AMD FX-8350 Review: Does Piledriver Fix Bulldozer's Flaws?

[mlcaouette]

Finally looks that AMD released some with enough merit to be in my sabertooth 990fx. Too bad I was let down last year when I bought the board months prior to bulldocrap release.

Seems like the 8320 will be a fair replacement once I get sick of my 955 @ 4Ghz.

Good job AMD, hopefully this will help with the financial woes you've been seeing as of late. After all whats good for AMD is good for a competitive market and is a win for everyone.

 

[Yargnit]

No tests on Windows 8? Didn't Bulldozer gain another 7% or so in 8 compared to 7? If that trends continued with Piledriver, another few % on single threaded applications could make a huge difference.

That said, Huge improvement, and frankly, what Bulldozer should have been. The price is spot on, and the multi-core advantage to single core disadvantage trade off is now a much more even scale. There is now a valid reason to consider this over a 3570k depending on your individual needs.

 

[bystander]

[citation][nom]EzioAs[/nom]To compare directly a CPU with a $200 MSRP with a $300 CPU isn't really fair imo (though I do get your point). And for multi-threaded workload, the 8350 isn't really far off from the 3770K. It usually came very close (7-15% slower) but for the price, it really is the better price to performance CPU (compared to the 3770K that is)[/citation]

The pricing is a result of their performance. I was just commenting on the fact that per core, AMD is still lagging far behind. 4 Core processors with hyperthreading really shouldn't be equal to 8 cores. For gaming, they still aren't all that great, but the extra cores for the cost, they will do well in desktop applications at least.

 

[Guest]

You seem to forget that unlike Intel's Ivy compared to Sandy, Vishera versus Zambezi leaves Vishera the superior overclocker as well as cooler-running and with superior overclocking price/performance ratios. There's also the fact that AMD did this on the same process node, not that that matters as anything other than a foot note.

not forgetting anything; like not just sandy/ivy but c2d/nehalem/sandy/ivy, no not all the same performance power differences but you get the hint.
so my not as sensitive as the usual AMD fanboy that posts here* friend; the point is at least this is progress that has been successful in the past. however your little footnote makes me wonder how far the insanity of fail can spread. really how successful is a mantra of "overclock for performance"?

i'll skip the, "whoppee! so it beats a PIIx4 now, break out the party hats!"


*seriously, if the butthurt pillowbiters didn't thumbdown to hide my comments; i'd seriously wonder if their sanity was restored.

 

[Benthon]

Congrats AMD. Not bad guys. One more generation and hopefully you'll have mostly caught up. I'm happy for all of us consumers who benefit from this.

 

[dscudella]

They should have built on the 965 BE. Take out the modules, drop it down to 4 true cores and start over with enthusiasts in mind.

 

[Bloob]

[citation][nom]dscudella[/nom]I would have liked to see more Intel offerings in the Benchmarks. Say an i3-2120 & i3-3220 for comparisons sake as they'll be cheaper than the new Piledrivers.If more games / daily use apps start using more cores these new AMD's could really take off.[/citation]
It is not just about applications using multiple threads, it is also about you using multiple applications ( that might use multiple threads ) simultaneously.

It is much better effort this time around with proper pricing, although not quite beating Intel yet ( that power consumption is really painful ).

P.S. I would have liked to see some Win 8 tests, as I've seen it implied that AMD would do better there

 

[ronch79]

I hope retailers aren't gonna go way beyond the $195 MSRP for the FX-8350. Stick to the MSRP, kids, because I'm buying one.

 

[sugetsu]

I am very happy with this new development. Given the recent line of FX processors would you consider them for video editing over intel if you are on a budget? Will the release of windows 8 give AMD the edge? I am about to built an $800 machine for video editing purposes and I would like to read your comments.

I know that AMD's main weakness is single threaded performance, but isn't multi threaded performance the way of the future anyway? Aren't future programs supposed to be design to take advantage of multiple cores? If I am correct on this then AMD will finally give intel a run for its money.

