AMD Bulldozer FX Processor Benchmarks Leaked

Subject: Processors | October 3, 2011 - 12:29 AM |
Tagged: sandy bridge, Intel, i7 2600k, FX 8150, FX, cpu, bulldozer, amd

Intel has held the performance lead for several processor generations now, and while AMD is still technically in the game for home theater PC and budget builds, many enthusiasts have moved to Intel for gaming and high performance computers. Many of those people have also held hope that the chip manufacturer would eventually come back strong and maintain some level of competition in the industry. As we move closer to AMD's Bulldozer launch (which seems to have been confirmed for October 12th), enthusiasts and reviewers alike are clamoring to answer a long awaited question: "will Bulldozer give Intel a run for its money?"

According to website Donanim Haber, enthusiasts’ high hopes may finally be realized. The site has posted several benchmarks results that indicate Bulldozer is not only cheaper than Sandy Bridge, but performs on par with Intel’s top end Sandy Bridge chips. In many tests, the AMD FX 8150 CPU’s eight core performance matches the multi-threaded (8 threads, 4 cores) performance of Intel’s high end Core i7 2600k processor.

View Full Size

In the benchmarks that the site performed, the AMD FX 8150 was tested against the Intel Core i7 980X for 1080p gaming and the Core i7 2500k and 2600k for multi-threaded performance. In the graph shown above, the AMD Bulldozer CPU was roughly on par with the i7 980X, trading wins in some games and providing a similar level of performance in others. The AMD processor won in the Metro 2033 and Lost Planet benchmarks, but was slightly slower in Civilization V and F1 2010. In AVP and Batman (among others), the two competing processors saw equal results.

View Full Size

They also ran several benchmarks using highly multi-threaded programs to take advantage of the many-core designs of the AMD and Intel processors, including WinRar 4, Handbrake, 7zip, and wPrime 32M. The eight core AMD FX 8150 Bulldozer processor was tested against both an Intel Core i5 2500k and a Core i7 2600k. The AMD CPU came out ahead in 7zip, wPrime 32M, and Bibble 5.0. It was slower than the Core i7 2600k in the WinRar 4 tests and slower than both the 2500k and 2600k in the ABBYY OCR10 benchmarks. In the other tests, the AMD processor kept pace with or was only slightly behind the top end Intel 2600k CPU.

From the leaked benchmarks (which you can read here), AMD’s new Bulldozer CPUs have made an admirable showing. Should these benchmarks hold true, Intel will have some serious competition on its hands, something that the company has not had to deal with in a long time. Whether Bulldozer will result in price cuts or ramped up production on the Intel side remains to be seen; however, the results are not going to be easy for Intel to ignore.

Stay tuned to PC Perspective for more Bulldozer news in the coming weeks.

Source: Donanimhaber
October 3, 2011 | 02:50 AM - Posted by Anonymous (not verified)

I think I know whats going in my next Linux box...

December 16, 2012 | 03:41 PM - Posted by Cathern (not verified)

Hi there matеѕ, how iѕ the wholе thing, and what уou deѕirе to say concernіng this aгticle,
in my view іts in fact amazing іn support of me.

My blog post ... link popularity service

October 3, 2011 | 03:39 AM - Posted by Anonymous (not verified)

with kernel 3.0

me too

wipe my Intel box with feces.
F*** you Intel after using for 15 years.
wouldnt mind buying qualcomm desktop chips if they ever come out but never Intel.

October 3, 2011 | 04:38 AM - Posted by n.n. (not verified)

lets asume this is not a fake (i personaly think it is) and bulldozer iz faster then the sandy bridge...but what about the sandy bridge-e? can bulldozer beat it..?! now theres something to think about

October 3, 2011 | 05:02 AM - Posted by Anonymous (not verified)

Anybody who buys an Intel extreme cpu either is very rich and doesn't mind wasting money, has a very small dick and buys something that they can brag about, or is just stupid. Why would you buy a 1000 dollar cpu when next year Intel will have a 250-300 dollar cpu that's faster than it. Talk about a waste of money. Example 990x vs 2600k. 2600k is actually faster and you can save 700 bucks. And 2600k came a year after 990x. Why do people even insist saying this all the time your ignorance is not welcome and it only makes you look like a fool that knows nothing and has no logical thinking. "Will bulldozer compete with sandy bridge e" hmmmm let me think about that one cpu costs 270 but the other costs a 1000. I don't know duhhhh. What a retarded statement to make. It is equivelent to "yea that subaru wrx is as fast as the Mitsubishi lancer but will that wrx beat a ferrari". Why would anyone say this. It would be fun to compare the 270 cpu against a 1000 cpu for shits and giggles but it isn't intended to go against sandy e. Btw the cheapest sandy e cpu @ 300 bucks looks to be slower than 2600k but has quad channel mem. Even the 1000 dollar sandy e cpu is steps behind the 2600k in some tests which is shitty. There has been leaked benchmarks

October 3, 2011 | 06:00 AM - Posted by zicoz (not verified)

Now all they need is high-end mATX boards.

