[VR-Zone] Intel Core i7 3770K@4.6GHz vs AMD FX-8150@4.6GHz

[AtenRa]

You know it is funny that before DB launched, people were comparing it to the Intel "E" class of CPUs (980X, 3960k, etc) due to its core count. Then when it got crushed by those CPUs, all the AMD fans started posting about market segments and such saying that BD should be compared to the 2500k and 2600k series of CPUs, in other words mainstream. Now that BD is getting crushed by those mainsteam CPUs, now we have AMD fans saying that is not a fair comparison either.

Personally i dont care if a CPU has 1000 cores or 2. It was always performance that drive prices. If BD was faster it would cost more.

It is fine to compare the two if you like, but people should not forget the price difference.

So bottom line is the AMD fans will ONLY compare BD to something that BD can beat, regardless of core count, market segment, price, etc. Good to know.

Everyone, not only fans, should compare with price to performance. BD has a price of $199,99 because that's how fast it is. If it could beat the 3770K it would cost more.
Unfortunately for everyone, BD is not competitive with Core i7 socket 1155 CPUs not to mention socket 2011.

 

[dastral]

I don't think this comparison is "too bad" for bulldozer, basically it shows that for "well threaded apps" it will probably compete with a 3570K (which costs the same).
While being hopelessly behind on poorly threaded apps and power consumption (big problem here).

If AMD can pull another improvement like it did with the Phenom vs PhenomII we may have a contender for the 150-200$ crown for "every type of load". But Intel will remain King of the Hill for a long long long time.

 

[Homeles]

Yet , it was enough to be understood , so you re just contradicting
yourself with this kind of trashy remarks.....

I was being nice by participating in conversation with you at all. Regardless, good communication skills are important for making any argument.

Cinema4D rendering is massively floating point , yet BD does
not do well in this bench as well as in anything that is FP dependant.

Cache is a huge issue here. Sandy Bridge has nearly twice the hit rate, and suffers a much lesser penalty for misses than Bulldozer. Also, because of the small size of the L1 data cache, Bulldozer has to use its terribly slow L2 cache more frequently.

With that said, yes, software optimization would help significantly, but the hardware issues are there. Optimization is also a two way street -- Cinebench isn't fully optimized for Ivy Bridge either, even though (I believe) it is compiled on the Intel compiler.

NO , it s enough hollow wordings with no usefull point....

They wouldn't be "hollow" if you had an idea of how CPUs work. Also, it's a bit hypocrital for you to say there's no useful point, yet agree with it in the quote below: "hence" implies you agree that there are many variables, which was my point.

Hence the necessity of specific optimizations for each uarch.....

Yes, optimization is important, but it can't overcome weaknesses at the silicon level.

Optimizations are still lacking for this CPU , otherwise , its
main problem are internal intermediate latencies that had
to be hugely increased to get te silicon working properly.

These shortcomings wont be completely adressed before
steamroller cpus.

Yes, Bulldozer lacks software support, and likely won't get it for some time. But who buys a processor for tomorrow, when there are processors that handle today's workloads much better and will still handle tomorrow's workloads very well? Tomorrow may never come, however the present is certain.

Bulldozer is a processor for a certain niche; Ivy Bridge is better at handling a wider variety of workloads. AMD intentionally made that decision. This weakness was supposed to be overcome with higher clock speeds, but they did not achieve this.

 

[Vesku]

Lack of optimization sure can make it look worse, though. Phoronix.com if you want to peruse Linux results where more effort was spent optimizing than with Windows thanks in large part to it's open development model.

 

[gramboh]

Since you like percentages so much, here's one for you:
FX 8150 is 250% the price of Pentium G850, while being 15% slower in games.
http://www.anandtech.com/bench/Product/434?vs=404

Just wanted to highlight this as no one addressed it. Mind blowing.

Who are the FX chips meant for? They seem to be marketed to gamers/"extreme" home users. This is a sad state of affairs.

 

[Homeles]

Lack of optimization sure can make it look worse, though. Phoronix.com if you want to peruse Linux results where more effort was spent optimizing than with Windows thanks in large part to it's open development model.

