Intel vs AMD benchmarks

[Titan]

I haven't been on AT in a long, long time but I am really annoyed right now and looking for some simple answers to this one question. How does the new AMD Bulldozer CPU lineup compare to Intel's stuff? I have been googling and browsing AT and Tom's looking for answers and Tom's desktop CPU charts haven't been updated since 2010. I am really starting to smell conspiracy here wondering if either AMD has bought off some publications, or if they nurture a fanboyish hope of AMD's product succeeding where it isn't. Youtube is filled with videos of complete idiots who know nothing about CPU architecture or performance, the highest rated videos are crap, or in German. I don't speak German being a unilingual American, but I am tempted to learn it now if the IQ of my country has dropped 20 or so points in the past couple years where I haven't paid attention to computers.

Ok, I am apparently in rant mode, sorry about that. I think it's going to get worse in a minute.

The core of my question is simply, how does each CPU stack up in synthetic benchmarks? How do they fare in integer versus floating point calculations . I would like to see a chart of single core performance so I can scale it up myself to multi-cores, or better yet a review that does this. Give me a solid benchmark so I can figure out price/performance myself, since prices change all the time. As a bonus, I would like to know the pipeline length of each CPU.

Years ago, I checked Tom's every day for decent articles and had their widget on my iGoogle page. I had to take it down as the writing became crap and I stopped going there. I don't even like going there now looking for answers but I am desperate. Anandtech, on the other hand, usually makes me feel like I am in college again in my computer architecture class. I am a software engineer, but I studied and understand CPU design in general terms. I have been looking on AT, and not seeing much in the way of comparisons here.

I caught an article when Bulldozer first came out, and the general idea was that this product newer than Intel's design doesn't stack up clock for clock. The low price point may make it worth it. But I am having a hard time figuring out which CPU is a sweet spot for price/performance. A few years ago I upgraded my dual socket opteton system to an i3 Clarkdale and was satisfied, also saving about 15 bucks a month on my power bill (mainly because server boards love to eat juice). I was down to two cores and hyper threading at 3ghz and doing all things fine.

Ok, I guess I don't have a lot of points here. I am really looking for an objective view from someone who knows what they are talking about and not a fanboy. Sorry I haven't searched a lot yet here, I am just fed up with the rest of the internet.

I think I am still in ass kicking mode so get ready for a rant punishing fanboys. I have just seen too much of this and need to vent.

Worshiping a brand is stupid. It's idiotic. Any company can make crap regardless of their brand. Pure and simple. Saying (AMD/Intel) is the best and always has been equals saying nothing.

I give credit where it is due. I give credit to AMD for not going out of business in a recession after buying ATI and having no money left; I give them credit for having a nice, low price point. I give credit to Intel for eventually getting it's head out of it's ass and making a good performance/clock CPU starting with Core 2, which was based off of the Israeli designed Pentium-M.

A brief history lesson, from about the year 2000, Intel made a series of bad decisions as the CPU market leader. In the early days of P4, the CPU was slower than the P3, and they were locked into a contract with Rambus to only use expensive RDRAM on their platform. Meanwhile, AMD benefited from the new, cheaper, and easier to produce DDR memory. Intel had to catch up later to switch to DDR. AMD's CPUs were better clock for clock, so they were the first to start labeling the CPU with something other than clock speed, and came out with the performance number. Eventually, Intel went this route. After the Itanium failure, Intel was reluctant to do what had always been done in x86 CPUs - widen the registers again, this time to 64 bit. AMD led the way, launching some nicer server chips called Opteron so we could get more than 4GB of ram. Intel had to copy this. Intel kept pushing for more ghz with prescott having a 32 stage pipeline over northwood's 20 stage, reaching 3.8ghz at a million watts and less clock-for-clock performance. AMD first decided to integrate the memory controller on the CPU die to improve overall system performance, and eventually Intel caught up. Intel pushed themselves to the wall like idiots, paying the price to AMD along the way. I'm not sure who was first in the dual core thing, but you get the idea.

But those days are over. The romance is long since done. Has been for several years now. Intel still has a virtual monopoly on the desktop CPU market, and with that much money they managed to not suck after years of bad decisions. The Core 2 was better clock-for-clock and low power consuming becoming an overclocker's wet dream. They are always ahead of the curve in manufacturing process so it always seems that at any given time, they have the best OC chip. Glad that Bulldozer is at least 32nm.

