Why is the response to Bulldozer so overwhelmingly negative?

[frostedflakes]

BD was overhyped like crazy, I knew a lot of people were setting themselves up for disappointment. I didn't even buy into all the hype, though, and I was still really underwhelmed by the performance. Lower performance per clock and per core than Phenom II kind of caught me off guard, wasn't really expecting that.

Yeah, it performs well for some heavily threaded applications, but this kind of stuff is a pretty small niche for desktop users. For what most of us actually do day-to-day with our computers, Sandy Bridge is the clear winner, and the real kick in the nuts is that our old Phenom II CPUs actually seem to perform better in these single and lightly threaded tasks. At best BD is a sidegrade, at worst it's a downgrade.

And with respect to the heavily threaded performance, if you compare power consumption of the 2600K and FX-8150, any money saved by going with the Bulldozer might be offset by the energy savings with the 2600K.

And as far as the low-end BD, they look even more underwhelming. It's a leaked benchmark, but all the other leaks ended up being true so I have no reason to distrust these. Here is an FX-4170 (4.2GHz base, 4.3GHz turbo) getting 15.28x in Fritz chess benchmark. For comparison, my Phenom II X4 955 at 3.2GHz gets 15.03x. Even in heavily threaded benchmarks 2M/4C BD looks terrible compared to Phenom II. This is just one benchmark, but all the ones this guy posted seemed to tell the same story. And of course single threaded performance will be crap just like it is with the FX-8150. And keep in mind this Fritz score is for the 125W 4170 that's clocked at very high frequencies. The FX-4100 will only be clocked at 3.6 base and 3.8 turbo and will more than likely be slower than the X4 955 in both multi threaded and single threaded situations.

 

[Phynaz]

Are there really so many enthusiasts here and elsewhere who are so myopic they fail to see the flagship Bulldozer SKU beating the flagship Core SKU in a few relevant, real world applications?

There's your answer. If your job depends on a few cherry picked applications then BD may be viable for you.

For the rest of the world it's basically a failure.

 

[Hypertag]

I suppose I'm just a positive thinker, but perhaps there isn't much (if at all) optimization for a brand-new CPU like the BD?

Sure SB has ownership of all current "benchmarks", but what about a few years from now? Software has to keep up with hardware, we all know that right now hardware is advancing faster than the software that utilizes it...

My other train of thought is:

Wouldn't 2 billion transistors be good for SOMETHING? I can't see AMD putting out a 2 billion transistor chip without seeing a massive improvement.

Seeing the overwhelming lead in 7-Zip makes me wonder if in time it will do the same performance in other applications?

I think the file compression tests can be explained by the enormous amount of L3 and L2 cache the processor has. File compression tests have always been more sensitive to Cache sizes. I remember this going back to 2002 with zip compression speeds.

 

[Avalon]

Because half the time it can't even beat the X6 in multi threading apps, let alone the 2500k. Sometimes it does, but sometimes doesn't cut it, especially when it uses double the power. To top it off, lightly threaded apps (not even single threaded) perform horrendously compared to what's currently available.

 

[Gigantopithecus]

So even though we've had to put up with bulldozer thread after bulldozer thread, it isn't going to stop now that it's obviously a massive flop. (Well, it was obvious months ago really.)

Now we are going to get endless fanboy damage control threads? Here is a tip, self-identify with a faceless corporation less, and you will actually be able to make smart buying decisions.

...Did you miss the part where I explicitly stated that I have very good reasons to believe that the FX-8150 will be a better performer than even an i7-2600 for what I do with some of my computers? Here's a tip - read first and make condescending remarks second, and then you won't sound like an ass. You must not have read the new sticky at the top of the forum, either.

This thread was posted simply to point out that for all its flaws - which are numerous - it has strengths, too. CPUs are becoming increasingly niche-specific. I'm sorry, but I do not understand why so many people dismiss CPUs built for other people as 'failures.'

 

[Gigantopithecus]

And as far as the low-end BD, they look even more underwhelming. It's a leaked benchmark, but all the other leaks ended up being true so I have no reason to distrust these. Here is an FX-4170 (4.2GHz base, 4.3GHz turbo) getting 15.28x in Fritz chess benchmark. For comparison, my Phenom II X4 955 at 3.2GHz gets 15.03x. Even in heavily threaded benchmarks 2M/4C BD looks terrible compared to Phenom II. This is just one benchmark, but all the ones this guy posted seemed to tell the same story. And of course single threaded performance will be crap just like it is with the FX-8150. And keep in mind this Fritz score is for the 125W 4170 that's clocked at very high frequencies. The FX-4100 will only be clocked at 3.6 base and 3.8 turbo and will more than likely be slower than the X4 955 in both multi threaded and single threaded situations.