October 3, 2011 | 06:01 AM - Posted by Mark (not verified)

All this tells me is that they wouldn't realease Bulldozer till they matched the 2500/2600 in almost everything.....which tells me they had to do ALOT of work just to get there.....

October 3, 2011 | 09:02 AM - Posted by JSL

For being almost a full year later than sandy bridge (even with the mobo chipset recall), an 8 core processor shouldn't be "on par" with a quad core with hyper threading wouldnt you think?

The second question is, how will they perform against Ivy Bridge with the new trigate process, or the 6 core variant of the Sandy Bridge E.
(only reputable sources for benches, not amd made "charts" like the ones "leaked" as they're as believeable as intel ones, and with detailed information how each cpu/mobo was set up)

Im neither an AMD or Intel fanboy, but it leaves me to question if AMD is simply trying to play catch up and is targeting current intel selection, or an early bid to gain some revenue before intel starts selling their next "tick"

October 3, 2011 | 11:40 AM - Posted by fuzznarf (not verified)

The "8 Core" is just marketing. Its really 4 core (or module). Each module has 1 integer unit, 1 floating point unit, one memory unit. and the ONLY reason they call it 8 core is because it has a shared extra floating point unit. So its not a true 8 core processor.

Intel uses hyper threading. AMD uses an extra floating point unit. Both both are quad cores (minus the marketing BS)

October 3, 2011 | 01:40 PM - Posted by Anonymous (not verified)

Your right to a point. Each module* has two integer cores and single 256-bit FP unit which can be split into two x 128-bit units. There are shared "front-end" components that feed each integer pipeline (core) as the L2 cache is shared as well.
* AMD should have called each "module" a flexible-core, or Core with hardware based threading -hardthreading technology and put 4C(cores) - 8T(threads) on a single MCM -just as they do now, but call it 2 cores x 4 modules = 8C-8T MCM.

October 3, 2011 | 08:14 PM - Posted by looncraz (not verified)

Wow, you REALLY don't understand what is in a module:

Each module contains all of the following:

2x Integer Schedulers
2x Integer Cores
1x FlexFPU capable of 2 concurrent threads at 128-bit
1x Prefetch
1x L1I Cache (64kb)
1x L2 cache (2048MB, mostly inclusive, non-victim)

This is actually only a "slightly" more integrated than the Core 2 Duo (which had shared L2 and some other rear-end items).

Bulldozer just takes the Core 2 Duo level of sharing to the next level, and includes much of the front-end as well.

By definition, a core is an execution unit that can operate independently, and concurrently, with other execution units. Bulldozer fits this definition perfectly for an 8-core CPU, except in 256-bit FPU:

Concurrent Operations of Bulldozer:

8x integer threads
8x 128-bit or less floating point threads
4x 256-bit floating point threads

Further, thanks to IPC design, in OPS, we have the following, per cycle, in ideal circumstances, CONCURRENTLY:

8*2 ALU ops
8*2 AGLU ops
8*4 32-bit flops
OR
8*2 64-bit flops
OR
8*1 128-bit flops
OR
4*1 256-bit AVX

The ONLY time Bulldozer isn't a true 8-core CPU is with AVX instructions. Then, 8 AVX threads will be fighting with the FPU resources of only 4 concurrent operations.

Good idea to be educated on a topic before posting...

--The loon

October 4, 2011 | 04:28 AM - Posted by Aharon (not verified)

Before making any education claims toward others, let me correct you as well:
A core is an independent unit that can process a given task ("process") and do not share any hardware units (resources) with other cores.
The moment some of the resources are shared, thus the 2 units (or more) cannot process independent data or instructions.
Farther more, once a shared memory/fetching mechanism (cached or registered) are shared, thus the memory bandwidth of the 2 alleged processing units is halved!
AMD's bold claim for 8 coress, is by definition a false claim - that might put them one day in the court.
AMD's bulldozer cannot process simultaneously 8 different process, it can process simultaneously only 4 process.
Sending the bulldozer cpu 8 process at once by the operating system, will cause a context switch - same as processing more then one task on a single core has always been.
before preaching to the others as being uneducated, I preach you to go to the university and master a degree in computer science and mathematics, something it seems you have no fulfilled so far.

