Introduction
There have been a lot of products
and inventions through the years that have been misunderstood and not
given the credit they deserve. Looking as far back as the invention of
the phonograph, even it was looked at as simply a quaint oddity. A
parlor toy, if you will. At the time, most people’s exposure to music in
the home was singing around their own piano, or simply singing in the
shower. Then came the phonograph. It made recording voices and music
possible, and with the advent from wax cylinders to reproducible media,
it made it distributable. Think about something, at the time of the
invention there were no such things as recording artists. Thomas Edison
had to create the entire beginnings of the recording industry in order
to support his parlor toy. We all see where that has gone….now turn off
the iPod and listen to me.
At some point (or points), a product comes along that changes an
industry, or in the case of the simple architecture of the phonograph,
creates one. In technology, there are constant creations and advances,
but oddly, the actual architecture of the CPU hasn’t really changed all
that much. The benchmarks we were using ten years ago are, for the most
part, still applicable as performance measurements today. This is simply
because the way CPUs operate hasn’t changed all that much. But how is
this possible in a world of ever changing software looking to grow more
complex all the time? At what point does the architecture of the CPU
need to change in order to keep up with the software and give the
software a chance to grow?
AMD’s new FX-8150 is the first release in their new Bulldozer line,
featuring an all new and radically different architecture. The AMD
FX-8150 is the first eight core desktop processor to market, but there
is much more going on beneath the skin. The eight cores of the FX-8150
are built into four modules on the chip, each with two physical
processing cores. These two cores share resources in a way previously
unseen, sharing not only L2 cache, but FPUs, as well. This makes for a
highly efficient unit capable of incredible multi-threaded capabilities.
In addition, the FX-8150 is a Black Edition CPU with an unlocked
multiplier. This obviously makes for easier overclocking and the FX-8150
is designed with extreme overclocking headroom in mind. Recently, AMD
set a world overclocking record with a Bulldozer CPU with the OC topping
out at an astounding 8.4GHz. This, naturally, is not what should be
expected from an ordinary home build, but it does show that the FX-8150
has extreme overclocking potential.
Technologies
AMD FX Series Processors
Seamless integration is AMD’s approach as they introduce the latest
technology and offer a flexible upgrade path towards the new AMD FX
platform. The award winning AMD 6000 series GPUs debuted in Q4 2010,
delivering top of the line DirectX11 performance and unparalleled
multi-display gaming capabilities to desktop PCs. The 9-series chipset
was released soon after, in Q2 2011, laying the groundwork for the AM3+
infrastructure while offering full compatibility with the current crop
of AM3 processors. The third part of the puzzle completing the new FX
platform is the AMD FX processor itself and it has finally arrived.
Unlocked and highly overclockable, these new FX processors feature the
world’s first octo-core desktop processor and AMD microarchitecture
developed from the ground up. Combined with a 9-series motherboard,
the new AM3+ platform supports increased ILDT current for higher
frequency HT link up to 5.2 GT/s. The AM3+ socket infrastructure also
adds support for CPU voltage load line and increased DRAM current, now
supporting up to DDR3-1866. Each 9-series chipset is CrossFireX ready
with two PCIe x16 lanes, supporting the latest 6000 series Radeon GPUs,
while also natively compatible with NVIDIA’s SLI technology.
The Bulldozer architecture focuses on efficient resource sharing to
deliver well-balanced performance and power consumption on
multi-threaded applications. Each Bulldozer core is armed with two
independent integer clusters each with a dedicated L1 data cache
(essentially two physical cores in one Bulldozer core) and modularly,
shares L2 cache, Floating Point Scheduler, as well as two 128-bit FPUs
supporting up to 256-bit floating point execution. Sharing resources
reduces not only the power consumption but also the die space, therefore
lowering the cost.
