Observations with an FX-8350

[ctsoth]

CPU frequency most certainly is relevant, especially given chips that overclock to roughly the same frequency. AMD putting the 8350 @ 4ghz is their attempt to hotrod the chip as much as possible, whereas the 2600k @ 3.4ghz is a very very conservative effort on Intel's part. You're arguing the MT performance of an architecture, which you simply cannot do without discussing frequencies.

You're looking at the .1% of time that Piledriver beats two year old i7 tech and calling it a win. Bulldozer and Piledriver are both poor implementations of a great idea.

Any CPU benches of higher mhz are irrelevant eh? Who's a fanboi, the guy who points out the one thing bulldozer has going for it, or the guy who says it doesn't count? Overclocking isn't the be all end all. Lots of people who value stability and data integrity do not overclock. Yes, base frequencies do count.

 

[ozzy702]

Any CPU benches of higher mhz are irrelevant eh? Who's a fanboi, the guy who points out the one thing bulldozer has going for it, or the guy who says it doesn't count? Overclocking isn't the be all end all. Lots of people who value stability and data integrity do not overclock. Yes, base frequencies do count.

You can't seriously claim that an 2600k @ 4ghz has a problem with stability or data integrity . :whiste: 4.2ghz is usually doable on stock clocks. Why didn't Intel release them at a higher frequency? Simply because they had zero competition and wanted to keep using super cheap heat sinks (maximizing profits) and offering far superior power efficiency.

If you're arguing architecture go ahead and do so, but clock frequencies set by the factory have had little to do with the factors you site for quite a long time at Intel.

^ This all coming from someone who has built five times the AMD setups as Intel and who's a long time AMD fan. Plain and simple Bulldozer and Piledriver are poorly implemented.

 

[inf64]

@evilwhitey

You assume majority overclocks their CPUs(even K/FX) while the opposite is true. Wast majority does not thinker with their machines,let alone OC them. Stock clock does matter and even if you say OCing is important you can OC FX8350 close to 5Ghz on good air/WC and have a 20-25% performance jump(similar could be said for OCing 2600K).

 

[ozzy702]

@evilwhitey

You assume majority overclocks their CPUs(even K/FX) while the opposite is true. Wast majority does not thinker with their machines,let alone OC them. Stock clock does matter and even if you say OCing is important you can OC FX8350 close to 5Ghz on good air/WC and have a 20-25% performance jump(similar could be said for OCing 2600K).

I'm not assuming that. AtenRa said he wanted to talk architectures and made claims that for MT Bulldozer and Piledriver are somehow superior, which they most certainly aren't.

Now if you want to just compare chips and their factory binning/clock speed then yes, Piledriver in MT is more or less equal with a 2600k which given AMD's aggressive pricing with Piledriver makes their chips roughly equal to Intels in price/performance.

Now if AMD can find a way to minimize the CMT tax, squeeze a little more IPC and further increase clock speeds (which may not happen until a move past 28nm) then they'll have a budget friendly winner.

I've recommended the 8320 and 8350 to several friends that play BF3 and if I was looking to build a purely budget machine and didn't play poorly threaded games I'd run one in a second. I was more than willing to pay for my 3770k four months ago and would probably make the same decision again today.

 

[Idontcare]

I have an old 790FX mobo with an SSD as boot and some other 5 drives for storage in RAID. Obviosuly I had to use AMD's drivers in that configuration. To check if TRIM was working I tried this: "fsutil behavior query DisableDeleteNotify". If the result is '0' TRIM is enabled. The result was 0 for what it's worth.

Trim is working, no issues with the drivers :thumbsup:

Results with ATTO:

HD Tach:

HD Tune:

And everybody's favorite - crystaldiskmark:

For comparison here are the numbers using the same Vertex 3 on my 2600K:

The numbers are quite comparable. Slightly lower on the AMD platform but not drastically so.

I do wonder why though, SATA speeds are a standard so the only reason the AMD platform is pulling and pushing data to the Vertex 3 slower than what the Intel platform does must come down to the SATA overhead on the AMD platform is eating up more bandwidth than on the Intel platform.

Does that even make sense? What other reason could explain it?

 

[sequoia464]

Quote: Originally Posted by sequoia464
Right now I have 2 sticks of Samsung 1600 memory @1866 1T (9-9-9-24-45). Just ordered a couple more, be interesting to see how things work with 4 of the Samsung - I know that my GSkill and Kingston would not run @ 1T even with just 2 sticks.


