Llano sans GPU. AMD Athlon II X4 631

Yeah, but that is an A6, is the IGP still active? If so that thing is eating 1/2 the memory bandwidth and adding to the TDP. 4.2ghz is still impressive if the intergrated GPU is still active.
 
Zarathustra[H];1037910297 said:
Ahh, so FM1 has fewer PCIe lanes than AM3?

The reason I ask is that my AM3+ board has enough for 2x 16x SLI or CF... If I do tri-SLI/CF its 16x 8x 8x. Not sure what happens if I go 4 way. The manual doesn't document this.

Here is a block diagram from when TechReport reviewed A75 boards from Asus, Gigabyte and MSI: http://techreport.com/articles.x/21207

Yes, FM1 only has 16 (PEG) + 4 (GPP) PCI-E lanes originating from the CPU socket and the A75 FCH has an additional 4 GPP PCI-E lanes from it.

As far as I know, AMD never really designed Llano for anything else than "Dual Graphics" mode, seeing as not many FM1 board makers support dual x8 PCI-E modes.

Basically, FM1 has the same amount of lanes as an AM3 board using the G, GX, or X chipsets.
 
the guy from the link is getting some funky bench results.
for example he's getting 24sec on SuperPi 1M (@ 4.2GHz) which is horrible.

on my stock PII555 (3200MHz), i get 21.8secs.
on same PII555 (4core @ 4100MHZ), i get 16.7secs.

something isn't right with this dudes system.

1M superPi may fit the entire problem in PIIs L3 cache.
 
Good news and bad news.

Bad news: i copied the exact bios settings from chinese website and it wont post. just beeps. i lowered FSB by 5 until 155FSB then it boots to windows.

SuperPI 2.6ghz = 28 seconds
Superpi 2.86ghz = 25 seconds (110x26)
SuperPi 3.88ghz = 19seconds (155x25)

All voltages stock, retail heatsink. also -- retail heatsink is terrible - its same as sempron processor which has 3 less cores. CPU z showing CPU 1.0V even though i have cool n quiet, and other power saving options its not listening... i might have to force a Pstate?

I have the northbridge set to the lowest setting, so ill start moving that up a notch at a time and see what happens. im going to try 3dmark06 now.
 
That is possible, the anand guy was having P-State issues as well and couldn't figure it out.

Anyone want to give us some i5 numbers?
 
Windows 7 index shows score of 7.5 which doesnt mean much to me.

I kinda feel like its throttleing the CPU , cpuz is reporting 3.88ghz but i think its more like 3.4ghz realistically...

My benchmark scores didnt increase by a large increment that i expected, then when i went from 100 to 110.
 
Its throttling the CPU for sure. Boot into windows at 3.88ghz, 3.6ghz, 3.5ghz benchmarks dont change until i drop below 3.6.

fritz chess:
150x25: 8892
140x25: 8452

Super Pi
26x100 = 28
25x155 = 20
25x150 = 20
25x145= 20
25x140= 21

So im guessing no matter what it reports im hitting a wall around 3.6ghz.

I am going to try a few more bios settings, but id say this is a bust so far.
 
Good news, i was able to overcome my wall. I just set the FSB 150x26 (3900mhz) now chess score is up to 9143...this forced me to turn ram down a notch. HW monitor if reporting correctly -- under full load its going to 45C with retail heatsink at 3.64ghz. Im impressed.

What i did to make it stable was go into the Power state and set Pstate to 0. this forces voltage to hard set and not jump around. going to keep bumping it up and see what happens from here.

Now im at 155x26 and fritz is up to 9579. Yay! i got it working!@!
 
cant get 160x26 to post. will try voltage bump. my board says 1.41 in bios, but HW monitor is showing 1.36V in windows...
 
I dont feel like the heatsink is keeping me back because if it posts, it boots to windows and runs stable -- otherwise doesnt post. im at 152x26 -- i think this is about all im going to get out of it. I tried increase NB and cpu vcore but cant post at 155x26. tried 160x25...
 
In the end, im happy with this cpu for $89, but im unhappy with the MSI a75ma-g55 for $89.. I really wish i got a full size ATX with more gap between the slots. a few minor qwerks layout, but even overclocked -- nothing on the motherboard or even heatsink on CPU is hardly warm to the touch! and retail heatsink is pathetic...

Just scored 19961 on 3dmark06 with one 5830@900mhz/1050memory -- interesting CPUZ says im at 3968mhz. 3dmark06 says im at 3783. strange things still happening...

CPU score 4899
SM 2.0 8073
SM 3.0 9893
 
Well sad news on this end. It looks like I got a dud board from newegg. Comp powers on for 1/2 second then turns off. PSU works when I switch back to the old board. Tried resetting CMOS via jumper and battery and still no go. Any ideas?
 