Bulldozer has a deeper pipeline that relies on improved branch
prediction and pre-fetchers, aiming to resolve the bottleneck caused by
incorrect branch prediction. Unlike the previous architecture, the
predict and fetch pipelines are decoupled. A queue of future
fetch-addresses is created by the predictor and allows the fetch logic
to go through this queue and compare it to what's in the instruction
cache. Each Bulldozer module is capable of decoding up to four
instructions per cycle, compared to three on AMD Phenom II processors.
Another major change that Bulldozer brings to the table is Power
Gating, when paired with a 900-Series Chipset. Each module can be
clocked and power gated independently, allowing unused cores to be
powered off while the other active cores can be driven up in frequency.
This new Turbo Core technology, enhanced for the new FX processors, has a
new mode which can run ALL cores for a time in Turbo mode where there
is sufficient TDP headroom. There is also a MAX Frequency mode where the
FX processor can increase the frequency on half the cores and can
remain in a higher frequency state longer compared to previous Phenom II
processors, resulting in improved performance in lightly threaded
applications.
Built on the latest 32nm silicon on insulator technology,
Bulldozer-based chips can have up to 8 cores (each core is seen by the
Operating System as a logical processor) comprised of four Bulldozer
modules sharing L3 Cache and NB resources between them. Bulldozer is
also designed to dynamically switch between shared and dedicated
components to maximize efficiency, while featuring new x86 instruction
sets, like SSE4.1, SSE4.2, AVX, and XOP including 4-operand FMAC
allowing for greater performance and flexibility. These new instruction
sets will find use in the next generation of applications that will play
a prominent role in the near future. While some new instructions, like
SSE, AESNI and AVX, useful for video encoding, data encryption and FP
intensive apps, are both supported by the latest Intel and AMD
processors, the FX series processors are uniquely capable of processing
FMA4 and XOP instructions as well, useful for HPC and numeric
applications.
Putting it all Together
The much anticipated launch of Bulldozer has been long coming. There
have been small pieces of information available here and there, some
confirmed, some merely speculation. One thing that AMD never hesitated
to share was that Bulldozer’s Zambezi core was going to be an entirely
new architecture, unlike anything we had seen previously. That is
certainly the case. Eight cores in four modules, with shared resources
in each module for two cores and a boatload of L2 and L3 cache makes for
an impressive looking new architecture.
There is also a new approach to the Turbo Core in the FX-8150 and the
whole Bulldozer lineup. The FX-8150 sports a stock clock of 3.6GHz.
When power headroom allows, all eight cores will ramp up to 3.9GHz under
load. That is pretty standard. Where it gets interesting is when only
four threads are required. In this case, the FX-8150 will address those
four threads with four cores boosted to 4.2GHz provided there is
sufficient voltage headroom. That is a big boost with no overclock
involved, and the amount of boost can be altered for either state in
AOD, should you decide to do so, allowing for what is essentially an
overclock when necessary, but a cool running stock chip when the power
isn’t needed.
At no point did AMD anoint Bulldozer as a Sandy Bridge killer. Even
now, they are quick to point out certain things that it is more capable
of, due to its unique architecture, but also are willing to admit where
it falls behind a little, due to the same. The question becomes, what
are the strong points of this new architecture, and where does it seem
to fall short? Another question being, is it the architecture that falls
short in some respects, or is it simply more advanced than the software
that has perceived lags in?
Breaking down what you are getting into with the AMD FX-8150 is a
little more complex due to this. As a result, the platform has enormous
strengths and some minor weaknesses. However, the weaknesses seem to be
amplified by how well it hits its strengths, but there is no area where
you can call Bulldozer an underperformer. Like I said, Bulldozer is a
complicated chip in more than its architecture.
We can talk about synthetic benchmarks, and all the GOPS you are
getting, or your memory bandwidth, but that really isn’t what most
people are looking to know. Most people don’t run Sandra and AIDA
benchmarks for entertainment (or work) purposes, so let’s focus on
applications you may actually be using, or benchmarks that simulate real
world usage. Besides, if you are running Sandra benchmarks for
entertainment, you don’t need an explanation here.
