Observations with an FX-8350

[bgt]

changed my fans for the noiseblocker B12-2/3. The 3 is on the CPU. They are really quiet and powerfull. Temp dus not get over 55C with prime95@full speed(fans). Speedfan is 75% max on the two case fans and 50% for the cpu fan. I let it get to 62C for less revs.

 

[guskline]

changed my fans for the noiseblocker B12-2/3. The 3 is on the CPU. They are really quiet and powerfull. Temp dus not get over 55C with prime95@full speed(fans). Speedfan is 75% max on the two case fans and 50% for the cpu fan. I let it get to 62C for less revs.

bgt: Is your 8350 overclocked or stock?

 

[CTho9305]

Commenting before I've read the whole thread...

Actually, that graph is extremely telling in regards to the claims that a module is a full 2 cores. I don't believe I've ever seen that particular set of data before.

This is a special case though - we know the FPUs are shared, and LINPACK is about as pure-FP as you can get. I see IDC posted some cinebench and maxwell scaling numbers that show much better scaling.

Do you happen to know anything official regarding TJmax for the FX-8350? Know anyone that would know?

When you posted your 200 watt results I searched a bit - AMD used to publish thermal and electrical specifications that could list absolute maximums, but I couldn't find specs for anything more recent than 65nm parts. Disappointing.

I have no issue believing this to be the case, but damn, that had to save them how much die-area? Maybe 0.1mm^2? The x87 instruction set itself is rather darn small, so small in fact that even back in the time of 486 processors it was viable to incorporate it onto the die.

Removing it now has got to be the equivalent of writing this post but leaving off the period at the end of the sentence just for the sake of saving the effort of pressing one more key



If it is then that is some darn impressive emulation considering performance is just lowly, not miserable or abysmal.

It wouldn't be a matter of just die area - it would be a matter of complexity and performance. Arithmetic units scale noticeably with operand size, and critical paths also scale. That said, I have no idea whether that actually changed here and wlee15 seems pretty sure it wasn't this.

 

[bgt]

bgt: Is your 8350 overclocked or stock?

stock

 

[frozentundra123456]

Wheres the cherry picking? Piledrivers architecture was clearly designed for MT apps in mind and it succeded there, the FX-8350 is a stronger chip than Intels quadcores, it stays in Core i7 class in the most heavy multithreaded apps and usage, case closed. Lets move on.

Hmm, equal or slower, or in a few isolated cases, slightly faster, means "stronger"... interesting concept.

I am comparing to i7 since that is a quad core, and you are claiming 8350 is stronger than intel quads.

And BTW isnt considering only one type of app the very definition of "cherry picking"?

 

[Hitman928]

I'm not so sure this is true (as others I think have pointed out already).

The FPU throughput should remain unchanged from previous generations. The only reason it would perform slower than a 6-core phenom is because the phenom can execute 50% more FP instructions (since the FPU is shared in a module on bulldozer).

And even using SSE2 for floating point math doesn't make bulldozer any faster.

That wouldn't explain why then Bulldozer is significantly slower in pure single threaded x87, like super-pi or why specific x87 instructions saw a large increase in latency through the core. No CMT tax in super-pi, but still much slower. In other SSE2+ benchmarks, the first gen Bulldozers were equal to a little slower than their phenom counterparts. So, there's other things going on, but x87 definitely took the biggest hit in fpu performance (20%+ slower than phenoms for x87 versus 5-8% slower for sse).

 

[Face2Face]

So do the new Piledriver CPU's have faster IPC than Phenom II?

 

[Homeles]

So do the new Piledriver CPU's have faster IPC than Phenom II?

No. They won't have that until Steamroller.

 

[Ajay]

No. They won't have that until Steamroller.

2014 is a long way off for AMD at this point. We'll just have to wait and see if the 'big core' group still exists in 2014.

 

[Face2Face]

No. They won't have that until Steamroller.

So for gaming a 8320 would be a downgrade from the Phenom II I have and draw a lot more power at the same time?

So a Phenom II @ 4.0Ghz is equal to a 8320 @ _Ghz? IPC only

 

[AtenRa]

So for gaming a 8320 would be a downgrade from the Phenom II I have and draw a lot more power at the same time?