As I said, I don't disagree. Putting the blame solely or even primarily on software is just silly though.

 

[Abwx]

I was being nice by participating in conversation with you at all.

Your participation is for your own sake , otherwise you wouldnt
be there , so please no more of these kind of arrogance , no one
is absolutely necessary in this world...

Regardless, good communication skills are important for making any argument.

Certainly when it comes to the marketing departments but in the
engineers s side the nicer english in the world wont be enough
to overcome a mistake in an equation.

Cache is a huge issue here. Sandy Bridge has nearly twice the hit rate, and suffers a much lesser penalty for misses than Bulldozer.

Rather , it has half the miss rate of BD , and surely not twice the hit rate.

Yes, Bulldozer lacks software support, and likely won't get it for some time. But who buys a processor for tomorrow, when there are processors that handle today's workloads much better and will still handle tomorrow's workloads very well? Tomorrow may never come, however the present is certain.

The same that are saying that an i3 do better than a BD or a X6 in games
are downplaying the fact that the i3 is already used at its full potential with currently lightly threaded games , but in two years , because most people buy a PC for more than two years use , we ll see how the things work with better multithreaded games , with theses same CPUs...

Bulldozer is a processor for a certain niche; Ivy Bridge is better at handling a wider variety of workloads. AMD intentionally made that decision. This weakness was supposed to be overcome with higher clock speeds, but they did not achieve this.

Surely that the things didnt work as expected but overall it seems
that AMD was too early on suggesting ever more multithreaded
softs as a rapid evolution , wich is not the case.

 

[pantsaregood]

I don't think this comparison is "too bad" for bulldozer, basically it shows that for "well threaded apps" it will probably compete with a 3570K (which costs the same).
While being hopelessly behind on poorly threaded apps and power consumption (big problem here).

If AMD can pull another improvement like it did with the Phenom vs PhenomII we may have a contender for the 150-200$ crown for "every type of load". But Intel will remain King of the Hill for a long long long time.

Phenom II was a die shrink of Phenom with an additional 4 MB of L3 cache. The only thing it really accomplished was drastically improving clock speeds. Other than the units with the TLB bug, the original Phenoms weren't really that bad - they just had a hard time getting clocks up.

Phenom II didn't really do much for reducing power consumption.

An improvement like that of Phenom to Phenom II wouldn't even be close to enough. I doubt they're going to pull a 42% clock speed increase out of Bulldozer like they did with Phenom.

 

[EltonL]

Hmm, it seems that it isn't just a cache issue, but a problem stemming since the the Althlon 64 X2s; branch prediction. The slow cache has to do with the flushing of the pipeline. What AMD never did enough of was invest in advanced branch prediction. Something which intel dumped tons of money into researching.

Hell, just compare the branch predictors between Conroe/Merom and AM2 and look at their respective IPC performance.

 

[pantsaregood]

K8 and K10 cores (they're almost identical) really don't look that bad by modern standards, honestly. Llano boosted performance modestly, too.

Eight-core 32nm Phenom with 8-12 MB of L3 and some sort of AVX/AES/SSE4 implementation would've kept the gap between AMD and Intel from widening. A CPU with those specifications at ~3.8 GHz could likely outperform a 3770K pretty consistently.

 

[EltonL]

K8 and K10 cores (they're almost identical) really don't look that bad by modern standards, honestly. Llano boosted performance modestly, too.

Eight-core 32nm Phenom with 8-12 MB of L3 and some sort of AVX/AES/SSE4 implementation would've kept the gap between AMD and Intel from widening. A CPU with those specifications at ~3.8 GHz could likely outperform a 3770K pretty consistently.

The yields would've been pretty bad. And it isn't as simple as adding cache. Part of the performance deficit was present also has to do with K10's lower IPC. The reason being that their pipeline in terms of branch prediction (how accurate it is and how well a processor predicts certain loads, errors causes a pipeline to be flushed consuming more time to complete the workload) is still inferior to Intels. Couple it with memory management upgrades and the entire QPI thing Intel created and you have a pretty large gulf.

