| AMD FX-8350 Vishera Desktop Processor |  |
| Reviews - Featured Reviews: Processors | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Written by David Ramsey | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Tuesday, 23 October 2012 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Page 13 of 15
AMD FX-8350 OverclockingLike Intel's Sandy Bridge, AMD's new Piledriver processors are fabricated on a 32nm process. This means they use less power and generate less heat than the older Phenom II CPUs, which were based on a 45nm process. Of course Intel has since moved on to a 22nm process with their Ivy Bridge CPUs, but enthusiast hopes for new overclocking records were dashed when testing showed that the Ivy Bridge CPUs overclocked no better than-- and in most cases not as well as-- Sandy Bridge. This leaves an opening for AMD, who has promised improved overclocking with this generation. Benchmark Reviews was able to overclock our original FX-8150 Bulldozer-architecture CPU to 4.8GHz. The Piledriver-based FX-8350 did even better, chewing through our benchmark suite at a dead stable 5GHz with the help of a Corsair H100 water cooling system. To reach this overclock I had to set the CPU voltage to 1.48V and adjust all of the power/VRM settings on the ASUS Crosshair V Formula motherboard to their maximum. CPU temperatures peaked at 58 degrees Celsius at an ambient temperature of 23 degrees. While many overclockers prefer to disable some cores to reach the maximum possible GHz, I prefer to leave all cores enabled as I did here. 
It's not clear how much overclocking headroom remains in the FX-8350 for exploitation by more exotic cooling methodologies. When the FX-8150 was introduced, AMD's reviewers guide gave a range of overclocking targets for air, water, and liquid nitrogen cooling; this time around, they simply said 5GHz should be possible for many processors using a water cooler, and it seems they were right. I was able to boot into Windows at 5.1GHz but most benchmarks would crash after a short time. Intel's newer chipsets derive most system clocks from the base clock, so the classic overclocking mechanism of raising the base clock no longer works, since doing so raises every other clock in the system (memory, PCI-E, etc.). This typically means that you can raise the base clock by 3-5MHz before encountering crashes, leaving adjusting the multiplier as the single overclocking mechanism. AMD 990FX systems, on the other hand, are built "the old fashioned way", and so raising the HyperTransport clock remains a viable option for those sub-50MHz increments if you're working to get the last few MHz. The chart below shows how overclocking affected the 8350's performance on each benchmark, with the stock clocked benchmark score normalized to 1.0 and the overclocked benchmark score represented as how much faster it was than the base clocked benchmark.
The FX-8350 can boost its clock speed to 4.2GHz under normal use, and my 5GHz overclock represents a 19% boost, so a 14% performance improvement isn't too bad although I'd hoped for some more.
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Comments
the threading and core parking updates can make a bit of a difference to benchmarks especially in lightly threaded situations
The extra performance is not Earth shattering, but it is good enough to consider buying into. I will not be selling my i7-2600k system anytime soon, but I can see an upgrade to this new FX-8350 happening within a few months.
Thanks for the review David.
When we review motherboards, processors, and the like, test systems are built, a fresh copy of Win 7 is installed, and we're good to go. Since Windows 7 will run for 30 days without requiring an activation code, it's easy to create fresh installations as needed. After the review is done, test systems are generally torn back down and the parts saved for future tests/comparisons.
Windows 8 requires a valid activation code just to install it-- you can't skip this step as you can with Win 7. Then it's locked down to the system it was installed on.
Right now that means we'd have to buy a new copy of Windows 8 for every video card, motherboard, or CPU we test. This isn't viable, and we're trying to figure out what to do about it.
'cause I just looked at their review and they used a different motherboard, different memory, different hard drive, and different video card, as well as an almost completely different set of benchmarks.
The only two benchmarks we have in common are PCMark 7 Pro and Handbrake. You can't compare the Handbrake results since we used different input files with different settings.
PCMark 7 Pro tests the entire system, not just the processor (as I specifically mention in my review), so it's not surprising that with different hardware they get different results. Our Creativity scores were virtually equal (less than 0.5% difference); our Computation score was much better (6468 vs. 5083) and our Productivity score was less.
