Return of a Legend - no, not me: AMD FX-8300, 95W TDP 8 core awesomeness :D

[FlanK3r]

yes I agree, this is the right true about stability = stability for each is individual. PC must by OK for usage (someone want BOINC program, some 48h PRIME , next games, next videoencoding...)

[Fr3ak]

Yesterday, I wanted to officially ask ASRock what kind of trash they designed with that board and it ssems like a lot of people have this issue with many different hardware combinations, so it seems like the board's design is flawed. But since none of the official representatives in their support forums logged in in the last 4 months, I didnt start my flamewar.

However, I can see the light at the end of the tunnel, Asus Germany sent the Crosshair V Formula-Z on the way, which will be here tomorrow! Great support by Asus, couldn't be any better - thank you Christian! First, I invited myself to the last ROG Camp here in Berlin in the very last minute and helping me out with the board now - perfect

Can't wait to properly max the FX-8300 out and even fine tune the whole setup with a board that does what it is supposed to do - set the values of my desire and at least attempt to boot with them even if the board thinks I am the biggest retard in the universe

[informal]

Crosshair V Formula-Z is probably the best board for AMD FX . Good luck man!

[Zeus]

Pretty sure you're gonna like that board. The UEFI bios needs a little getting used to but looks so much better than the oldschool bluescreeen bios' we used to have before. Just be carefull with auto settings for voltages, the bios does set some voltages pretty high at auto.

If there's something you need to know about the board, just ask. Not saying i know everything about it but i have at least some experience with this board and i have also owned the CVF nonZ version before.

[mongoled]

I have also owned the CVF nonZ version before.

Seeing you have used both boards, do you have an opinion on if there is a clear difference in CPU clocking ?

I have the non Z board, but saw FlanK3r saying that the Z is better for clocking CPU's, but would like to hear other peeps opinions on this.

[tbone8ty]

Curious for owners of CF5...does the 2nd 4-pin cpu power plug help at all? Its located next the the cpu fan header

[The Stilt]

Curious for owners of CF5...does the 2nd 4-pin cpu power plug help at all? Its located next the the cpu fan header

The 8-pin EPS12V connector is rated for 36A (9A per contact).
After VRM losses this means around 330W consumption allowed for the CPU.

Some of the PSUs with multiple 12V rails have per-rail OCP configured to significantly lower than that (e.g. <25A or so).
Connecting the additional 4-pin ATX12V connector will balance the current draw between rails and therefore prevent the OCP from kicking in under high loading conditions. Also while both of the connectors are connected the maximum current capability will 56A which means the CPU can draw around 517W power (in theory).

Having both of the connector attached does not do any harm.

[Zeus]

On my old CVF i only had the 8 pin connected initially. Later i also connected the 4 pin hoping for better OC but it didn't make any difference. For this CVF-Z i have 'm both connected.

[Fr3ak]

Crosshair V Formula-Z is probably the best board for AMD FX . Good luck man!

Pretty sure you're gonna like that board. The UEFI bios needs a little getting used to but looks so much better than the oldschool bluescreeen bios' we used to have before. Just be carefull with auto settings for voltages, the bios does set some voltages pretty high at auto.

If there's something you need to know about the board, just ask. Not saying i know everything about it but i have at least some experience with this board and i have also owned the CVF nonZ version before.

What can I say? I am totally in love with the Asus Crosshair V Formula-Z. Now I understand what you meant by getting a real board. This thing is a beast. So many great features - not justfor benching, also for daily usage. The AMD board which impressed me the most, followed by the Abit NF7-S

The 8-pin EPS12V connector is rated for 36A (9A per contact).
After VRM losses this means around 330W consumption allowed for the CPU.

Some of the PSUs with multiple 12V rails have per-rail OCP configured to significantly lower than that (e.g. <25A or so).
Connecting the additional 4-pin ATX12V connector will balance the current draw between rails and therefore prevent the OCP from kicking in under high loading conditions. Also while both of the connectors are connected the maximum current capability will 56A which means the CPU can draw around 517W power (in theory).