I won't say I have tried everything. I tried several variations of 170FSB -- mostly used 5mhz increments though. i bumped NB voltage to 1.2 CPU voltage to 1.45, ram was never overclocked, i set the timings to 9-11-9-27-40 as specificed by manufacture with voltage to 1.65 Ram was always 1866mhz or lower.
 
Well sad news on this end. It looks like I got a dud board from newegg. Comp powers on for 1/2 second then turns off. PSU works when I switch back to the old board. Tried resetting CMOS via jumper and battery and still no go. Any ideas?

bad ground? check the posts, make sure there isnt metal touching your motherboard where it shouldnt be? reseat ram.. ..
 
What is your ram set at? Have you tried to jump to the 170s? Maybe there is a hole?

Can you have that guy on anatech with the a4-3400 overclocked to 4.7ghz provide some simple benchmarks to prove the cpu is actually working on displayed speeds?

super pi / fritz chess should tell us if its working. I had my system post over 4ghz a few times, but benchmarks remained unchanged until i change pstate to 0. which case i couldnt get higher then 3952mhz
 
Here is a block diagram from when TechReport reviewed A75 boards from Asus, Gigabyte and MSI: http://techreport.com/articles.x/21207

Yes, FM1 only has 16 (PEG) + 4 (GPP) PCI-E lanes originating from the CPU socket and the A75 FCH has an additional 4 GPP PCI-E lanes from it.

As far as I know, AMD never really designed Llano for anything else than "Dual Graphics" mode, seeing as not many FM1 board makers support dual x8 PCI-E modes.

Basically, FM1 has the same amount of lanes as an AM3 board using the G, GX, or X chipsets.

Hmm...

I would LOVE to see a Llano core with the GPU disabled in an AM3 (or AM3+) package...
 
Zarathustra[H];1037913843 said:
Hmm...

I would LOVE to see a Llano core with the GPU disabled in an AM3 (or AM3+) package...

The problem there is that Llano does not support HyperTransport in any shape, form or fashion.

Question: If AMD did make a 32nm 4+ (8, 10, 12, 16?) core 'Stars' CPU, how high would it have to clock to even come within 10-15% of SB? Also, would it be possible for AMD to add AES, AVX and full SSE4 to the Stars FPU?.
 
I got my FM1 machine back and am pretty sure my mobo crapped the bed.

Anyone had RMA with ASrock before? decent?
 
The problem there is that Llano does not support HyperTransport in any shape, form or fashion.

Question: If AMD did make a 32nm 4+ (8, 10, 12, 16?) core 'Stars' CPU, how high would it have to clock to even come within 10-15% of SB? Also, would it be possible for AMD to add AES, AVX and full SSE4 to the Stars FPU?.

I dunno, you could do it in an independent translation system but that's not quite optimal nor is re-engineering to support the new instruction set especially if the K-10 is optimized to hell. You'd basically have to roll back a bunch of stuff, redesign, and reoptimize so you're almost better off doing something form scratch.
 
ive owned many asrock boards, but never had to RMA one... let us know how that goes.
 
The problem there is that Llano does not support HyperTransport in any shape, form or fashion.

Question: If AMD did make a 32nm 4+ (8, 10, 12, 16?) core 'Stars' CPU, how high would it have to clock to even come within 10-15% of SB? Also, would it be possible for AMD to add AES, AVX and full SSE4 to the Stars FPU?.

That is too bad. I hadn't read up on the design, but I was thinking/hoping it - electrically and logically - was equivalent to an AM3 design, just with extra pinouts for the GPU.
 
Zarathustra[H];1037914749 said:
That is too bad. I hadn't read up on the design, but I was thinking/hoping it - electrically and logically - was equivalent to an AM3 design, just with extra pinouts for the GPU.

Actually, FM1 has fewer pins than AM3: 905 vs 938/939/940/941/942 for AM2(+)/3(+). I guess a majority of pins were just repurposed for PCI-E and the FCH interconnect.
 
Guys, i noticed a discrepancy on the chinese website. Possible photoshop or fraudulent pictures.

In the picture of the BIOS. Almost all of his settings are on auto -- which is a stretch to believe considering its a 1.8ghz overclock. But the biggest thing that jumps out at me is his NB multiplier is set to auto. I have the same motherboard and when i set my FSB to 170 and NB multiplier to auto this makes my northbridge frequency 1083 or 1183 -- something REALLY high. in the screenshot his northbridge is 680mhz which is the lowest possible setting for NB multiplier (not auto which seems to be highest possible setting)

I saw another website that tested an AMD engineering sample 3870 with unlocked multipler -- they were also only able to get to 3.9ghz -- i have a feeling thats about the limit for these llanos with current FM1 motherboards...
 
That's annoying... So the final overclock is roughly 150%?

That's not terrible, about C2D level right there. You should try and bench against some other computers.
 