The thing is I know these 4 GSkill sticks will do 1T at these timings and volts because they do it on my Intel rig. So it can't be the ram that is the issue in preventing 1T operation for my specific ram sticks, can it?

I installed the other 2 sticks of the Samsung, set it at 1T and no problems.

The GSkill that I have should run at 1T also, but I couldn't even boot in to windows with it unless I ran at 2T.

 

[Puppies04]

2600K/2700K loose in 99% of highly MT applications even in Cinebench.

Take those AMD tinted specs off, your own link shows the 2700K beating the 8350 in cinebench 11.5.

You do make me lol sometimes.

Edit.

Not only is the intel chip sucking a vast % less power than the piledriver chip whilst doing this it is also managing this with a .5ghz non turbo or .3ghz turbo clockspeed defecit.

 

[Puppies04]

This is not the best case scenario, it is the only scenario i was talking about. I have clearly stated from my first post that i was talking about the MT scenario only. I havent talked about power consumption or anything else, i was purely talking about MT Performance, the reason of BullDozer(CMT) design. In those applications the FX8350 is faster than Sandybridge and if PileDriver was released in 2011 things would have been different.

No it isn't, your own link shows this. Go back and check the numbers if you don't believe me. Ignoring cinebench single threaded because AMD gets its ass handed to it on a plate I see the 2700k going 2 for 2 with the 8350 even with its clockspeed disadvantage.

 

[AtenRa]

The numbers are quite comparable. Slightly lower on the AMD platform but not drastically so.

I do wonder why though, SATA speeds are a standard so the only reason the AMD platform is pulling and pushing data to the Vertex 3 slower than what the Intel platform does must come down to the SATA overhead on the AMD platform is eating up more bandwidth than on the Intel platform.

Does that even make sense? What other reason could explain it?

Most probably because Intel 1156/1155 platform only has a single Chipset communicating directly with the CPU (lower latency) when AM3+ platform has two Chipsets (North and South).
The SATA-3 Controller in the AM3+ is situated at the SouthBridge (SB950) that communicates with the CPU through the NorthBridge(990FX). That makes for a higher latency thus lower performance. It is the same with PCI-e and GPUs as well.

 

[AtenRa]

@Puppies04

Im not going to continue this here and ruin IDCs topic.

 

[wlee15]

You can see the instructions going through the FMAC. Also of note, if you look at the bottom instruction, the latency of this instruction (far right number) more than doubles from K10 (114). Across most of the more complex instructions I looked through, the latency took a dramatic hit from K10.

So, perhaps emulation wasn't the right word (though I'm still not convinced that at least some of the instructions aren't done through emulation), but to me it is pretty clear that AMD's new fpu can do x87 for compatibility reasons only and did not give it much dedicated resources for fast execution.

Note the latency for that same instruction on Sandy Bridge is 726 cycles.

 

[Puppies04]

@Puppies04

Im not going to continue this here and ruin IDCs topic.

I'm not surpised, you contradict yourself constantly, you should have given up 10 posts ago.

 

[AtenRa]

I'm not surpised, you contradict yourself constantly, you should have given up 10 posts ago.

Well, i was talking about MT performance only and people started talking about Single Thread, Power Consumption, IPC, CPU frequencies and more. Continue counter each of you here would ruin the topic, im not going to do that no matter what you people think.

 

[Abwx]

No it isn't, your own link shows this. Go back and check the numbers if you don't believe me. Ignoring cinebench single threaded because AMD gets its ass handed to it on a plate I see the 2700k going 2 for 2 with the 8350 even with its clockspeed disadvantage.

It s not the 2700K that is great , it s the task that has been
Intel optimized for instructions , paths , disposable exe ressources ,
cache sizes and so on....

 

[Arkaign]

Trim is working, no issues with the drivers :thumbsup:

Results with ATTO:

HD Tach:

HD Tune:

And everybody's favorite - crystaldiskmark:

For comparison here are the numbers using the same Vertex 3 on my 2600K:

The numbers are quite comparable. Slightly lower on the AMD platform but not drastically so.

I do wonder why though, SATA speeds are a standard so the only reason the AMD platform is pulling and pushing data to the Vertex 3 slower than what the Intel platform does must come down to the SATA overhead on the AMD platform is eating up more bandwidth than on the Intel platform.

Does that even make sense? What other reason could explain it?

That's very interesting, and mirrors my own experiences.

Can you bench your results if you have an external USB 2.0 and/or 3.0 HDD on both?

In my experience, both USB and SATA performance is a little bit behind on AMD setups. Not so much to be a huge problem, but it does bug me.