Having both of the connector attached does not do any harm.

The Stilt?! Thanks buddy for shining some light into the dark with that perfect and easy to understand explanation The PSU I am using right now doesnt have a second 4pin connector and since it is single rail, I dont think I would benefit from using the 2nd one, but I will give it a shot with a different PSU. Theoretical 517W CPU power draw sounds amazing, considering my whole system is being powered by a 450W model


Now onto the news:
After finding my Windows 7 DVD in one of my moving boxes (took me almost 2 hours >_<), I was finally ready to get the ball rolling. It was kind of late when I started and the board comes with a zillion settings, so I am sure, there is some room for improvement once I get to know the board better and understand the behaviour of Piledriver CPUs. Unfortunately my multimeter seems to be broken, so I have to get a new one first in order to make sure, what exact voltage I am running on the Formula-Z.

I skipped the near stock testing and started with 1.45V trying to find out, what settings can be used to get that extra 0.23% better OC, but even after hours, I am still on my search for the holy grail

Anyways, enough of the talking, I guess you only want to see some pictures with overclocking pr0n on them, so here they come (with just a tiny bit of more talking):

Before you draw your own conclusion about the FX-8300, please keep in mind that this CPU is quite an early model, so I dont think it will be comparable to the retail CPUs in terms of overclocking behaviour. And apart from that, I am a OC noob, who can make everything in his hands look bad

Setup:
Asus Crosshair V Formula-Z (Alpha BIOS 0078 and later on 1201)
AMD FX-8300 (batch 1237PGN)
2x 4GB G.Skill TridentX DDR3-2000, CL9
NZXT Kraken X60 (2x 140mm all-in-one watercooling)
Mach Xtreme Technology DS Turbo SSD 128GB
Silverstone Nightjar 450W
Additional fans for VRM area, memory and PSU
Windows 7 Ultimate

Ambient temperature 17C, 1.45V vcore:


Temps at 4.9 GHz with 1.45V and ambient 17C running Cinebench 11.5:


Ambient 17C, 1.45V, 1 core (Cinebench 11.5 takes forever - dont try this at home!):



After playing around at 1.45V for quite some time getting a touch for everything, I opened the window for a -8.6C chilled air intake - benching with cloves and thick jacket is always more fun and sends the right picture to the new neighbours

For unknown reasons, I didnt take a screenshot of the temps, probably because of a brain freeze. The wiindow opens to the wrong side, so the full blast of arctic air runs right over me, while my rig is happely laughing at me in the warm corner. I added some fans to get some of the cold air to the system - I didnt feel like moving half of my furniture around, at 2 AM

But using my improved version, MB temp went down to -1C, CPU idle was 12-14C depending on the voltage. At 1.5V max load temp for Cinebench 11.5 was 28C. That was the point when I regretted having sold my nice, little Single Stage. With it I would have still opened the window in order to get from -50 to -60C, but my FX-8300 seems to scale way better with temp than with voltage.

Ok, more pr0n right here:
-8.6C chilled air, 1.5V:


-8.6C chilled air, 1.55V:


-8.6C chilled air, 1.6V:


-8.6C chilled air, 1.625V:


As you can, see, next to no scaling from 1.6 to 1.625V, so I went for a couple suicide runs and called it a night:

-8.6C chilled air, 1.5V suicide:


-8.6C chilled air, 1.6V suicide:


-8.6C chilled air, 1.625V suicide:


-8.6C chilled air, 1.625V suicide, 1core:

[Zeus]

Woohooo, a fellow cold air bencher!!

Not bad at all buddy!

Told you you're gonna like the CVF-Z! It is indeed a wonderfull board.

One thing though and i can't shoot you for that as you're just starting with this setup...but.. your efficiency in CB is not great. You're at 8.99 at 5267 whereas you should be able to run 9.0 to 9.05 with just 5.2G.

Pretty sure you're gonna get that figured out soon enough so i'm not gonna give you any pointers here as i don't want you to beat my CB 11.5 score.

Did you notice any difference between bios 0078 and 1201?