Aharon Y – BA at Mathematics and Computer science.
Israel.

October 4, 2011 | 01:10 PM - Posted by Josh Walrath

I really doubt that AMD will be addressing the definition of a "core" in court. As technology changes, the boundaries of such definitions will inevitably change as well.

In the classical sense, you are correct in the definition of a core, but I don't think that it much applies anymore. Intel's Hyperthreading was a step in the direction, and AMD's implementation is a further advance (AMD claims they only inflated the die size about 5% by adding the second integer unit and beefing up the fetch, decode, and branch units).

You assertion that the two units cannot process independent data or instructions is somewhat misleading. Multiple instructions are always in flight and scheduled for the different cores. So in a true multi-threaded application, we will not see one of the integer executions units in each module idle while the other is working on new instructions. Same goes for the FPU/SIMD unit. It can handle 2 x 128 bit instructions at any one time, or 1 x 256. So even though it shares the same front end, instructions and data are always lined up and scheduled for the first unit to finish the previous workload.

This stuff is not really new. Chip designers have always worked on ways to cover up latencies by always having work available for the execution units. It does not pay to have a unit idle while waiting for data and instructions. So to avoid such stalls chip designers have aggressive prefetch and scheduling. It is not much of a leap from these types of architectures that were introduced 15 years ago to increase their throughput to keep two separate execution units at work in the same module, even though they share components.

October 5, 2011 | 07:22 PM - Posted by dew (not verified)

The front end fetch/decode are shared, but AMD has made them wider than in Phenom II. The memory bandwidth for the fetch unit is twice as wide as in Phenom II (256 bits vs. 128). The decoder has an extra general-purpose decoder. Also, the context switch occurs in a single clock cycle, which is as good as it gets. My source is this article: http://www.realworldtech.com/page.cfm?ArticleID=RWT082610181333&p=5

Before you get all elitist, actually read about the microarchitecture.

-Darin W. B.S. Computer Engineering

October 5, 2011 | 07:26 PM - Posted by dew (not verified)

Sorry, the bandwidth for the fetcher is the same, but the ITLB's are larger and it has better branch prediction and prefetching.

October 9, 2011 | 03:15 PM - Posted by Jimbo Jones (not verified)

I remember you - the guy who tried to sound more credible by stating that since Intel has an R&D facility near your home, you know more than everybody else ... once again I won't bother to finish reading your post ... ;-)

October 4, 2011 | 02:04 PM - Posted by Anonymous (not verified)

Each integer "execution unit" (an apt term) is clearly not inter-dependent on its neighboring "unit", however it is neither an island unto itself. Per AMD's own admission their approach will be a net gain of around 80% increase over a single BD "execution unit" (alone) with the same frontend and cache circuits -all for the cost of few percentage points of die space. To that end we'll see 90% execution unit performance x 2 (180%)over a single core which would be considered 100%. So adding the second "execution unit" does negatively effect performance of the first "execution unit". Execution unit performance is not 100% x 2 and therefore cannot be considered fully independent nor can they execute at 100% concurrently.

Furthermore, though not specified, my original observation was intended to be more of a marketing strategy as is implied with "AMD should have CALLED each" and not "AMD should have MADE each". This is mainly to curtail the Intel fans from declaring 1 intel core = 2 amd cores, while conveniently forgetting about the gains of hyperthreading.

Finally, I do hope AMD has some mechanism in place to allow running 4 threads on 4 separate modules allowing all the L2 cache to be utilized instead of overclocking half the cores and using half the cache on the processor. I can imagine a few workloads where the former would be preferable to the latter.

October 5, 2011 | 07:02 AM - Posted by Mayford5 (not verified)

This is really what it comes down to. Intel fanboy says"No there are not 8 cores because AMD lies and sucks" AMD Fanboy says " Yes it is 8 cores and we did it first on the desktop." Who really freakin cares whether it is 8 or 4 w/some sort of crappy HT. This should be a performance discussion which is only speculation at this time. So what if AMD says it's 8 core and it isn't. This argument sounds just as dumb as when AMD stated that Intel's first dual core wasn't a true dual core. I get sick of people who don't know jack getting in these discussions and make dumb statements on either side and don't show any schematics or anything to back up the stupid claims about the architecture.

Ok so now that my rant is over. I think these are another fake. I don't know if we will have true benchmarks until the enthusiasts get them and can give us some benchs for Toms and PCPER and other reputable sites. That being said I don't care who is faster, I just want true benchmarks so I can either buy an FX or wait for Sandy-e. I am due for an upgrade anyways.