Which proves precisely nothing, except that you have no idea whatsoever what you're talking about.
it's silly for you to make claims without a single modicum of proof to support the argument. As the author points out, you can't compare completely different computer systems and expect the same benchmark score. Perhaps these in-depth technology reviews require a level of understanding you lack.
SPECS.DDR3 GSKILL 2133MHZ.. PHENOM 3.9-4.2TURBO. 3 SSDS-RAID0
My personal philosophy for CPU reviews is to run a LOT of benchmarks, since different benchmarks show different things.
The reviews I've read on other sites so far seem to have all reached similar conclusions: faster than Piledriver, competitive in most uses with midrange Intel CPUs.
The 8150 came out claiming native 1866 support, which to most implies one can fill the DRAM slots and run 1866, after untold complaints from people trying to run just 4x4GB (16GB total) of 1866, the advertising was changed to reflect 1866 at 1 DIMM Per Channel. AMD's BIOS and Kernel Guide for Developers reflects the same info (and that was based on 4GB DIMMs)....
I ask about the DRAM as AMD has been notorious for having weak Memory Controllers, (i.e. the initial 965 in general, couldn't carry 1600 DRAM until the C3 revision), so the CPU itself, primarily it's Memory Controller, does have a major say in what DRAM (frequency, density, total) it can run.
I, and I know of many others, would love to give this CPU a try, and see how it would do with the 8 cores....if it can support a sufficient amount of DRAM of a high frequency to run multiple VMs and applications. If in fact it runs 1866 as 'native' as AMD claims would be nice to see it run 32GB of same and if it CAN run faster DRAM.
Also of note: Would be interesting to see how the 8350 stands straight up to a 2500K with both OCed - most 2500Ks will run easily stable at 4.7 or better - vs the 5.0 of the 8350, keeping in mind of course the 2500K is now almost 2 years old, or better yet vs the 3570K (generally will run 4.5-4.6 with few problems or heat), which as of today at both the Egg and Tiger is about $5 less than the 8350
Newegg "$15 off w/ promo code EMCJNHD27, ends 11/1"
##newegg.com/Product/Product.aspx?Item=N82E16819116504&Tpk=3570K
As far as memory frequency if there's nothing to be concerned about between 1600 and 1866, then why not test at the CPU's true 'native' freq of 1333? Regardless, my question was how about large amounts of DRAM at high freq, i.e. can it handle 32GB @ 1600, 1866, 2133, 2400 etc. I originally brought up the question as the testing was done with 8GB, not the 16 you have loaded currently. I know most of my customers, look for 16GB to start and even systems that are planned primarily for gaming, those folks want 8GB min to start....and when looking to enthusiast systems, most want to get the most out of the system they can, whether it be gaming, rendering video, imaging work, GIS with large data sets, VMs where higher freqs = greater MTs (Mega Transfers per sec) or higher bandwidth. Maybe it's just me, but when I pull my DRAM out of one of my systems 2133 or 2400 and put in 1600 or even 1866 and run that in a like amount, I can 'see' a difference, especially when I start opening up say Win8 in a VM and play with images and be working with GIS or video in Win7
Here at Benchmark Reviews we have had occasion to test high frequency memory, and we've never found any quantifiable real-world difference in performance other than on synthetic memory benchmarks. If you can suggest a real-world application that benefits, with some numbers to back it up (rather than your 'feel'), we'd be interested to here about it.
I guess our thoughts as far as real world computing or quantifiable examples differ, I look to utilizing the cores of multiprocessor CPUs as an example, which is why I mentioned opening Win8 in a VM and rendering a video, while at the same time being in Win7 and running GIS, where as you talk about running a single app or benchmark at a time (i.e. how many of those 8 cores were actively doing anything in any one of the benchmarks, ever consider running multiple benchmarks at the same time?...might be surprised what you find). Further I said I could ?see? a difference when using faster DRAM, as in, things get done faster, not ?feel? as you imply somewhat sarcastically.
Again, if you can suggest some real world tests that will show the advantage of high speed memory, we're all ears.
