEM market is far bigger for AMD but this is not going to help at all for enthusiast systems.
Watch this space for a big price cut because FX is simply not competitive in the marketplace or appealing.

I def made the right choice going for a more modest AMD cpu that quad core has a good bang per buck ratio and is pretty fast (fast enough for me) Shame though as I might have updated to an FX but now nope..not even slightly interested. I will continue to build more value based pc's using AMD so don't write them off yet..but they need to get their act sorted out and pretty fast.

Agree with the above the PhII 840 is well priced and I picked that one. AMD can't upsell me to a 4 core FX because it's not an upgrade (loses on many benchmarks despite the higher clock rate and it costs more!) The 6 core FX is more expensive than the 6 core PhII and can't really beat it either so if you want to upgrade their you'd likely just get the PhII 6 core over the FX esp as the prices have fallen of late

8 core FX is just too pricey at £200 odd try £150 ish for that £110 for the 6 core and £70 for the 4 core that would make a lot more sense to buyers and upgraders.
AMD could have just done as the poster above says moved to the new 32nm fab with the older Athlon II and PhII designs slapped some more lv 2 cache in there and some tweaks and they would have beaten FX no problems hands down. New design or not FX is a bit of a turkey esp at it's current price level.

I have noticed a lot of similarities between the "Bulldozer" and the Pentium 4.

They under perform generally when compared to their predecessors and their competition.

They run hot.

They are designed to clock high - and that is the only way they can perform.

They have a low IPC.

They are over-priced.

They seem to have been designed my the marketing team rather than engineers - with the P4 the goal was clock speed, with the "Bulldozer" the goal is core count (which to be fair is exactly what the server market is looking for, along with strong IPC, low power, cheapness and existing socket compatibility).


Andy

Yes you are right in some ways the Pentium 4 range went up high in clock speed and were heavy on power draw with fairly big TDP's. More to the point at the time the Athlon 64 single and dual cores ran cooler, used less power and performed far better even at much lower clock speeds.
I ran an Athlon 64 x2 4200 for a long time and back when I got it it's performance was easily as good as a dual core P4 at 3.2 GHz running at 2.2GHz so the gap AMD had was significant at that time.

Years on the situation is reverved now Intel have better performance with less cores and less heat. I'm sure I am not alone among AMD fans wanting the company to get back to the old days. I was expecting the 6 core to appeal the most to me...sadly it can't get near the i5 2500k and that was the goal for it IMO hell it can't even seem to beat out the 6 core PhII either!. Not a hope in hell I'd shell out £200 for the 8 core fx. Some hefty price cuts might help a bit but the most worrying part is that even with increased clock speeds the FX range does not seem to scale well in performance. I doubt the next batch of FX CPU's will be a significant improvement unless a heck of a lot of work goes on in the next few months. None of the FX models are well priced and the slower clocked 8 core isn't likely to do that well either.

Or maybe someone will release a software patch to add 50% performance boost to the FX models

The only good news is that AMD seem to want to stick with AM3+ for a few more years (really they have to being honest) I'm also not sure the socket FM1 is a good idea either. Budget wise I can't build an FM1 socket machine for the cost I can with a budget AMD2+/AM3 board with an Athlon II. Lots of questionable choices from AMD of late the budget stuff is the bread and butter AMD builders diet, well least for me it is.

It is a bizarre change with the Bulldozer, its a modern role reversal of the Athlon 64 vs the P4 that you have mentioned.

I was looking at the quad core and the 8 core as the Bulldozers to look at on the basis that they were going to have the same IPC as the PII, but with a higher clock (which is where the other 8-core and the 6-core fall down), new instructions and so on, but alas its turned out to be pants.

I would disagree with your opinion about the scaling of the Bulldozer, it scales almost linearly providing that the software can run with 8-cores - this bodes well for the Bulldozer design principle for the future i.e. keep on adding pairs of cores and so long as the software can use the extra cores it will continue to scale well.

The problem is the IPC plain and simple, there is no serious scaling going from the quad-core, to the hexa-core to the octo-core CPU's with just 1 or 2 threads because a single thread cannot get any real boost from additional cores unless that is waiting for something else to be processes - but that then instantly makes it dual-threaded anyway.

As far as the FM1 socket is concerned its almost dead already........ It will be replaced with the FM2 socket and there is not going to be any compatibility between them (as far as I know), with the FM2 socket coming out with a "Bulldozer" derived (and hopefully massively improved) CPU core instead of the "Stars" core that is in the FM1 socket currently. AMD claim a big jump in overall CPU performance (wont be difficult) and lower power consumption.

I will wait and see what the FM2 platform will really be like, I don't believe what AMD say after being lied to by them several times about the performance of the "Bulldozer" architecture.


Andy

Power efficiency is worth exactly nothing without accompanied performance.
Bulldozer should dig a deep hole and throw itself in it. AMD made an enormous mistake putting this thing out, maybe 4 years ago it would have a place in the world but today it already obsolete at the day it was launched. Sad days for all enthusiasts around the world but there it is.
Intel, here I come.

Agreed.



Its got the right equipment

I do think that AMD has a decent chance in the server market with this product, but desktop software and the way people use desktop machines is still mostly about a few fast cores, and the Bulldozer is not that. If AMD can release a dual-die 12-core CPU next year for the server market they will still be fine even if they don't change anything else at all. But in the desktop market they have simply got the wrong product for sale at the moment - although in a few years things might be different, all software might be heavily threaded and everything is about having lots of cores and not a few individually fast cores as it mostly is now.