The FX-8150 has eight cores….and it likes to use them. During testing
the FX-8150 produced some absolutely jaw dropping results in
multi-threaded applications. Rendering in Cinebench actually stopped me
in my tracks during the test with the speed I saw, and the benchmark
numbers certainly back that up. All eight cores worked quickly and
efficiently in unison for an incredibly fast 3D render, and the final
scores certainly backed that up.
Multi-threaded video applications, such as Handbrake and x264, were
absolutely no problem, as the FX-8150 put up outstanding transcoding
numbers in both. The FX-8150 split tests in x264 with the Sandy Bridge
processors. In Handbrake 0.9.5, the FX-8150 did what really matters – it
cleanly transcoded a DVD mpg to mp4 faster than either the 2500K or
2600K at stock speeds, and scaled appropriately faster when overclocked.
All in all, the AMD FX-8150 is an incredibly powerful CPU for
multimedia creation, and using today’s multi-threaded apps only adds to
its power.
In our WinRAR tests, the FX-8150 was able to really shine. This is a
multi-threaded productivity app and the FX-8150 took full advantage.
Showing that it’s not just an entertainment piece, the FX-8150 put up
numbers on par with the 2600K in compression with even better throughput
at stock clocks.
Gaming with the FX-8150 is absolutely solid and showed an improvement
over the Phenom II and Thuban in almost all cases. Again, the Bulldozer
CPU seemed to show an aptitude for more complex and CPU intensive
games. Metro 2033 saw almost a 25% performance increase over a Phenom II
980 using the same video card. Real world mimicking benchmarks, like
3DMark 11, saw a huge jump in physics computation, and Unigine and Stone
Giant displayed beautiful rendering. One immediately noticeable, but
not measured, trait I saw was in S.T.A.L.K.E.R. CoP. It has always been a
sticking point that vegetation was seemingly rendered only as you got
right on top of it. The FX-8150 rendered the scene significantly further
into the distance, making for a much more realistic playing field.
It would seem that the FX-8150 has only one area where it may not
meet expectations, and that is with older and single threaded software.
Clock for clock, the FX-8150 shows minor improvement over the Phenom
II/Thuban, but not the huge leaps we see in multi-threaded apps. Single
threaded apps run very capably, but there are no mind boggling numbers,
nor are there any performance issues of concern. As I stated previously,
it is more of a situation where “runs single threaded apps very well”
seems to pale next to its stellar multi-threaded performance. As apps
become more complex, the FX-8150 will not only be ready for them, it
will be eagerly awaiting them.
Reviewer's Opinion
It’s been a long wait for Bulldozer’s arrival, with the date being
pushed back again and again. Naturally, that had me going in looking for
the flaw that kept it from coming to market on any of the previous
announced release dates. Was the architecture not yet ready? Were there
microcode issues? Did they inexplicably catch fire when exposed to lemon
meringue pies? Who knows, it was something, and after playing with the
chip, my gut hunch is that “something” was a question of the right way
to market it.
Bulldozer is new architecture, all new from the ground up. There
isn’t another processor out there like it. Usually, there is something
on the same page around, but this time there isn’t even anything in the
same book store. It’s different and it is incredibly forward thinking.
However, with different also comes a different type of performance, and
the FX-8150 certainly has anything but traditional performance
characteristics. Looking at synthetic benchmarks, they were all over the
board in ways that made very little corresponding sense, so it wasn’t
until I got to use the processor in actual applications that I got a
feel for what it was all about.
The AMD FX-8150 absolutely excels in a multi-threaded environment. It
seems the more you throw at it, the better it responds. Look at the
WinRAR test, this was always a place that Intel came up big. It uses
every available thread, and pushes them. Surprisingly, the FX-8150
absolutely chewed it up and spit it out, posting numbers better than
either of its Sandy Bridge competitors. It’s been a long time since AMD
has held that distinction.
Multi-threaded video encoding was again a big winner for the FX-8150.