Performance is NOT a warm fuzzy feeling you get, it's a quantifiable result. Anyone with an understanding of DDR3 memory will tell you that clock speed isn't important in contrast to latency. I even wrote a long in-depth technical article about: bit.ly/PoNdE8
Stop commenting, and start reading about the technology behind these items. You're asking us to run DDR3-2800 for the sake of a CPU test, which doesn't make sense to anyone who understands the pipeline architecture of a PC. Even if we wanted to run such a ridiculous test, finding compatible hardware components that consistently performed to this level would be extremely difficult.
As far as the technology goes, I'm rather comfortable with my knowledge, which is one of the reasons I didn't ask you to test with 2800 as you claim, in fact for 2800 the 3770K is about the only CPU that can carry 2800, so it's ridiculous for you to claim I said that (yet both you and David seem to delight in twisting what people say here). I simply asked if faster sets were looked at, one reason being you all tested the 8150 with 1866 sticks, so why drop to 1600 when you have a newer more powerful CPU (and in fact the 8150 had better bench results running with 1866 that it does here running with more DRAM 8GB vs 4GB in the original test/review), yet only using 1600 here.
Also in reading the reviews here, as long as you brought up reading about the technology here, in the review on the 1866 16GB Ares, it states as a CON the DRAM XMP profile doesn't set the specified speed and timings, which as you should know, the DRAM itself doesn't actually set things, it provides information in the SPD and it is up to the BIOS to take in and implement that info (i.e. it's not the RAMs fault), in the same review it's said it's not as cost effective as buying two 8GB sets, which may be considered a con-but most DRAM manufacturers recommend against mixing two sets or more, even of the same exact model - the XMP programming is done by the set which will differ between 2 and 4 stick sets, often leaving one to have to manually set up the sticks (in particular the advanced/secondary timings like tRFC (which can be a bear for most all users). There's also other DRAM reviews that have questionable items
Another good example is: earlier you commented "That wouldn't be a part of any CPU review, since it's more of a motherboard feature/limitation. The CPU can operate with any frequency, but the motherboard needs to support the speed and density first"
Which I also differed with and explained, and you failed to reply to. According to that statement you imply an entry level socket 1155, I3 can run 32GB of 2800 DRAM...if the motherboard supports it...I disagreed and again explained myself, Could you also explain your premise to that statement? Just curious.
For example, as you point out, in my FX-8150 review I used DDR3-1866 memory on the AMD processor and DDR3-1600 memory on the Intel processor. The idea was to show each CPU in the best light with its highest officially supported memory frequency.
Wow, you should have seen the comments I got on that. For example, one commenter "Jim Reece" said "I'll give you that you did your best to cripple the 2500K", because I didn't run 1866 DRAM with it. He also took me to task for using a 2500K as the comparison (AMD's stated performance target) rather than a 2600K or 2700K, apparently under the impression that I have every Intel CPU ever made in a box here or something.
In the end this commenter was so disgusted with me, certain that I was simply a shill for AMD, that he said "No need to reply, I'm gone, I'll stick with sites that ACTUALLY perform reviews that are objective rather than w/ bias."
But you know what? He came back. Not only did he come back, he applied to be a reviewer here. After missing several promised dates to complete his first review, he informed us that he wouldn't be a "good fit" because he's a "straight shooter" (and, as he informed us a year or so ago, we're biased) and launched into a bizarre screed attacking several of our other reviews on technical points, all of which were wrong.
We hope he's gone for good this time.
#memory.dataram.com/products-and-services/software/ramdisk
You can select how much of drive you use while retaining programs on it.
Windows recommends using a Flash Drive that can double the size ofyour RAM minus any programs U want to keep onit
Just been reading your benchmark test on the new trinity APU's (very good article), I used the A10 in a recent mini-itx build (Bit_fenix prodigy).
Ive had a request to build a rig for someone on a tightish budget, my first thought given the money available was to put in a AMD Piledriver FX-4 Quad Core 4300 and a 1gb 7850 gpu - total price: 210gbp, then I thought what about the A8-5600k crossfired with a HD6670 at approx: 134gbp.
I haven't been able to find any benchmarks unfortunately, I think its a pretty relevant question at the moment.
Many Thanks
Joel
Thanks,
- Aaron
If you do run apps that can spawn a lot of threads, well, 8 cores rocks. But aside from transcoding and rendering, there aren't a lot of apps that can really leverage this power.