Phenom II was competitive with Penryn to an extent. It didn't exactly stand a chance against Nehalem, god forbid Sandybridge.

 

[Hatisherrif]

...2500k and 2600k series of CPUs, in other words mainstream. Now that BD is getting crushed by those mainsteam CPUs,

Dear haters,

Your mom didn't find me mainstream.

Sincerely, i5 2500K.

 

[Hatisherrif]

Phenom II was competitive with Penryn to an extent. It didn't exactly stand a chance against Nehalem, god forbid Sandybridge.

It was absofreakinglutely competitive with Core2 Quads, and by the way I'm guessing you've never owned one, so you're just big on words and internet benchmarks. It stood a chance against Nehalem in gaming, and considering that the six core version reached Nehalem performance it wouldn't be too optimistic for a well-designed eight core version to reach Sandy Bridge. This was the original argument, and it is valid.

 

[EltonL]

It was absofreakinglutely competitive with Core2 Quads, and by the way I'm guessing you've never owned one, so you're just big on words and internet benchmarks. It stood a chance against Nehalem in gaming, and considering that the six core version reached Nehalem performance it wouldn't be too optimistic for a well-designed eight core version to reach Sandy Bridge. This was the original argument, and it is valid.

I had one for a bit, it was blazing fast. But they were at pretty much the end of the line (AFAIK) of the K10 architecture. I will gladly concede that I was speaking in terms of Server/HPC benchmarks and performance more than gaming performance. So your point stands decently.

At an architectural level though, Bulldozer had some awesome ideas that were marred by an odd repeat of the NetBurst era. Mind you it took Intel some years to realize that branch prediction and error elimination within a processor was necessary.

All I'm saying is; while they are close (close enough to not be noticeable unless you need that much CPU bandwidth or you need power efficiency), AMD's architecture is still a bit rough around the edges.

Note that the TDP of the PIIs at the time weren't as power efficient. But the original premise of including more cache and clock speed is correct I will admit. Although I wonder how far they could've gone with Phenom III..

 

[SlowSpyder]

Note that the TDP of the PIIs at the time weren't as power efficient. But the original premise of including more cache and clock speed is correct I will admit. Although I wonder how far they could've gone with Phenom III..

I see what they were trying to accomplish in BD's design, at least I think I do... an efficient way to handle more threads, I think AMD was trying to out Hyperthread Intel. But somewhere along single and lightly threaded performance went too far down hill.

I have to think a PhIII with some tweaks for ~15% better IPC, more/better L3 cache, higher clock speeds for both the cores and the NB/L3 cache, and of course an octo-core version eventually would have put AMD in a better place now than BD has. If they could squeez 10-15% more IPC and mid 4GHz clock speeds, they wouldn't surpass Sandy Bridge but it would eclipse BD by a good margin.

And of course if AMD made it AM2+ compatible for me, that'd be great. I suppose I won't hold my breath. It's getting harder and harder to hold off the urge to get myself a new Intel set up. I'd love for AMD to make a comeback with Piledriver and show us all why they went with BD in the first place, but something tells me all Piledriver will do is cement me as an enthusiast squarely in Intel's camp for CPU's. But you never know. And with that semi-rambling stream of conciousness, I'll end this post now.

 

[Hatisherrif]

Fair enough. I've never owned a Bulldozer so I can't say anything good or bad about it, but I can say I'm extremelly happy with the i5. It's a little monster (at least for what I do).

 

[EltonL]

Bulldozer was a victim of planning ahead too much if you ask me. Create 8 3/4 cores (well akin to 3/4 cores) and rely on parallelism to bolster your performance. It's a good idea in theory, but you run into latency issues when you start to divide workloads and such.

On the other hand the other problem was finding a way to increase clock speed on the K10 architecture without suffering massive diminishing returns. On a realistic level, a six core or 8 core K10 chip (Thuban derivative) wouldn't be able to clock as high unless they managed to create a good balance between yields and the die shrink. Factoring all that in and you might have had a recipe for a better chip.