In the AMD glory days, there were nothing but Intel fanboys trashing AMD chips for catching on fire. But AMD had the better product and it costed less, period. Intel stood by saying it had more ghz. Now the shoe seems to be on the other foot. AMD has something with more ghz (breaking all sorts of OC speed records measured in ghz!) while intel's product is better clock for clock. All I have been hearing is a lot of loud AMD fanboys and a lot of hype about bulldozer and it seems that it did not deliver. I can't tell by finding any benchmarks over the yelling of "more cores!"

AMD has more cores. I am not so dumb as to believe an 8-core CPU is twice as fast as a 4 core CPU. I understand real world applications barely use four on a good day. The geek idiot in me does drool over having 12 or 16 cores, but then I catch myself and realize that the number of cores doesn't matter as much as performance per core, especially when we are talking drastically different architectures. I have heard claims that 6 intel cores beats out a 12 core amd chip on some things. The benchmarks are hard to find. If this is true, it is huge news that could be shrouded by the noise of marketing.


Cliffs:

So this is a very long rant by a man enraged by his own inability to find benchmarks and make any sense out of the CPU world. Maybe I just suck.

 

[Jovec]

inability to find benchmarks and make any sense out of the CPU world.

AnandTech's Bench works very well

Intel's Low End*
BD 8150 vs 2100 - http://www.anandtech.com/bench/Product/434?vs=289

Intel's "Sweet Spot"*
BD 8150 vs 2500k - http://www.anandtech.com/bench/Product/434?vs=288
BD 8150 vs 2600k - http://www.anandtech.com/bench/Product/434?vs=287

Intel's High End*
BD 8150 vs 3960x - http://www.anandtech.com/bench/Product/434?vs=443

*Roughly speaking

 

[grkM3]

What is your budget? and is there a microcenter near you?

A new sandy bridge cpu will out do any AMD cpu at the moment,Im talking the whole range from low end to top end so if you want a game system a 2500k is your number 1 bet.

What benches do you need?its pretty simple,amd spent 6 years on BD and when it came out it was slower than there older cpus,its the P4 version on the amd platform.

just think of the sandy bridge cpus as the athlon 64s

Also I paid 900 bucks for my FX opteron at newegg...when AMD had the performance it was just as bad as intel was with pricing.

 

[Maximilian]

http://www.anandtech.com/bench/Product/288?vs=434

Pick yur CPU and go to compare them.

 

[Titan]

Thanks for the bench tool link. Either I am blind, or it isn't easy to see in the CPU category. My one question, is any one of those benchmarks (such as AES encryption) a single threaded bench? Honestly, I would like a couple (one for floats, another for integers). As a programmer, I don't always multi-thread stuff and sometimes lazily like to brute force things. This is a nice tool in that it lists ghz with the CPU info, could they also list number of cores? It would be handy.

No budget yet, Honestly I am just trying to shop without the noise of marketing.

Thanks guys.

 

[grkM3]

before you buy anything wait for intel 22nm ivybridge cpus to start hitting review sites.

 

[Titan]

OK, apparently I am blind. Must be the rage.

Seriously though, I miss the days where my CPU or video card name wasn't masked under a marketing code that doesn't really mean much. I know things are more complicated now. I used to like watching the technology makers duke it out with raw skill and science. Now I am tuned out because marketing finally showed up there and it's a whole lot more to wade through.

The reason I came here is because I know I should have came here first. Thanks for still being there, AT!

 

[Maximilian]

Some of them are single threaded, im not 100% sure which.

Cinebench is one people often quote when showing single threaded performance. To find out the number of cores googling the chip and following the link to the first intel.ark.com result is the easiest way. AMD havent got a site like that that's as good but their naming is a bit more sensible. phenom II X6 - 6 cores, phenom II X4 - 4 cores, FX 8150 - 8 cores, FX 6100 - 6 cores.

 

[Titan]

Ok, moving the discussion along, an 8150, 3.6 ghz 8 core CPU is still $250. An i5 2500k, which is 3.3 ghz and only four cores, basically beats the pants off the 8150 while costing 30-40 bucks less.

http://www.anandtech.com/bench/Product/434?vs=288

Fair question: do the multi-threaded benches use all 8 cores? Because if they do, this is really sad.

Looks like a 20-50% performance increase clock-for-clock single core in Intel's favor. That's saying something about architecture difference.

 

[degibson]

http://www.spec.org/cgi-bin/osgresults?conf=cpu2006

 

[Ventanni]

Anand's bench tool is definitely the place to go for comparing CPUs. The only disadvantage to it is that not all CPUs are tested using the same benchmarks, so some comparisons may provide more or less benchmark information (like comparing a Core2 to Bulldozer may yield fewer results than comparing a Bulldozer to a 2600k).