So a Phenom II @ 4.0Ghz is equal to a 8320 @ _Ghz? IPC only

FX8350(4GHz) has the same Single Thread performance as 3.7GHz Phenom II in Cinebench 11.5.
I will say that at 4.3GHz it will be in par with your Phenom II @ 4GHz.
As for the power, at idle the FX will have lower consumption. It will only have higher consumption when all cores are active but then it will have double the performance than your quad core Phenom, meaning it will consume the same.

 

[Hitman928]

Actually, gaming (in general) is one area where the Piledriver cores match or are faster clock for clock over the Phenom's. It's probably not a big enough jump though to justify the upgrade price from an oc'd Phenom X4. Other workloads, the Piledriver cores roughly match a Phenom. Sometimes their a little faster clock for clock, a little more often they are a slightly slower.

http://www.tomshardware.com/reviews/fx-8350-vishera-review,3328-15.html
http://www.anandtech.com/bench/Product/186?vs=699

(remember those games that use 1-2 cores, piledriver will have a turbo advantage as well, but even when clocked the same, PD typically does better than x4 phenoms).

 

[SlowSpyder]

So for gaming a 8320 would be a downgrade from the Phenom II I have and draw a lot more power at the same time?

So a Phenom II @ 4.0Ghz is equal to a 8320 @ _Ghz? IPC only

A while back I remember someone posting an article or thread on PD single threaded performance. If I remember correctly, PD was ~7% slower in single threaded performance. Since Vishera can typically easily clock well over 7% more than your average overclocked PhII, you should expect better overall performance. That, and up to eight threads, if you need them.

I think a lot of us wish AMD would have had something better than ~7% less IPC combined with 15% higher clocks three years after PhenomII launched, though. But in the end, performance should be faster with Vishera. If my motherboard would have been AM3+, I'd buy one.

 

[Face2Face]

Good info guys thanks. I was just curious, I am pretty sure I am going Intel next round.. Believe it or not, I am still pretty happy with the performance of the 4.0 Ghz PHII

 

[guskline]

IDC: Check your PM for the Bios setting.

 

[bgt]

My sons pc has a 1090T@stock with a 880chipset with an Asus mb but really is no match with the 8350. We tested it with Adobe photoshop, he is a web designer. The FX really runs circles around it even if we clock it down to 3.2Ghz.
He also wants to upgrade his pc so we are testing a lot of stuff on the 2500K and the FX CPU.

PS does the 3770 still gets very hot when under full load? I've read something about the heatspreader not being optimal or is this cured now? Thinking of buying 1 to test.

 

[guskline]

bgt: Thanks for the comparison of the 1090T and the 8350 as to Adobe photoshop.

 

[jvroig]

FX8350(4GHz) has the same Single Thread performance as 3.7GHz Phenom II in Cinebench 11.5.

Don't you mean 4.2Ghz since I assume TurboCore has been left on?

Deneb X4 980, 3.7GHz: 1.1
Thuban 1100T, 3.3GHz (but turbos to 3.7): 1.1
FX3850 (4Ghz, but turbos to 4.2): 1.1

Given the data points above from the chart (I added the turbo speeds myself, however), it makes sense to conclude that TurboCore was left on, otherwise there is a huge discrepancy in the Thuban and Deneb scores if Turbo was off.

 

[SPBHM]

PS does the 3770 still gets very hot when under full load? I've read something about the heatspreader not being optimal or is this cured now? Thinking of buying 1 to test.

as far as I know nothing has changed, but it's not really a problem since the CPU can work at some pretty high temperatures...

also you shouldn't really compare temperature readings from different CPUs/sensors like the FX and your i5... I think the only way would be using an external sensor (the same) for both.

 

[AtenRa]

Don't you mean 4.2Ghz since I assume TurboCore has been left on?

Deneb X4 980, 3.7GHz: 1.1
Thuban 1100T, 3.3GHz (but turbos to 3.7): 1.1
FX3850 (4Ghz, but turbos to 4.2): 1.1

Given the data points above from the chart (I added the turbo speeds myself, however), it makes sense to conclude that TurboCore was left on, otherwise there is a huge discrepancy in the Thuban and Deneb scores if Turbo was off.