I tried cpu multiplier of 20 and FSB of 170 3.4ghz with voltage bumps and it doesnt work. i tried multiplier of 29 -- but performance didnt change from 26.

im thinking really serious about sending this motherboard back and getting a full size one, with ample gap between the 16x slots for my 5830s...
 
I picked up the Frys combo deal on the A6-3650 + MSI A75MA-G55 for $149.99 - $20 MIR today. I can only get it up to 3.7GHz (143x26) so far. Graphics get wonky over 145 HTT.
 
I picked up the Frys combo deal on the A6-3650 + MSI A75MA-G55 for $149.99 - $20 MIR today. I can only get it up to 3.7GHz (143x26) so far. Graphics get wonky over 145 HTT.

People are testing the version without graphics (Integrated graphics processor) because this Chinese website showed some screens of it running at 4.4ghz. People speculated that this may be due to the disabled IGP, thus less TDP which results in higher OCs.

So far tests haven't proved very favorable : /
 
Copy and pasted from my original post @ AT:

I think my board is maxed out at 161 HTT. What I've discovered so far is that I'm able to hit 3.7GHz using the iGPU (I'll have to keep playing around to see if I can hit higher) and 4.18GHz using discrete video. I'm very surprised at how little power the A6-3650 sips at idle. It is idling at 46w per my Kill-a-watt meter (64GB M4 SSD, Geforce 210GT, and Antec Neo 520c PSU) and while running Cinebench it tops at 109w. Gaming uses around 80w of power consumption (Using integrated graphics). I'm really happy with this purchase. Anyway, here are my results of tinkering:

Stock:

i-GjSRw5x-X2.jpg

i-wXdZ8m3-X2.jpg


4.18GHZ:

i-fkbX3xp-X2.jpg

i-rTNSLzd-X2.jpg

i-RGSfPfG-X2.jpg

i-BNFvdZQ-X2.jpg

i-4fJL3SX-X2.jpg



I'd say a 1.5GHz overclock is damn fun on a budget. i3-2100, you are no match for the Llanos once overclocked. And power consumption is really respectable as well. I'm really thinking about selling my i3-2100 + mobo to get another A6-3650 combo because it barely uses more power but has a better iGPU. If you have a Frys near you, jump on the A6-3650 + MSI A75MA-G55 deal.
 
I do have a kill a watt. ill test it today. Rich, you did this on stock voltage? did you bump nb or cpu voltage at all? sometimes my system will boot at FSB over 155, but sometimes it just beeps three times. What I find really strange is if my computer boots, it runs fine....
 
My computer doesn't cold boot @ this high of HTT overclock, so I reset CMOS, boot into BIOS, load my overclock setting and reboot and it posts. The 3 beeps sometimes occurs (memory error in BIOS). I bumped up cpu-nb to 1.22v, CPU voltage was left on auto, but I'm testing it at 1.407v setting in BIOS now.
 
People are testing the version without graphics (Integrated graphics processor) because this Chinese website showed some screens of it running at 4.4ghz. People speculated that this may be due to the disabled IGP, thus less TDP which results in higher OCs.

So far tests haven't proved very favorable : /

Less TDP, and it seems the IGP craps out before the CPU. The other guy above got to around 3.7Ghz with IGP, and almost 4.2Ghz on descrete. This it really convincing me that that 631 is capable of 4.4 easily because the IGP is hardware disabled.
 
I dont agree. The highest i can reach with the 631 is 3952mhz and i wouldnt call it rock stable by any stretch. 75% of the time it refuses to post at these settings. Rich can only reach his overclock by setting his bios back to default settings and then chosing overclock settings and doing a warm reboot (not from a cold start). Granted I am on the stock heatsink, but these chips dont seem to run hot at all. I tried bumping voltage from 1.36 to 1.44V and the highest temps i could get in windows running benchmarks was 49C according to HW monitor that was with 1.3ghz overclock too.

At this point i dont know if its the motherboard or limitation of the FSB on chip holding us back. Still for $89 i paid for the CPU is perform really well at 3.64ghz (my 24x7) setting on stock voltages.
 
Less TDP, and it seems the IGP craps out before the CPU. The other guy above got to around 3.7Ghz with IGP, and almost 4.2Ghz on descrete. This it really convincing me that that 631 is capable of 4.4 easily because the IGP is hardware disabled.
If disabling the IGP magically makes a CPU a monster overclocker then Intel should just disable the GPU in the 2600k and wed all be hitting 5.5ghz.
 
If disabling the IGP magically makes a CPU a monster overclocker then Intel should just disable the GPU in the 2600k and wed all be hitting 5.5ghz.

I'd assume there is very little commonality between AMD's on chip IGP solution and Intel's.

The closest we may get on the blue side will be the SB-E chips. Intel makes igp disabled 1155 Xeons but they have a locked multi and no bus clocking.
 
Back
Top