In general though, the benchmarks reinforce the overall summary of Bulldozer chips:

-They're power hogs in comparison to Sandy Bridge chips.
-Their single thread performance is subpar even compared to Phenom II performance, and especially compared to Sandy Bridge chips.
-They do perform relatively quite well in multi-threaded threaded applications. They also scale very well in multi-threaded applications.

My firm belief is that like Phenom I chips, Piledriver will be what we hoped Bulldozer was. AMD has a tendency to release subpar first generation chips, but their second generation are usually pretty good.

 

[pelov]

Ok, moving the discussion along, an 8150, 3.6 ghz 8 core CPU is still $250. An i5 2500k, which is 3.3 ghz and only four cores, basically beats the pants off the 8150 while costing 30-40 bucks less.

http://www.anandtech.com/bench/Product/434?vs=288

Fair question: do the multi-threaded benches use all 8 cores? Because if they do, this is really sad.

Looks like a 20-50% performance increase clock-for-clock single core in Intel's favor. That's saying something about architecture difference.

It depends on the benchmark. Multi-threaded Cinebench 11 uses up to 40(? or was it 64?) cores. Because of the IPC gap, the benchmarks where the 2500K and 8150 perform roughly the same it's safe to say that all 4 modules are filled in the 8150.

That 20-50% gap is quite massive and affects even the heavily-threaded benchmarks. Where the BD chip does win it doesn't win by much and it only wins because it's got twice the amount of threads of the 2500K.

It's a server CPU

 

[bononos]

I..... Cliffs:

So this is a very long rant by a man enraged by his own inability to find benchmarks and make any sense out of the CPU world. Maybe I just suck.

The cliffs for the current situation is that the Bulldozer is considerably less efficient per clock than Intel and consumes more power/heat at competing performance levels. And the recent Windows patch to improve BD core scheduling only gave it a very very small(low single digit) improvement in benchmarks.

BD doesn't completely suck since its generally faster (15% according to AT) than the PhenomII since its clocked higher than its predecessor.

Alot of the fanboyism and negative publicity was self inflicted by AMD after it misled the public into believing that it was much faster than Bloomfields. The expectation of a generational leap ahead of the PhenomII (and minor but significant advantage over the Sandy's) led to the crushing disappointment and extreme defensiveness over BD. This is coming from a fan of AMD since their K5 days.

 

[AtenRa]

Have a look in the link on my signature bellow, i believe you'll find it helpful

 

[grkM3]

Have a look in the link on my signature bellow, i believe you'll find it helpful

Your test has a major flaw its comparing a quad cpu to a 6 core cpu and to an 8 core cpu.if you were to use a 6-8 core intel it would of had a higher scaling because you only have 4 cores.

Thanks for the review tho.it shows how well a quad core cpu can compete with a 6 and 8 core.can you redo that test with an 8 core sandy xeon with quad channel memeory?

Im curious to see how intel would do with 8 threads using hyperthreading

 

[blckgrffn]

Your test has a major flaw its comparing a quad cpu to a 6 core cpu and to an 8 core cpu.if you were to use a 6-8 core intel it would of had a higher scaling because you only have 4 cores.

Thanks for the review tho.it shows how well a quad core cpu can compete with a 6 and 8 core.can you redo that test with an 8 core sandy xeon with quad channel memeory?

Im curious to see how intel would do with 8 threads using hyperthreading

I think the point was 8 thread Intel vs 8 thread AMD. It is a thread scaling exercise, after all.

An eight core Xeon is likely to scale like a 2600k to four threads, continue that until eight, then have similar scaling from threads 9-16 compared to threads 5-8 on the 2600k. And well is should, given the price premium.

OP, I believe your rage is well founded. I think, as others have pointed out, the x86 CPU race is simply boring and doesn't pull in the page views like some other topics (cell phones? jeeze...) and so gets less coverage as a result.

For that lazy brute force programmer in us all, Intel continues to deliver

 

[PG]

A Pentium G620 edges out the FX-8150 in single threaded performance: http://www.legitreviews.com/article/1741/11/

Tom's cannot recommend any AMD cpu at any price point:
http://www.tomshardware.com/reviews/gaming-fx-pentium-apu-benchmark,3120.html

"With the sub-$100 Pentiums performing so well, Intel's $125 Core i3-2100 easily beats more expensive Phenom II and FX models. And the $190 Core i5-2400 dominates the sub-$200 landscape without challenge, really. As such, we're almost-shockingly left without an AMD CPU to recommend at any price point."