Yes you are right, I always forget about the turbo at 4.2GHz. At 4GHz it produces 1.06 points

 

[Idontcare]

I am getting started on the OC'ing fun, but I don't want to burn this chip up while still using stock HSF and stock TIM.

I know I asked earlier, but I cannot seem to find the response (thread is unwieldly) but what is the safe voltage for these pilderiver chips? We know AMD knows but isn't telling, but what is the general concensus?

Is 1.5V safe? What about 1.6V?

^ the red dots are data for my chip, optimized Vcore as needed to remain stable while running at least 5 passes of LinX.

The extrapolation to 4.5GHz reads out at 1.49V, and 1.68V for 5GHz D: Are those considered silly stupid voltage for a 32nm AMD chip, or is that "safe and reasonable"?

(by comparison I had no qualms over shoveling 1.5V into my 32nm 2600K and running it at 93C with LinX...perhaps arguably "silly stupid" of me, but that is my threshold as a baseline in answering my question above)

 

[2is]

I have a feeling that much after 4.6GHz there's going to be a steeper voltage curve that the current graph suggests.

 

[guskline]

I am getting started on the OC'ing fun, but I don't want to burn this chip up while still using stock HSF and stock TIM.

I know I asked earlier, but I cannot seem to find the response (thread is unwieldly) but what is the safe voltage for these pilderiver chips? We know AMD knows but isn't telling, but what is the general concensus?

Is 1.5V safe? What about 1.6V?

^ the red dots are data for my chip, optimized Vcore as needed to remain stable while running at least 5 passes of LinX.

The extrapolation to 4.5GHz reads out at 1.49V, and 1.68V for 5GHz D: Are those considered silly stupid voltage for a 32nm AMD chip, or is that "safe and reasonable"?

(by comparison I had no qualms over shoveling 1.5V into my 32nm 2600K and running it at 93C with LinX...perhaps arguably "silly stupid" of me, but that is my threshold as a baseline in answering my question above)

From the digging I've done on the forums voltages up to 1.5 V are probably safe. HOWEVER, the stock HSF is not up to the job of taking away all the heat as you go much over 4.3-4.5 Ghz. Have a Corsair H100 with 4 fans makes a BIG difference along with excellent cooling on the VRM area. The stock HSF is adequate for defaults but run a benchmark on OC mode and even the PileDriver temps can shoot way up. You've got to dissipate the heat quickly.

 

[Idontcare]

From the digging I've done on the forums voltages up to 1.5 V are probably safe. HOWEVER, the stock HSF is not up to the job of taking away all the heat as you go much over 4.3-4.5 Ghz. Have a Corsair H100 with 4 fans makes a BIG difference along with excellent cooling on the VRM area. The stock HSF is adequate for defaults but run a benchmark on OC mode and even the PileDriver temps can shoot way up. You've got to dissipate the heat quickly.

Yeah I have an H100 but first I wanted to flesh out the OC capabilities of the processor at stock (without the aid of a $100 3rd party cooler ).

AMD's stock HSF is heads-and-shoulders above Intel's stock HSF and the max OC's with stock show that. I can't take my 3770K above 4.2GHz with the stock HSF (it starts thermal throttling), likewise I can't take my 2600K above 4GHz with its stock HSF.

But I have been running benchmarks at 4.5GHz with the piledriver on a stock HSF. Granted I didn't test for LinX stability, that is coming up, but still, pretty good for a HSF that comes with the CPU.

 

[guskline]

Yeah I have an H100 but first I wanted to flesh out the OC capabilities of the processor at stock (without the aid of a $100 3rd party cooler ).

AMD's stock HSF is heads-and-shoulders above Intel's stock HSF and the max OC's with stock show that. I can't take my 3770K above 4.2GHz with the stock HSF (it starts thermal throttling), likewise I can't take my 2600K above 4GHz with its stock HSF.

But I have been running benchmarks at 4.5GHz with the piledriver on a stock HSF. Granted I didn't test for LinX stability, that is coming up, but still, pretty good for a HSF that comes with the CPU.

Too bad AMD didn't put more into the CPU and less into the stock HSF!