It ripped through videos in a flash. The x264 and Handbrake numbers
speak for themselves. However, even more impressive was actually sitting
with it working in Adobe Premiere. The snappy feeling during editing
was outstanding, adding filters on the fly was flawless and when it came
time for conversions or transcoding, the FX-8150 whipped through it
like nothing. The feeling and speed was drastically different, and for
the better.
Gaming was satisfying, and sometimes spectacular. Again, it seemed
the more complex the game, the more outstanding the performance became.
With all of these, it was interesting watching Turbo Core work. With a
CPU monitor set up, you could see the speed and voltage boosts at
appropriate times. Though Turbo Core is not a new idea, the two step
Turbo mode is. Only need four threads? Well, four threads at 4.2GHz all
being used is much better than eight at 3.6GHz, with four sitting idle. A
bit like an instant OC, with flexibility added to allow you to tweak
the boost.
Where the FX-8150 doesn’t make improvements in leaps and bounds is in
older and/or single threaded applications. It seems curious, but it
seems to be a characteristic of the architecture. It doesn’t lose ground
here from previous generations, the improvements just aren’t as drastic
as they are in multi-threaded apps. I’m sure some are going to throw
out the numbers from these and point to them as an issue. But my feeling
is this, when was the last time you were typing a letter in Word 2003
and wished your CPU was faster? I am going to guess never. I sure
haven’t.
On to the fun stuff….overclocking was a breeze. Like all of the Black
Editions I have used, the FX-8150 much preferred a multiplier OC to a
Ref Clock boost. 4.9GHz was stable on air, with normal operating temps
staying within limits. Only once did it get dicey, and that was on a
prolonged eight thread video transcode, and even then temps peaked at 66
centigrade. That may sound a bit hot for an AMD, but we are talking 8
cores, 1.525v under load. It’s going to generate heat, no way around it.
OC performance scaled nicely, though to the naked eye it may not seem
as good as it actually is. You need to keep in mind that with stock
clocks of 3.6GHz, the Turbo Core allows the CPU to jump to 4.2GHz. So
much of the OC when under load is actually a bit smaller than the 1.3GHz
you assume.
The AMD FX-8150 is possibly the most advanced processor on the market
right now, and brings absolutely stunning performance to the latest and
greatest apps out there. Do older apps get the same type of stunning
upgrade? Well, no…but software is changing, and you have to make the
break somewhere. Trying to be a “one size fits all” really means that
“one size doesn’t fit anything correctly”.
The new AMD Bulldozer/Zambezi/Scorpius (call it what you will)
architecture is forward thinking. The FX-8150 delivers smoking
performance on the latest apps and games and, just as importantly, has
the promise to evolve. Right now, it’s time for developers to get the
software caught up to the FX-8150’s capabilities. It’s a top performer
now and looks to be for some time to come. You can look at all of the
out of context benchmarks you like, but in real world use, the AMD
FX-8150 is ready to Bulldoze a lot of sand out of its way. If you
appreciate the AMD FX-8150 for what it is, and what you are really going
to be doing with it, it was worth every second of the wait.
Pros:
- Incredible Multi Threaded Performance
- Excellent Price/Performance Ratio
- Smooth Working Turbo Core For Speed Boost
- Dual Turbo Modes For Increased Boost In Lightly Threaded Apps
- Maintains Excellent Temps At Stock Speeds
- Black Edition With Unlocked Multiplier
- Eight Physical Cores
- Easily Achieves High Overclocks
- Excellent OC Performance Scaling
- Easily Tweaked With AOD For Maximum Performance
Cons:
- Performance In Single Threaded Apps Not On Par With Multi-threaded
The memory bandwidth of the Core i7-3960X is surprisingly low given it
was tested with quad-channel DDR3-1866 memory. The read throughput
reached just 18.1GB/s, while the write performance was limited to
15.1GB/s. This made the Core i7-3960X roughly 9% slower than the Core
i5-2500K when measuring read performance and 18% slower for the write
performance.
DC Computer Solutions 
“Big data” analytics, security and cloud computing will be three of the most significant drivers of technological change in 2012, according to Ovum.