I see Bulldozer as like the Phenom I. Wait for the second go around and see how it performs.

 

[grkM3]

I think the point was 8 thread Intel vs 8 thread AMD. It is a thread scaling exercise, after all.

An eight core Xeon is likely to scale like a 2600k to four threads, continue that until eight, then have similar scaling from threads 9-16 compared to threads 5-8 on the 2600k. And well is should, given the price premium.

OP, I believe your rage is well founded. I think, as others have pointed out, the x86 CPU race is simply boring and doesn't pull in the page views like some other topics (cell phones? jeeze...) and so gets less coverage as a result.

For that lazy brute force programmer in us all, Intel continues to deliver

I think everyone knows that hyper threading is no where near the full gains you would get with an actual core.

 

[blckgrffn]

I think everyone knows that hyper threading is no where near the full gains you would get with an actual core.

Given that CMT is another variation of the way HT extends the resources of a CPU, contrasting their scaling abilities is the interesting part of the research done.

ie, CMT is very close to the scaling you'd get with full core, HT much less so. You can see where AMD found themselves in a pickle when it came to marketing, as saying four cores with CMT would have simply registered was four cores with HT for many. As you say, most aren't that impressed with HT, although it is certainly better than it was in the P4 days and does bring tangible gains in many scenarios.

 

[sm625]

AMD has basically gone 3 years without any performance improvements. You can literally take a 3 year old cpu and compare it to their newest stuff and not lose in 1-4 threaded benchmarks. Overclocked or not.

And the really sad thing is that their 3 year old stuff isnt much faster than their 6 year old stuff. They've just been stagnant for so long it is a wonder they arent chapter 9.

 

[Denithor]

One thing to note is the '8' core Bulldozer consists of 4 true/full cores and 4 neutered cores that are only helpful with certain kinds of tasks.

It's kinda like the ATI versus nVidia architecture difference all over again. nVidia has X number of full-fledged cores while ATI has 3-4 times the number of cores but only 1 in 5 is a 'full' core - the others kick in if certain conditions are met in the workload.

 

[dealcorn]

I would like to see a chart of single core performance so I can scale it up myself to multi-cores, or better yet a review that does this. Give me a solid benchmark so I can figure out price/performance myself, since prices change all the time. As a bonus, I would like to know the pipeline length of each CPU.

I do not get the part where you think the results you get by scaling up single core performance have real meaning. As soon as you scale single core you may or may not run into issues with memory bandwidth, I/O bandwidth and substantial helpings of god knows what. Multicore is ubiquitous. Thats the way cpu's are sold and how they should be measured if your interest is in purchase.

 

[blckgrffn]

One thing to note is the '8' core Bulldozer consists of 4 true/full cores and 4 neutered cores that are only helpful with certain kinds of tasks.

It's kinda like the ATI versus nVidia architecture difference all over again. nVidia has X number of full-fledged cores while ATI has 3-4 times the number of cores but only 1 in 5 is a 'full' core - the others kick in if certain conditions are met in the workload.

Hmm... I am not sure that is really accurate but I suppose it roughly equates to the situation at hand...

All the cores in BD are neutered equally, it's just when they have to share less they can do a bit better Also, when the threads are scheduled right Turbo is much more effective (which is to say, they are forced to share a FPU.)

 

[AtenRa]

Your test has a major flaw its comparing a quad cpu to a 6 core cpu and to an 8 core cpu.if you were to use a 6-8 core intel it would of had a higher scaling because you only have 4 cores.

Thanks for the review tho.it shows how well a quad core cpu can compete with a 6 and 8 core.can you redo that test with an 8 core sandy xeon with quad channel memeory?

Im curious to see how intel would do with 8 threads using hyperthreading

The review was about Thread scaling between CMP, SMT and CMT. There is no flaw in the thread count, Phenom 6 threads CMP vs BD 8 threads CMT vs Intel 8 threads SMT(HT).

CMP has the better scaling followed by CMT and with HT having the lowest.

Even if you have an Intel 12 Threads CPU (6 cores + HT) you will not get more than 30% from the HT. So if a quad core has a 400% scaling with 4 threads a 6 core will have close to 600%.

Now add the HT scaling (30% max) and you have 400+30% = 520% max for 8 threads or 600+30% = 780% max for the 12 Threads Intel CPU. That is the maximum theoretical SMT scaling with 12 Threads.

 

[Edrick]

I have read almost all comparisons I could find. But I have not see one done with F@H. If I were to build a dedicated F@H rig, would the FX-8120 be better than a 2600?