October 3, 2011 | 09:14 AM - Posted by Imperfectlink

One may consider it to be a paltry showing for an 8-core processor to be matched up with a 4 core with HT but it isn't. The module system is AMD's answer to Intel using HT so a 4 module system is actually meant to compete with a 4-core 8-thread processor.

Consider though that AMD is working with a design that is typically lower in performance per clock so this compensation is called for.

I expect Sandy-E to outbench the FX for sure (since I do 3d rendering I'll probably buy Sandy-E) but that doesn't make it substandard. Allow me to put two FX processors into a system without charging enterprise pricing and I'm there.

October 3, 2011 | 09:38 AM - Posted by Anonymous (not verified)

It all comes down the price! At a good enough price point AMD could win.

October 3, 2011 | 10:04 AM - Posted by TheWrightMatt (not verified)

Great, marketing slides... I'm sure those wont be skewed in any way.

Chart #1: FX-8150 has a natural clockspeed advantage and the 980X is an old architecture (Gulftown which is based on Bloomfield that is over two years old). Would have been more fair to leave the 980X out of it but, they did it for a reason (cherry picking).

Chart #2: Why is everything normalized to 2500K? I'll tell you why, it makes little differences look bigger. Take with a leathal dose of salt.

More fun comments that aren't jerking off AMD check out here. Congrats AMD, you're releasing a product that is finally catching up with Intel, just as their about to launch a new line.

October 3, 2011 | 10:32 AM - Posted by JSL

wow, talk about a fanboy war over in that thread.

October 3, 2011 | 08:21 PM - Posted by looncraz (not verified)

You **ALWAYS** normalize results to the baseline comparator. If you are comparing your product to the i2500k, then you normalize your charts to that.

That is that the i2500k values will be 1.0, and ALL others will be some deviation from that.

--The loon

October 3, 2011 | 11:07 AM - Posted by Mark (not verified)

article is from the 24th, over a week old....

October 3, 2011 | 02:22 PM - Posted by sandybridgeE (not verified)

so?

bulldozer barrely manages to beat the entry level sandybridge. with the sandybridge E processors coming out in a few months loaded with 6 cores + HT, intel will easily extend its margin. Not to mention ivy bridge will roll around by the time CES his and further extend intel's lead. This shows AMD reamins a full generation behind of intel...yet again.

October 3, 2011 | 04:11 PM - Posted by Esbornia (not verified)

Full generation behind or not, for a gamer like me, bulldozer is much much better then SB. Cheaper, and trust me, I have a Phenom II x4 955, a Phenom II x6 1090t, an E8500 and a I7 2600K. Apart from 3D rendering untill now, the difference is not that big and Intel is way overated. CPUs apart, the best upgrade you can give your PC nowadays, if you have a decent CPU/ Video card, in terms of performance, is a SATA III SSD.

October 3, 2011 | 05:10 PM - Posted by Devlin (not verified)

Now this is getting funny its your 8 cores vs are 4 core with ht who cares if it's just as fast as the SB and cost me 50 dollars or more less who gives a crap about how many cores it has hell if it had 16 cores and sells for 270$ and still holds its on with SB i buy it its cheaper lol

October 3, 2011 | 07:52 PM - Posted by Anonymous (not verified)

Yeah I know who really cares if it has 8 cores technically anyways it is 4 cores. All anyone should care about is performance at the end of it all. I think Intel fan boys are just jealous that they don't have an 8 core cpu. But anyhow if intel did have an 8 core cpu and amd had a 4 core which was as fast as the intel 8 core then Intel fan boys would be saying the opposite of what they are now saying who cares about cores it is just as fast. Intel fan boys are funny and they entertain me at the least. I think I may buy a new 8150p later this month but I am going to wait to see actual benchmarks before buying. If it is as fast as 2600k and overclocks to nearly 5 ghz like 2600k then I will buy. If it isn't all hyped up to be what its suppose to be then I will ditch my x6 1090T for 2700k when it is released by Intel. I buy both depending on price/performance. But I will say one thing I like a lot about AMD is compatibility with sockets. That is something Intel should take note of.

October 5, 2011 | 07:07 AM - Posted by Mayford5 (not verified)

Thank you for being reasonable. First reasonable posts I have read in this whole thread.

October 4, 2011 | 01:40 AM - Posted by Tomek (not verified)

News:

1)Sandy Bridge-E processors has a Vt-D bug.

http://www.techpowerup.com/152978/Sandy-Bridge-E-VT-d-Broken-In-C1-Stepp...

2)Sandy Bridge-E has been pushed back to Q1 2012

http://www.techpowerup.com/148282/Sandy-Bridge-E-Delayed-to-January-2012...

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.