I hadn't seen that benchmark - thank you! ...And, I agree - if the FX-4100 ends up being inferior to the Phenom II X4 955, then that result is overwhelmingly negative. :lol:

 

[Vesku]

Do you know how the 1100T will compare to the 8150 for your purposes, Gigantopithecus?

 

[BTRY B 529th FA BN]

Bulldozer succeeded in failing 2-fold

 

[4ghz]

I don't understand the logic behind your argument. People shouldn't care if bulldozer is inferior unless they are able to experience and feel the inferiority right now? Just because Bulldozer's weaknesses are currently being hidden behind other bottlenecks doesn't mean that will always be the case. Some people keep cpu's for 4 to 5 years or more. You still regularly see people on these forms asking is X video card will be bottlenecked by an Athlon 64 x2 or Core 2 duo.

Personally the biggest failure for bulldozer is the ridiculous gobs of power it sucks when overclocked. You should not need a 800W+ power supply to power an overclocked cpu and a gtx 460. That's just ridiculous. My antec 750w supply powers my overclocked 2500k and twin gtx 560 ti's and 6 hard drives with room to spare...

 

[Arkaign]

I would also caution gamers against looking at a few GPU-limited scores being close to SB and coming to the conclusion that BD makes a decent gaming choice.

The fact is that as GPU power increases, the 2500/2600k will be able to continue to deliver better performance, and the BD will hit CPU cap sooner. A good example is games that are already starting to show lower minimums and average framerates with C2Q and PhII vs. 2500/2600k.

 

[Gigantopithecus]

Do you know how the 1100T will compare to the 8150 for your purposes, Gigantopithecus?

My research entails a lot of MCMC (though not in Excel - yikes!) and the FX-8150 looks to be about 15% faster at MCMC - but I do not know what the parameters of the Anandtech Monte Carlo sim are, so I don't know for sure yet.

The reason the 7-Zip test is relevant to me is that the genomic data with which I work are heavily compressed, and since we don't have hundreds of terabytes of hard drives, I have to decompress them, analyze them, then recompress them, rinse, repeat. It's not an elegant solution but I'm an anthropologist, not a computer scientist. Again, looks to be about 15% faster than the 1100T.

I didn't order the FX-8150s from Newegg (price gougers!), but I should have everything up and running by the end of the month.

 

[Ferzerp]

If you're doing Monte Carlo sims, Bulldozer still is an awful choice. That belongs on a video card.

http://gpgpu.org/2011/01/23/trng-library-parallel-monte-carlo

http://www.atomic.physics.lu.se/biophotonics/our_research/monte_carlo_simulations/gpu_monte_carlo/

http://www.gpucomputing.net/node/6156

 

[OVerLoRDI]

My complaints with Bulldozer:

Thermals
Performance/Watt
Very little performance increase per thread vs Phenom II

The first 2 could be killer for this chip in the server market. I wouldn't care much at all if my personal CPU used more watts and pooped out more heat, but scale up the operation and power and cooling becomes a massive cost.

You have mentioned that most people won't take full advantage of a i5-2500k or a FX-8150 anyways, but most desktop apps still single threaded. Single threaded performance will always be important because parallelism doesn't exist for all tasks/software. Keeping that in mind why not pick the better rounded chip, which is clearly the i5-2500k.

The point about the lack of benches on the lower end chips is a very valid point. I would be curious to see, especially since the non-k chips can't overclock.

 

[sm625]

Why is the response to Bulldozer so overwhelmingly negative?

Because we now have proof that AMD is bankrupt. They cannot produce a 400 sq mm cpu/chipset that can compete with a 200 sq mm intel cpu/chipset. No way they survive that. Not when intel is about to turn around and offer the same performance at 140 sq mm. At that point it will take 3 times the silicon and 9 times the electricity to do the same workload.

 

[Gigantopithecus]

If you're doing Monte Carlo sims, Bulldozer still is an awful choice. That belongs on a video card.

http://gpgpu.org/2011/01/23/trng-library-parallel-monte-carlo

http://www.atomic.physics.lu.se/biophotonics/our_research/monte_carlo_simulations/gpu_monte_carlo/

http://www.gpucomputing.net/node/6156

Yeah, I know, but MCMC is a small enough part of my workload that it's not worth dumping the money into a dedicated GPU computing system (yet - though I have a few grants out that might change things if I get them). The vast majority of what I do is the genomics work.

