Testing does FX-83xx seem to flat line after 4.3GHz

[Johan45]

Man I almost gave up half way through that post. And you said you were roofing?? I have never has a bull dozer so I can't compare but I'm happy with what I have but I have put some work into making it go. That is true.

 

[storm-chaser]

Again well thought out and explained to highlight your point. 4.3GHZ yields a roughly 3.75% increase in cinebench score VS a 2% increase at 4.4Ghz. But there is no "leveling off" at all if you were to chart Cinebench scores based on CPU speed. It continues to trend upward but just not at the rate you see at the 4.3Ghz point. A Cinebench score at 4.4Ghz is going to be higher than that at 4.3, and so on. It does scale upward, proving that a higher clock speed does result in better Cinebench #s throughout clock speed range.

The main reason explained as to the sweet spot being 4.3 was heat. Given proper cooling it appears there is no other reason to stop at 4.3. Early on it was said only serious liquid cooling devices would be practical to go beyond that but we are now seeing air coolers pushing these cpus past the 4.6 GHZ mark, as you mentioned. So as far as I can tell, like any other CPU, these scale well and with a proper cooling solution the sky is the limit.

And bear in mind Cinebench is just one of many benchmarks out there, only one indication of how a CPU performs at a given CPU speed, or how well it scales.

Still, very impressive work RGONE and thank you for all the data collected and the incredible undertaking to compile and develop results.

 

[caddi daddi]

and it's more than core temps that come into play, the motherboard heats over time, the vrms realy heat over time the case heats and all that heat has to be gotten rid of and that's right into your office.

 

[Mandrake4565]

and it's more than core temps that come into play, the motherboard heats over time, the vrms realy heat over time the case heats and all that heat has to be gotten rid of and that's right into your office.

They do make for some good space heaters in the winter!

 

[ShrimpBrime]

Again well thought out and explained to highlight your point. 4.3GHZ yields a roughly 3.75% increase in cinebench score VS a 2% increase at 4.4Ghz. But there is no "leveling off" at all if you were to chart Cinebench scores based on CPU speed. It continues to trend upward but just not at the rate you see at the 4.3Ghz point. A Cinebench score at 4.4Ghz is going to be higher than that at 4.3, and so on. It does scale upward, proving that a higher clock speed does result in better Cinebench #s throughout clock speed range.

The main reason explained as to the sweet spot being 4.3 was heat. Given proper cooling it appears there is no other reason to stop at 4.3. Early on it was said only serious liquid cooling devices would be practical to go beyond that but we are now seeing air coolers pushing these cpus past the 4.6 GHZ mark, as you mentioned. So as far as I can tell, like any other CPU, these scale well and with a proper cooling solution the sky is the limit.

And bear in mind Cinebench is just one of many benchmarks out there, only one indication of how a CPU performs at a given CPU speed, or how well it scales.

Still, very impressive work RGONE and thank you for all the data collected and the incredible undertaking to compile and develop results.

I like cinebench. It really is fun. I usually run the bench custom thread count double that of which the processor has to work with. So on my 8 cores, I set for 16. My Llano (which I just broke 4.07) I run 8 threads. Not sure if it makes a better score or worse, I just know that it looks to be taxing the Cpu more and that's what I like to see. A benchmark that can go beyond the call of an 8 core cpu.....

 

[RGone]

The main reason explained as to the sweet spot being 4.3 was heat. Given proper cooling it appears there is no other reason to stop at 4.3. Early on it was said only serious liquid cooling devices would be practical to go beyond that but we are now seeing air coolers pushing these cpus past the 4.6 GHZ mark, as you mentioned. So as far as I can tell, like any other CPU, these scale well and with a proper cooling solution the sky is the limit.

And bear in mind Cinebench is just one of many benchmarks out there, only one indication of how a CPU performs at a given CPU speed, or how well it scales.

The easy part first. Cinebench R11.5 because with it I had available a CPU Performance Test segment aimed at testing the CPU only. Plus it is a 'free' and available to all to test their systems.

Now the harder part. Yesterday and today I would still personally consider very serious water-cooling as the very best method to cool an FX 8 core cpu up to about 4.9Ghz as Bassnut has his system doing, but a read of his water cooling journey would be in order to see what he had to do to get an FX-8350 to hold up for long time periods. His journey is described in the Water Cooling Forum section. Very descriptive of how much he had to do get cooling at the performance cooling level he attained. Not a cheap journey either. Plus Bassnut is a one-of and not just one of many. So any results attained long term by others is just not a sure thing.