That said, I do hope they learn from BD as it isn't inherently flawed, there needs to be a C2D-esque shrinking of the pipeline stages and better path/branch prediction based on my limited knowledge of CPUs.

snip

I have to add that while they were competitive, in terms of IPC/Clock per clock performance, the Phenom IIs still drifted a bit from comparative Yorkfield based processors. Not much, but still slightly.

 

[MustangSVT]

here we go again, why do these amd trolls even bother. It's not even a close race and they still bring up ridiculous reasons. not entertaining anymore

 

[nenforcer]

I have to think a PhIII with some tweaks for ~15% better IPC, more/better L3 cache, higher clock speeds for both the cores and the NB/L3 cache, and of course an octo-core version eventually would have put AMD in a better place now than BD has.

 

[pantsaregood]

K10 with 8MB L3, modern instruction sets, 10% higher IPC, eight cores, and full speed cache? At 4.0 GHz? I don't think anyone realizes how much trouble that would cause for Intel. Such a CPU could easily leave any current Ivy Bridge units behind in multithreaded situations, and it would compete pretty evenly in single threaded situations.

With those improvements, a Phenom III X4 would be very competitive with IB i5s. Full speed cache, 10% IPC increase, and modern instruction sets would give close to a 25% performance increase per clock. Suddenly you jump from Penryn-grade to being toe-to-toe with Ivy Bridge. That's without scaling clock speeds up, too. Manage an average clock increase of 10% per price bracket, and you're looking at a 33% leap from Phenom II.

Of course, anything that could work solely from L2/L1 would only gain around 10% performance per clock.

 

[ShintaiDK]

I´m sure it will be beaten by a 5Ghz 8 core IB at the same TDP

1100T is 125W at 3.2Ghz with 6 cores. And thats without the PCIe hub and iGPU that SB/IB got.

 

[dastral]

An improvement like that of Phenom to Phenom II wouldn't even be close to enough. I doubt they're going to pull a 42% clock speed increase out of Bulldozer like they did with Phenom.

Well a few corrections, a few tweaks, 22nm and i don't think it's too unrealistic.
I don't expect BD2 to beat Intel, but MAYBE being competitive on "decently threaded apps".

BD will always be horrible at games (specially on poorly threaded ones)....

 

[Lonbjerg]

We have no id how hasswell will do, not even a tiny clue. Haswell is just as bound to the 22nm physics as IvyB is. (haswell might even descrease performance at the same frequencies without special software optimizations)

Next issue: the test is ran on equal frequencies which is not the design goal of intel nor AMD to do that. So factoring out 1 of the key point of performance is sure to lead to misleading performance figures.

That said, bulldozer isn't even competitive to sandybridge, ivyB just makes things works. If BD would have been 1-1,5 year earlier, they would have had a much better run vs nehalem and its derivatives.

Haswell is designed for Trigate 22nm from the get go.
Ivy Brigde isn't...it's a refresh of Sandy Brigde.

You cannot compare ticks and tocks like that.

 

[Medu]

Well a few corrections, a few tweaks, 22nm and i don't think it's too unrealistic.
I don't expect BD2 to beat Intel, but MAYBE being competitive on "decently threaded apps".

BD will always be horrible at games (specially on poorly threaded ones)....

GF's 22nm? I wouldn't be surprised if the oil money behind them dried up before GF reach 22nm(I believe TSMC are going straight to 20nm). Even if they do reach it it's likely to be 2014/15 and it's highly questionable whether AMD will be around at that stage.

 

[Abwx]

Haswell is designed for Trigate 22nm from the get go.
Ivy Brigde isn't...it's a refresh of Sandy Brigde.

That s non sense.

IB was designed for 22nm , otherwise it wouldnt be more
power efficient than SB.

There s no way that a CPU could be half optimized for a given
process since it would not even reach the previous node perfs.

Layout has been deeply reworked to be put in adequation
with trigate fets geometry and electrical caracteristics.