 

[Gigantopithecus]

Because we now have proof that AMD is bankrupt. They cannot produce a 400 sq mm cpu/chipset that can compete with a 200 sq mm intel cpu/chipset. No way they survive that. Not when intel is about to turn around and offer the same performance at 140 sq mm. At that point it will take 3 times the silicon and 9 times the electricity to do the same workload.

That's an interesting interpretation. So how do the APUs figure into your consideration? Intel has no answer to Zacate and Llano is more frequently the superior choice for most mobile consumers (if I might speak with such generalities).

 

[Maximilian]

OP did you even read any of the reviews? This is what bulldozer sports:

- 2 billion transistors, about 2x what sandy bridge has
- First wholly new architecture in a decade from AMD
- 8 cores
- Higher clock speeds

Yet despite this it only sometimes outperforms its predecessor Phenom II?! Plus it dosent really do all that well in multi threaded stuff either which it was supposed to be built for. Never mind the thrashing it gets when pitted against a 2600k or 2500k.

Look at this... bulldozer vs nehalem bulldozer has a 1 ghz clock advantage and is 3 years newer and is on a 32nm process and is best in class against the slowest of the 9xx series

http://www.anandtech.com/bench/Product/47?vs=434

They trade blows. Despite all the advantages it dosent outright beat what intel had 3 years ago, that is a failure in my view.

 

[Dresdenboy]

maybe because 99% of users are better of with SB while the last 1% are better off with BD

This is one point. Another is: x% are better off with legacy software.

During a quick look at disassembled SuperPi code I saw enough pitfalls for BD so that it might often only be able to decode 1 or 2 instructions per clock instead of 4.

 

[THRiLL KiLL]

are there any reivews with a 68xx or 69xx video card and bulldozer?

i thought there was susposed to be a gpu boost

http://en.wikipedia.org/wiki/AMD_Fusion

 

[amenx]

As someone else said in another thread, its best not to try to convince people like the OP to switch, we need people like him to keep AMD afloat while we reap the benefits of SB.

 

[Ferzerp]

Yeah, I know, but MCMC is a small enough part of my workload that it's not worth dumping the money into a dedicated GPU computing system (yet - though I have a few grants out that might change things if I get them). The vast majority of what I do is the genomics work.

Important enough to waste money on these space heaters, but not important enough to put a $200 upgrade in to existing systems that will perform an order of magnitute or more above the space heater?

Are we being trolled?

 

[Maximilian]

are there any reivews with a 68xx or 69xx video card and bulldozer?

i thought there was susposed to be a gpu boost

http://en.wikipedia.org/wiki/AMD_Fusion

Thats only with the llano cpus, A8 etc, the FX stuff launched today dont have integrated graphics.

 

[Belegost]

So as far as I can tell, then FX excels in highly parallel integer operations... for which I would do far better off-loading to GPU (checkout MAGMA and other similar libraries for mixed CPU/GPU scientific/engineering coding - easy as pie)

So setting aside that advantage, the performance is overall worse than a 2600K, and sits near a 2500K for multi-threaded loads. For code that needs to be operated on sequentially, the single-core performance is abysmal.

Then there is the power issue... under load the FX is using 30-50% more power, and a bit more at idle. Even if I'm not paying the bills, the heat and noise in a small lab with a dozen or so machines adds up fast.

The price is on par with SB, so no win for budget machines.

So, I see no practical advantages, and a major disadvantage with going FX, so why would I not be negative?

 

[SithSolo1]

I can walk into Mircocenter and pick up a 2500k for $180 or I can get the FX-6100 for $10 more. Why anyone would pick up an FX for gaming at this point in time is beyond me.

Even if you buy online the 2500k is $20 cheaper than the 8120 and you can get a z68 mobo for roughly the same price as a 9xx series AMD board.


If all you do is zip, encode, and bench mark files all day maybe an 8-core BD might be better but is it worth the $20-60 price difference?

Lets not forget that in the long term the 2500k will save you money in power usage vs the 6/8-core BD OC'd or not. It might not be much monthly but over a 2 year period it could add up to $50-100 if you leave your computer on 24/7.

 

[sm625]

That's an interesting interpretation. So how do the APUs figure into your consideration? Intel has no answer to Zacate and Llano is more frequently the superior choice for most mobile consumers (if I might speak with such generalities).

Intel's answer to zacate is the celeron 847. But there isnt really a market for it because the B8XX/B9XX are so much cheaper, more powerful, and the power draw increase really isnt that bad. The B940 really is one of the best budget chips ever made, and intel is basically giving them away.