So no the sky is not the limit. It is the limit exhibited by those holding the world records with LN2, DICE and liquid helium since it would likely take a dual stage phase change cooler to handle the high heat loads of the 8 core FX processor. I don't believe a single stage phase has enough capacity to handle the HEAT output of a super clocked FX 8 core. When considered for long periods of time use, LN2, DICE and liquid helium, don't seem useful. Not for application use anyway. Certainly not useful for video editting.

Now this is an observation that I have seen with earlier Phenom 2's. Deneb processors of the 4 core variety. I have put those Phenom2's under single stage phase and they only responded somewhat to -40 temps. Those Phenom2's did not just take off like a rocket when dropped from 0c to -40c cooling. Somewhat faster the CPU ran but for all practical purposes, I would imagine that -100c would be required to see a SUPER CPU speed as compared to cool water or the extended time frame of use one might see with a dual stage phase change cooler.

I can run 5.2Ghz P95 stable for extended periods of time. However I find it hardly worth the effort and expense to remove that much HEAT long term. At this juncture for sure I am talking about using my system as a computer and not just a benching machine to turn in bench scores. It takes some serious refrigeration to run an extended period of time with an FX 8 core above 5.0Ghz. Serious c00l. The faint of heart need not even apply if they plan anything more serious than being able to run a few benches. I prefer being able to actually do work type computing on my rig.

I felt I needed to respond with the information as given above since it is those without preparation that read the words they "want to hear" and off they go. Well in my mind the sky is not the limit. HEAT is the limit and to get rid of the HEAT an 8 core FX can generate, takes very very serious cooling if one actually planned to run an application that takes a few hours to complete. Serious c00ling. Cooling few of us at the DIY level can obtain.

To obtain the trophy beast you have to know your beast. With an FX beast it is the HEAT that is at the heart of the beast. HEAT and loads of it. I noticed "johan45" has run his FX-8350 for a CPU Validation at 5.7Ghz on very good water but even that g00d cooling was not enough for much more than a CPU Validation at 5.7Ghz. Even his 5.7Ghz was not benching ready.

Because I have seen the BEAST up close and personal, I do not take him lightly. Not at all. He can make one crazy, dealing with his beastly heart which is HEAT.
RGone...ster.

 

[storm-chaser]

Sure Cinebench is a great benchmarking tool. And thats a great summation of overclocking in general. How far do you want to take it? Its the question that drives us on - each user / builder / overclocker is subjective in his or her own right. You can have a cool running moderate overclock on an HTPC build or go all way to liquid nitrogen cooling 8GHZ BEAST. Personal intrests / goals, funding, etc, all can play a factor.
And the Cinebench results show this processor scales effectively.

 

[Johan45]

Sure Cinebench is a great benchmarking tool. And thats a great summation of overclocking in general. How far do you want to take it? Its the question that drives us on - each user / builder / overclocker is subjective in his or her own right. You can have a cool running moderate overclock on an HTPC build or go all way to liquid nitrogen cooling 8GHZ BEAST. Personal intrests / goals, funding, etc, all can play a factor.
And the Cinebench results show this processor scales effectively.

I figured this would show you how well it scales for Cinebench in particular. These scores are from HWbot and selected for water cooling only.
Every bench I have run continues to scale upwards with a raise in CPU speed but as RGone says the heat is the killer. Before I upgraded my cooling from an AIO to real water I couldn't get over 5.2 in Cinebench and even very "good" water only gave me another 300 MHz and I reach the thermal limit on my cooling.

 

[RGone]

Johan45, you and mR Clean made a good showing at the HWBot. Good deal.
Bobert...

 

[Johan45]

Don't know if you recall me talking about the guy from TPU I had been poking at but that's D1nky. You'll notice he's right on my heels. He got the bug bad enough to almost single handedly resurrect the TPU benching team.

 

[Mandrake4565]

When considered for long periods of time use, LN2, DICE and liquid helium, don't seem useful. Not for application use anyway. Certainly not useful for video editting.

Funny stuff Gonester!

Johan45, you and mR Clean made a good showing at the HWBot. Good deal.
Bobert...

Just an FYI in my 9.22 Cinebench run, it was possible because it was 8 deg F outside. I was sitting in my office with the door shut and the windows open. I had a fan in one window blowing the cold air in another directing the air to my radiator. I was in full winter gear, hat, gloves, winter jacket, and long johns. It must have been 30f in the room. If you notice the max temp was 57c on the cores and 46 on the socket. I believe my water was at 16 c yes I had a probe in it. I had tried running the bench the day before when it was warmer out and my temps were going into the high 70's on the cores and the bench would just crash. What I learned was my 8350 requires less voltage to run those speeds when under cold temps. I needed 1.6 Cpu v to run 5.3 Ghz when the temps were warmer vs 1.58 @ 5.4 when colder.