Oh btw, regarding your Asrock rant, check this new baby from Asrock!: http://www.asrock.com/mb/AMD/990FX%20Extreme9/ 12+2 power phase design!

[Fr3ak]

Cold air benching - the secret OC weapon of the poor xD I am always trying to get the best possible temps. Whole rig in the freezer, radiator on the outside, dry ice chilled liquid - I used to get very creative when it came to lower the temps somehow

I need better memory, that's for sure. I have pretty much 0 knowledge about Win 7 tweaking apart from what you always do since XP, but yes, I noticed the bad efficiency. I also ran SuperPi, but I was shocked how slow Piledriver is on it. My personal best was twice as fast - done with a CPU 4 years ago

I didnt notice any difference between 0078, 0501 and 1201. Might make a difference when going subzero. I am already trrying to find some decent RAM. Until then, I will try to find out, what might make the CPU fly on the CVF-Z, then go for LN2 - everything else is just child's play But from the results I have seen, 8350 seem to clock 100-150 MHz higher on average than 8320 do, so that might be also the case with 8300 depending on what bin AMD uses as basis for the low vcore validation. Being able to boot at 4.8GHz with just 1.4V and run Cinebench at 5GHz with that vcore made me curious on how much more it could do. I will try some more, but I almost touched each and every BIOS setting, so I doubt I will find THE golden setting.

[Zeus]

Ok, thanks for finding out OC potential of the FX8300, it saves me the hunt for one. At OCN there are quite a few guys with an FX8320 and all OC far less than the FX8350 counterparts. So FX 8350 it is.

About your rams, what do you have now? I have to say the Gskill TridentX 2600C10 i have seem to run fine in the board. Since 2600 requires the CPU-NB to run at at least 2600 as well i decided to run them at 2400 and lock in the timings to Cas 9-11-11-25.

I don't know how much impact running tighter timings have on this platform but i think this Samsung based ram does fine looking at Aida's bandwidth test:



Don't know if PSC or BBSE based ram will do better though.

[Fr3ak]

On my phone right now testing RAM, so cant write much.

Are you running 1T or 2T? I was on 2T yesterday and my RAM is the DDR3-2000 CL9 version. It has been running on "auto" for the last 2 years, so I dont know yet what it can do. 1075MHz at 9-11-11-31-45 1T is the max for Cinebench 11.5 at 1.7V. 1102MHz at 1.75V :/

Did you notice the "OMG" feature on that ASRock board? hahaha "omg, it suxors"

[Don_Dan]

My memory is doing fine as long as I didnt even bothher to find its limits. So I am still hoping to hit crazy speeds with it. Until I prove me wrong, there I will stick with it

I need better memory, that's for sure.

I am already trrying to find some decent RAM.

I told you so!

[Fr3ak]

I thought it would be kickass - or better I was hoping for that. Best price was 300 Euro for that kit back then. Good that I ran it on auto settings for all this time xD

So MemMaster Don, what do you recommend for AMD?

[Zeus]

1T mate, over ~2540, you're gonna need 2T.

Daniel, go easy on Oliver, he's trying to get back after 3 years, we should cut him some slack. For now.

[Fr3ak]

1102MHz is the max with 1.75V on 1T and CL9 for me. Testing CL11 now for a laugh, but that memory sucksors. I officially admit, that you were right from the start Daniel (which I didnt doubt, but I was believing in a wonder). So I have proven myself wrong (2nd time in a short period after my board choice desaster - gah).

Eugene, what VMem are you using? Maybe I am too easy on voltage? My vcore max on water + chilled air was 1.75V, but since it didnt scale, I stopped at 1.625V. For that memory, I will try to see if 2.1V is of any help. If it blows up, then I will be on LN2 with new memory even faster

[Zeus]

Well, Samsung is actually low voltage ram but the chips seem to scale all the way up till 2.0V. For all the benches i've shown i haven't been over 1.67V set in bios. Now this board overvolts ram by quite a lot! 1.6V set in bios is 1.65V measured on the measuring pad with a DMM.