I would have liked to have seen some kind of hybrid CPU that has for example 4-cores, 2x low performance SIMD cores that are very good at heavily threaded things and for pulling up the rear, and 2x very high performance single threaded cores for churning out data in a serial fashion - I dont know how practical that would be to design and build, but I am sure that someone could have come up with something along that principle. Although the idea of a slow fat processor is called a GPGPU, and a fast singular processor is called a CPU, so somehow AMD's lines got blurred between those 2 ideas, and they really have not won either.


Andy

The problem with that sort of analysis is that it ignores the source of Intel's huge profits. A publication like Forbes should do better and look at the numbers!
2010 revenue: Intel 43.6, AMD 6.5
2010 income: Intel 11.5, AMD 0.5
It's not that Intel is bigger than AMD. It's that Intel is very profitable while AMD doesn't have enough income to compete.

x86 is still very important and Intel has effectively had a monopoly on it for a few years. That's why Intel is so profitable. Intel pays off AMD and prices itself out of some market segments to keep AMD alive. AMD doesn't compete for real anymore: it's a figleaf that allows those who don't want to see what's happening to pretend Intel is not raking in monopolistic profits at the expense of the public.

Andy, do you work in the IT market, there in the Essex (just to know)?

IMO Bulldozer has virtually no chance (other than an hypothetical price leverage) just in the server market, where power efficiency really matters (average "joe enthusiast" isn't a silencer nor a greener).

When you have to deploy in a massively dense server farm (or in some linux-based supercomputer), to say, 16384 CPUs, and the AMD ones will draw maybe 4 times the electric power and perhaps 2 times the space (not for the bigger dies, but due far more heat to dissipate) of some noticeably better performing Intel counterparts (right now and the same amount of money), which man would ever sign such a less-(or-not)-profitable contract with Sunnyvale?

Not to mention the upcoming LGA2011.

Intel will raise prices as high as it needs to keep AMD competitive in enough market segments to ensure its survival. Until such time as people are willing to move away from x86 or that some other x86 competitor shows up, that's hardly hypothetical.

More accurately put, Intel has had an x86 monopoly since Intel invented x86.



In a server farm, large business or for a supercomputer you are totally right, however most servers that are sold are barely different from standard desktop hardware, that is where AMD has a real fighting chance, that and Virtualisation where life is often simpler to have more cores that don't have to be very fast, they also don't have to use lots of power because power can be reduced at the expense of performance...... but anyway no-one will really know for sure for another 12-months when there will be a whole year worth of data to look at that will give a clear picture, either AMD's market share will have gone up, stayed the same, or gone down - regardless of what happens with the market share we all want to see AMD sell lots of server CPU's because they can make lots of money and then pump that back into R&D to compete better with Intel.

The success of Bulldozer is in everyone's interests, not just AMD fans, but Intel fans as well. If it wasn't for the massive success of the "Athlon" brand of CPU's Intel would be making the Pentium 8 and it would suck donkey-balls - that is ultimately why I and everyone else is so annoyed that Bulldozer has not kicked some serious arse like the Athlon did when it was first released, and then the Athlon 64, and the Athlon x2.


Andy

I'm curious to see how the Bulldozer-based Fusion APUs will be. AMD already bet their success on APUs the moment they bought ATI; it's possible that, much like Llano, the Trinity CPU performance will be "good enough" and the GPU performance will outdo what Intel can provide. That could get them a significant chunk of the low-midrange market, and could be popular for OEM all-in-one systems.

BD-based APU will be what BD is, inferior to last gen phenom, athlon x3 with HD7350 likewise. CPU performance per watt will be 3x times less compared to IVY, unless they employ 28nm but still who would want that.

The APU's are not overly exciting but are ok for what they are (if you care about budget 3d performance) I'd consider them for some build but at the moment the budget boards are not as budget as they could be leaving Athlon II's as a better choice for those types of machines. I would still prefer one socket for all AMD desktop processors.

It might be a bit tricky to make a profit on the 8 core FX the die size is massive anyway even on 32nm. What annoys most is you keep hearing about these comments evidently from AMD "performance will be better in windows 8" (hmm not even out yet maybe I'm happy with Win 7?) or software will be better optimised for BD with it's new instructions and when developers exploit all the cores on the FX series.

Folks don't care about "wait" we've already done that for ages with FX they simply want good, fast, and low power consumption right now. This won't sink AMD (watch the marketing machine at work) but it's not going to do well. I can see a lot of folks who were waiting for FX simply go ahead and get their 2500k or 2600k CPU and board in the last few days.

I started this thread and used the term power efficiency and linked to a review that looked at performance per watt; power efficiency doesn’t equal power consumption. The word efficiency relates to how it uses the power not how much power it consumes. I don’t know why so many people on this thread have missed this point!

As bad as BD is from the power efficiency perspective at the platform level if you look purely at the CPU it is even worse. So from a silencing perspective it’s a disaster as the CPU is about half as power efficient as SB. The over-clocked power efficiency is even worse, it’s staggeringly bad, it behaves more like a top end GPU that has been O/C’d.