 

[Puppies04]

It s not the 2700K that is great , it s the task that has been
Intel optimized for instructions , paths , disposable exe ressources ,
cache sizes and so on....

I see what you mean, lets use a car analogy I haven't seen one in this thread for a while.


It's not my car that is fast it is just designed to burn petrol efficiently while doing 50% MPG than other cars with the bigger engines, handling better, going faster and being better equipped inside.

 

[NTMBK]

AtenRa makes a good point about the outdated platform for the FX chips- they've not had a revolution there since they integrated the memory controller. I wonder what those SSD numbers would look like on an A10- they use a more modern chipset (single FCH, instead of separate NB and SB), so things might be better there.

 

[Abwx]

I see what you mean, lets use a car analogy I haven't seen one in this thread for a while.

It's not my car that is fast it is just designed to burn petrol efficiently while doing 50% MPG than other cars with the bigger engines, handling better, going faster and being better equipped inside.

You are seeing nothing , just displacing the debate by using
irrelevant car analogy that spot the wrong parameters.

If accurate it would say that both cars are not provided
with the same circuit , one being favored by a road that
is made as straight and short as possible trough specific optimizations...

For the rest , the slide say it all , whatever other dubious
arguments are in your biaised mind...

 

[SPBHM]

AtenRa makes a good point about the outdated platform for the FX chips- they've not had a revolution there since they integrated the memory controller. I wonder what those SSD numbers would look like on an A10- they use a more modern chipset (single FCH, instead of separate NB and SB), so things might be better there.

I think it's probably the controller/software itself, there is always some variation from MB to MB, but it doesn't look to different than anything else with the a85
http://www.anandtech.com/show/6360/asus-f2a85v-pro-review-a-look-at-fm2-with-a85x/6

if I remember correctly, AMD was really behind Intel and NV with how SB600 performed, I think the current chipsets are close enough to Intel.

 

[frozentundra123456]

You are seeing nothing , just displacing the debate by using
irrelevant car analogy that spot the wrong parameters.

If accurate it would say that both cars are not provided
with the same circuit , one being favored by a road that
is made as straight and short as possible trough specific optimizations...

For the rest , the slide say it all , whatever other dubious
arguments are in your biaised mind...

Disregarding the car analogy, it does seem to me that AMD should better optimize their processors for the current environment (lots of lightly threaded apps that need high ipc), rather than some idealized future one where everything uses 8 cores optimally, heterogeneous computing is the norm, and cold fusion has made power use irrelevant.

 

[Abwx]

Disregarding the car analogy, it does seem to me that AMD should better optimize their processors for the current environment (lots of lightly threaded apps that need high ipc), rather than some idealized future one where everything uses 8 cores optimally, heterogeneous computing is the norm, and cold fusion has made power use irrelevant.

I did adress a MT task claim in respect of optimizations while you are talking
of something different , i.e , ST perfs in current environment....

Anyway this is not the place for such an already beaten to death debate.

 

[Idontcare]

Most probably because Intel 1156/1155 platform only has a single Chipset communicating directly with the CPU (lower latency) when AM3+ platform has two Chipsets (North and South).
The SATA-3 Controller in the AM3+ is situated at the SouthBridge (SB950) that communicates with the CPU through the NorthBridge(990FX). That makes for a higher latency thus lower performance. It is the same with PCI-e and GPUs as well.

Ah yes! Excellent observation!

I had failed to realize the SATA datapath topology will be different (longer) with the AMD platform versus the Intel one because of the 2-chip solution.

No doubt the extra bit of bandwidth loss is due to exactly this. :thumbsup:

Each chip in the datapath has to catch and then retransmit the data forward to the next, and that can never be a zero-latency adder step.

Is AMD expected to reduce the datapath topology for their next platform refresh?

Not that it really matters, I mean there is no way I'd ever notice the performance difference with the SSD performance on either platform but does AMD intend to integrate to a single-chip solution eventually?

 

[ShintaiDK]

Is AMD expected to reduce the datapath topology for their next platform refresh?

They already did with the FM platform.

 

[guskline]

IDC: I doubt AMD will change the chipset for Bulldozer/Piledriver cpus.
IF AMD ever releases a Steamroller it might get a new socket and chipset to try to stay competitive. I have no regrets with my Piledriver but realize to come close to Intel's Sandy/Ivy high end chips AMD has to make some major mods.

 

[Idontcare]

They already did with the FM platform.

Interesting. I wonder how much of the existing performance lag in SATA bandwidth is due to the weaker CPU itself at that point?

I'm going to underclock my 8350 and re-run the SATA benches just out of curiosity.