 

[Katsu_rap]

Vishera Summary:

Homerun? No.
Improvement? Yes.
Power efficiency? Getting better
Overclockable? Yes to that
Pricing? Not bad at all

 

[cangelini]

[citation][nom]Yargnit[/nom]No tests on Windows 8? Didn't Bulldozer gain another 7% or so in 8 compared to 7? If that trends continued with Piledriver, another few % on single threaded applications could make a huge difference.That said, Huge improvement, and frankly, what Bulldozer should have been. The price is spot on, and the multi-core advantage to single core disadvantage trade off is now a much more even scale. There is now a valid reason to consider this over a 3570k depending on your individual needs.[/citation]
Nope. Even AMD's own guys are saying don't expect any improvement from Windows 8 versus a properly-patched Windows 7 machine (which mine is).

 

[de5_Roy]

nice review. enjoyed reading it.
seems like amd did improve from bulldozer. it's all fine and dandy until i think about how low they set the bar with bd in the first place...
still, considering launch price, overall performance - the 8350 seems better than fx8150. amd deserves credit for that.

 

[rohitbaran]

Only read the conclusion for now, but seems like AMD did good. Hey AMD, don't fire your engineers! Look, they are going on the right track.

 

[mayankleoboy1]

I see most of the commenters agree that PD is a leap forward from BD and PHII.

But i want to know who will go out and really buy the CPU ? I want to see people convert the comments into actual buys.

 

[mayankleoboy1]

None of the benchmarks use the advanced instruction set of PD. None can actually use FMA and BMI instructions.

 

[ukee1593]

^^ @sugetsu

Yes I'd expect that a FX8350/8320 to be a very good build for a (high end) video editing/content creation build. To hit ~$800 on an Intel build while getting performance comparable to the 8350 is very difficult. Better to use a Cheapo AMD processor than to cheap out on your Power Supply, Motherboard and Case just so you can afford an Intel processor!!

AnAnd did a bit better coverage of the idle power consumption of the 8350. http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/6 dropping to ~74 over ~77 Watts for the old dozer 8150. This is very impressive considering that most of the power will be wasted in the chipset (which doesn't change over the old FX)

I'd also expect Piledriver to be quite a hit in Server applications as those workloads are often more multithreaded.

++ for a video editing build ensure you get a nice GPU to accelerate rendering while you are working. This is worth a lot more than the CPU in many cases.

 

[cangelini]

[citation][nom]mayankleoboy1[/nom]None of the benchmarks use the advanced instruction set of PD. None can actually use FMA and BMI instructions.[/citation]

Actually, Sandra uses FMA3 for PD and FMA4 for BD. But we aren't even getting widespread use of AVX in these tests, and that was something I blew countless hours on for the BD introduction (go back and read it--even had custom code in there to see how AVX would help certain apps). Let's hope ISVs can put these new ISA extensions to use sooner than later, eh?

 

[sarinaide]

There was never going to be a instant golden ticket when moving to a new arch, take intels move from P3 to Conroe, took 3 generations of tweaking to finally reach Conroe, so in that regard PD shows significant gains over Zambezi, upping clocks and lowering power consumption at the same time and on the same strained 32nm process. 10-20% performance yields in certain workloads is good reflection. Vishera is much better all round.

 

[nibir2011]

So the thing AMD needs to do to be right where they were before the intel core(sandy bridge and ivy bridge) architecture is do something awesome at the same time frame of the release of haswell release.

AMD is in right track. I absolutely think that it should not think about single threaded performance but it must improve its per clock performance (yes the thing you are thinking is if per clock performance increases then single threaded performance will increase too, but i think that is not necessarily true because AMD is giving more core that is where it can make a difference so in a threaded application 1% of increase will give a overall increase of 8% increase in 8 core whereas intel's four core can only give 4%).And the total industry is moving toward threaded application even today's mobile apps are designed to take advantage of the dual or quad core processor of the phone or tablet. Most importantly the price is very attractive.That is a awesome choice for any professional or official uses. As in a company your target is to get maximum at the lowest cost possible. The only problem it has is its power consumption. Because in production people must think about bills.Which is also the reason why it did not get the Tom's Hardware Recommended Buy however it surely could get a tom's hardware Approved hardware award.I think this problem will be solved in the next release. If that happens then it will create the same height it created for the Athlon 64s.It feels very good hen there is a contender like AMD is the line, that's why Intel can not cut our pockets more aggressively. I am hoping to see a price drop in the i5 lineup for the release.