1.67V was just to be sure, i reckon i would have never needed anything over 1.63V. Not saying it's best ram to use but so far it has run 100% trouble free in this set up.

[Fr3ak]

I am at a set 1.8V CL11 2T now and I can already stop right here. Basically no scaling at all. But at least I now know for sure. Daniel, I need new mem. Why on earth didnt you tell me that? ;D

[Don_Dan]

I thought it would be kickass - or better I was hoping for that. Best price was 300 Euro for that kit back then. Good that I ran it on auto settings for all this time xD

So MemMaster Don, what do you recommend for AMD?

Are you still using the Trident kit with Elpida ICs you posted in my thread?

I'm not sure what's the best kit for AMD, but a Samsung kit like the one Eugene is using should be a pretty good choice.

Daniel, go easy on Oliver, he's trying to get back after 3 years, we should cut him some slack. For now.

I'm going easy on him, I didn't ask for LN2 results yet.

1102MHz is the max with 1.75V on 1T and CL9 for me. Testing CL11 now for a laugh, but that memory sucksors. I officially admit, that you were right from the start Daniel (which I didnt doubt, but I was believing in a wonder).

One does always need better memory...

I am at a set 1.8V CL11 2T now and I can already stop right here. Basically no scaling at all. But at least I now know for sure. Daniel, I need new mem. Why on earth didnt you tell me that? ;D

1.8V is probably too much for that kit.
Okay, okay, it's my fault. Try the TeamGroup 2400 CL9 kit, it should treat you well und you should be able to get it easily...

[Zeus]

Oliver, can you test if you can run 300 FSB on your board for me please? I was hoping it was my chip holding my FSB back but it's the same with this chip. Things come to a grinding halt at 275 for me. Maye it's just the bios?

[Zeus]

Sorry for double post but i just flashed back to bios 1002, same thing. Still no 300 FSB. EZ flash lets me flash back without having a moan about an outdated bios, good thing Asus!

[vario]

You have a good chip there, for comparison i can say that to finish cinebench at 4.9ghz i must throw 1.55v at it with some other options set at ultra high (LLC and the like), 5.0ghz is just for show, cant finish cinebench at any voltage.If these 8300 will show up in retail i may go for a change.
BTW winter coling rules, allowed me one time to see a 3ghz POST with my barton mobile back in the day :-). -18 in the night ;].

As for the samsung ram, if you want to be sure of the results dont get the basic OEM, i went that way and got burned , no scaling past 1.57v and cant get even 2133 fully stable even (no matter the timings), well its 1333mhz ram so maybe i shouldnt cry about it .

@FLANKER

Its absolutely not true about PRIME95 ,ive tested this thing in and out, it works perfectly good with a piledriver and bulldozer chips, and all other chips.Its a continuously developed piece of software with a big and good support community, its not even just a "hammer" to beat on a cpu, its design to do something valuable.That being said, its HARD, as it turned out it can beat a stock cpu with a stock bios settings, and even CnQ state changing can trigger a faulty operation.But its not fault of the software but motherboard/bios/amd issue, and every system can be tuned to be stable.After my initial shock that my cpu isnt stable at stock, ive read more about it, and did a LOT of testing,either way with OCCT you will get similar results.
To be honest after lotsa reading some people just arent happy with it because it doesnt allow them to brag with higher OCs ,so if results dont reflect their goals, they change the parameters those results are being picked at...
And im not doing 48hprime loads too if its gonna fail, its gonna fail pretty quickly with PD.

On teh stability front, samsung IC`s (at least mine DH0`s) are tricky to get fully stable also, ive run happily 2133 10-10-10 for a week(did some hours memtesting ofcourse) until my friend that i burned few dvds for, called me up and said two of them had CRC errors ;[.Prime blend+winmemtest plus running farcry at the same time seems to be the quickest way to finding errors ,still, can be few hours.

[FlanK3r]

Nice Freak man! Your score will be better with higher HT clock, FX has default 2600 MHz and can up working to 3200 MHz.

[Zeus]

Better CB score with higher HT link? Really?