 

[Johan45]

I like the long johns part. Amazing what we'll put up with for a few more Mhz

 

[ssjwizard]

What I learned was my 8350 requires less voltage to run those speeds when under cold temps. I needed 1.6 Cpu v to run 5.3 Ghz when the temps were warmer vs 1.58 @ 5.4 when colder.

This makes good sense if you think about it. Temperature and resistance move proportionately to each other. Looking back on my comments back in the spring about these CPUs you will see where I state that temp based stability can be combated with more voltage. The theory behind that is that if the temps rise to a certain point the resistance of the circuit will cause the internal vCore to drop below the point of stable operation at a given clockspeed. Increasing the voltage to compensate for that drop is no different that combating vDroop in the VRM directly.

Admittedly temps above 70C are playing with fire, but I have yet to see one of these chips die on reasonable voltages even when running in the 80c+ range. I would definitely say that above 80C is definitely not recommended, but again I had one FX under those conditions for nearly 30 days straight without any problems.



@RGone
What tool are you using to pull that P-State info, and if you want I can get you the #s from my various 8320 CPUs. I know you were comparing 8350s, but I figured it cant hurt to provide some data on its lil brother if you think it will help.

 

[Mandrake4565]

ssjwizard use Cpu-z for p-states but I believe you need to have it at default settings for them to be correct. Open Cpu-Z click on "about" the click "save report" txt

 

[RGone]

^+1^
What mR Clean said and it needs to be at default or all the P-States will not be listed. You know numb nutts here had to find out. Hehehe.

I think it is an awesome idea to have some P-States for FX-8320s since we then can compare them to the FX-8350s.

I know if the new cpu had the P-States shown on the box and I could pick...well I know what I would pick.
RGone...ster.

 

[caddi daddi]

if it's an fx I would like to see the pstates posted, the more data points we have the better we can figure out of it all.

 

[Kamel]

RGone, I feel like we're good friends even though you probably don't know or remember me. You helped me a great deal with my DFI LanParty (toaster certified hehe) and now you're helping me yet again many years later.

I just wanted you to know I appreciate you, and I appreciate the dedication you've always had to real world system performance, as opposed to just how high can I get this gauge to read. I recall very specifically how you stressed the importance of FSB speed with my DFI board when many people seemed to care less about it. I also recall how people told me I was an idiot for thinking I could clock my xp2500 to 250x10 and run prime95 stable for 8 hours - hah, fooled them! lol

 

[RGone]

Thank you sir for the kind words. Not sure I remember the name as well as your location in FL. That seems to ring a bell with me. Good to hear from you man.

Yes the toaster idea was neat back then. 8 hours of verified prime and I know it was two other captures had to be made for validation. Most of these guys would go nuts trying to get a toaster with todays stuff. H*ll I might also. Hehehe. Thanks again man.
RGone...ster.

RGone, I feel like we're good friends even though you probably don't know or remember me. You helped me a great deal with my DFI LanParty (toaster certified hehe) and now you're helping me yet again many years later.

I just wanted you to know I appreciate you, and I appreciate the dedication you've always had to real world system performance, as opposed to just how high can I get this gauge to read. I recall very specifically how you stressed the importance of FSB speed with my DFI board when many people seemed to care less about it. I also recall how people told me I was an idiot for thinking I could clock my xp2500 to 250x10 and run prime95 stable for 8 hours - hah, fooled them! lol

 

[Kamel]

Thank you sir for the kind words. Not sure I remember the name as well as your location in FL. That seems to ring a bell with me. Good to hear from you man.

Yes the toaster idea was neat back then. 8 hours of verified prime and I know it was two other captures had to be made for validation. Most of these guys would go nuts trying to get a toaster with todays stuff. H*ll I might also. Hehehe. Thanks again man.
RGone...ster.

I believe the screenshot was 8 hrs prime, CPU-Z, and a result from a 3Dmark. I included my signature for old times sake. I don't recall if I ever did the toaster thing twice, but I ended up being at 2500mhz after switching to water cooling. The signature shows my results on air cooling lol.

I was living in either TN or KY at the time, so the person from FL you're remembering probably wasn't me.

Either way, I didn't expect you to remember me, there were many users you helped daily on dfi-street.

 

[RGone]

I think I may have my "toaster" around somewhere. Although I did delete a bunch of older stuff about 8 mos ago. Seeing the "toaster" brings back some good memories of a time past. Thanks man.
RGone...