 

[mayankleoboy1]

1.The visual studio tests are not much conclusive. The compiling time depends on the project to be compiled. You comipiled Chrome, which can spawn multiple threads.If you try to compile Firefox, you will find that it uses multiple cores only 10% of the time. The rest is single threaded. So it will perform worse on PD

2. Can you provide the files you use in the compression benchy, so we can test our hardware too?

 

[razor512]

Push the X6 1100 to 4GHz and it will beat the 8350 in every test except the ones that require special CPU optimizations.

What tomshardware needs to do is another round of average overclocked CPU's to see if it is better for a user to get a X6 and overclock it or 8350 and overclock it.

I am currently using a X6 1075T and when overclocked, easily beats the 8350 by a good amount in cinebench.



Since your readers are likely to be the type that will build their own system and also upgrade their CPU, and most of all, overclock.

Push the x6 1100 to 4GHz, then push the 8350 to 4.8GHz, then repeat all of the current benchmarks then lets see which offers the most bang for your buck.

PS also do mixed multitasking, the problem with the FX chips come in mixed workloads, since instead of 6 discrete cores, the 8350 is really a quad core with a select few components doubled to handle an additional thread. the problem with this is that you only get a good amount of performance when all cores are doing the same thing but if you do something that has the CPU doing a mixture of processing that stresses multiple components within each CPU core, then performance drops to levels of a quad core due to the shared components.

AMD needs to refresh their line with a true 8 core chip that can meet or exceed the per core performance of the 1100t while also handling higher clock speeds. If they can do this then the CPU and improve the efficiency a little so that it can run cooler, then they can easily compete closely or even beat many current gen core i7's in many real world benchmarks (especially if they can maintain their high clock speeds)

the area where the FX chips perform well is on tasks that are computationally intensive but working with a small amount of data as this stresses mainly the parts that they doubled, but in balanced tasks, the performance takes a nosedive as the cores fight over use of components that each core traditionally has all to them self (no sharing)

With my 1075t overclocked to 4GHz, I get a CPU score of 7.2

Please tomshardware, try overclocking the 8350 enough to hit 7.2, then after that, push it to it's max overclock to see what those of us can expect over our old Phenom II x6 chips.

 

[cangelini]

[citation][nom]mayankleoboy1[/nom]1.The visual studio tests are not much conclusive. The compiling time depends on the project to be compiled. You comipiled Chrome, which can spawn multiple threads.If you try to compile Firefox, you will find that it uses multiple cores only 10% of the time. The rest is single threaded. So it will perform worse on PD2. Can you provide the files you use in the compression benchy, so we can test our hardware too?[/citation]
Remember that we used to compile Miranda. The audience asked us to step up to something more demanding, which is why we have this super-huge compile job now which eats up a tons of time during the benchmark run

Our packsource for the compression tests is 1.35 GB. Not sure the best way to host that, but I can ask, if it's really something you'd like. Alternatively, if you catch me via email or Skype, I could zip it all up and send directly, I'm sure.

 

[Guest]

actually this comparison form anantech is both sobering and all that one needs to know

Projected Performance: Can AMD Catch up with Intel?

http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/7

 

[cobra5000]

jbx007, there is ALWAYS a mark up, when the hot new product comes out. It dies down pretty quickly, as long as the supply meets the demand.

 

[Guest]

[citation][nom]razor512[/nom]...[/citation]

you hit it on the head, fx is basically little more than newer form of hyper-threading

so its not like you have a full 8 cores, core pairs have shared areas and so depending on what is being processed it can end up being little more then 4 core.

also, amd needs to make improvements to their implementation of the x86 instruction set, its been lagging bigtime for a while and it does not appear that they are addressing it, but working around the issue.

you can see just how sad when you run single core / thread benchmarks., there you will see no matter how much they spiff up multi-threaded, they are always behind because of their core's poor performance.

anyways those are my take of all i read.