- Apr 27, 2000
- 22,746
- 12,751
- 136
TL;DR summary: My Athlon II x4 635 was good for 3.7 ghz, and it sucks at memory overclocking. So far.
edit: Bios v1.10b2 has been a joy. I've started from scratch on overclocking, and have easily stabilized 3.8 ghz. Working on 3.85 now . . . it may take 1.51v vcore to do it, but the chip is priming at 49C with those volts so I'm not too worried about it just yet. 3.9 ghz seems to be a no-go (will not prime for more than a few minutes, even at 1.52v vcore). Will work on NB/RAM speeds once I get this sorted out.
edit edit: It's been easy to stabilize the cores at 3.8 ghz isolated, but in a consolidated overclock with OCed NB and memory, it is not stable. Even 3.744 ghz has been a pain-in-the-ass to stabilize. At least I can boot past 3.7 ghz, but getting it prime stable is another matter altogether . . .
Nice try bios rev 1.10b2, but you can't fix Propus. Well, not entirely.
Not many people around here talk about Propus overclocking, and apparently that is for good reason. While all the AMD OCers are atwitter about C3-stepping 965BEs or unlockable x2/x3s, I went and got myself an Athlon II X4 635 and paired it with an MSI 790FX-GD70 and G.Skill Pi series DDR3-2200.
Yes, I know its a bit silly, but I had my reasons that I'll ramble on about sometime in the future once I finish playing with beta BIOSes and suchlike on this board.
Back to the chip.
First impressions:
1). The HSF that shipped with my retail box Athlon II x4 635 is physically identical to the HSF that came with my s754 Sempron 2800+ (Palermo core, 256k l2) from 2005. I used the poor little thing for a bit and the temps were so high that I underclocked the chip and undervolted almost everything I could to get temps down to a managable level. Ridiculous.
2). When lapping the 635 (yes, I lapped it), it became readily apparent that the copper IHS was horribly uneven. Even more interesting was that AMD managed to "level off" the IHS by laying down extra nickel plating on the recessed pits present on the copper IHS. So, the unlapped chip was (mostly) flat, but the nickel plating had variable levels of thickness. My x2-3600+ had both an uneven copper and nickel surface, and was much harder to lap.
I also cut myself on the corners of the CPU and bled on the sandpaper . . . it didn't seem to make lapping any easier though. Imagine that.
3). It's a C2-stepping chip. Here I was thinking that one of AMD's most recently-released quads would *have* to be a C3-stepping chip . . . but noooooooo. Grumble grumble. Had I done proper research I would have known this in advance.
Overclocking:
The chip runs pretty cool until you start playing with vcore. Lapped 635 + Shin Etsu x23-7783D (still going after 3 years in storage) + Noctua nh-d14 = 27-29C idle, 36-37C load. Ambient is about 26-27C. Got the same temps at 3.35 ghz @ stock vcore.
At maximum overclock (3.7 ghz, 1.43 vcore, some other voltages jacked up, 2.85 gz NB, DDR3-1520 6-7-5-15 1T), Prime gets it up to about 46C.
All that extra cooling doesn't seem to do much to compensate for the fact that a). it's a Propus and b). it's a C2-stepping chip. At stock vcore (1.3v), it tops out at around 3.35ghz. It takes 1.43v vcore to reach 3.7 ghz.
The Northbridge can be stabilized at speeds as high as 2.95 ghz, which sort-of proves my theory that Propus should have a NB that is more overclockable because it doesn't have any L3. Sadly, the lack of L3 means that NB overclocking does a lot less for Propus than it does for Deneb. It also has a maximum NB multiplier of 10x which certainly makes life interesting.
Then there's the RAM . . . good old Propus can't post at RAM speeds higher than DDR3-1700, which is fantastic when you've gone out and bought DDR3-2200. Timings go straight to hell past DDR3-1620 or so (9-9-9-28 or worse). I'm told that BIOS rev 1.10b2 for the GD70 helps with RAM speeds but I have yet to test anything other than the stock BIOS (1.7).
As an added bonus, my chip's IMC refuses to POST if you:
Use the 1:2 (DDR3-800) RAM multiplier
Use CAS 4, 5, or 12 memory timings
As you can imagine, this chip is just a barrel of monkeys. And to think, I BLED FOR THIS CHIP!
In reality, though, I shouldn't be too bitter. Compared to my x2-3600+ I get:
twice as many cores
twice the l2 cache per core
500 extra mhz per core
improved ipc
And I have a pretty good ready-for-Thuban(Bulldozer?) platform set up in case I decide to replace this balky little thing.
Maybe once I try some new BIOSes and play with suicidally-high voltage levels, I'll try benching it, but for now it's nothing particularly wonderful. Alas. I *did* beat Anand's 635 OC by 200 mhz though . . .
edit: Bios v1.10b2 has been a joy. I've started from scratch on overclocking, and have easily stabilized 3.8 ghz. Working on 3.85 now . . . it may take 1.51v vcore to do it, but the chip is priming at 49C with those volts so I'm not too worried about it just yet. 3.9 ghz seems to be a no-go (will not prime for more than a few minutes, even at 1.52v vcore). Will work on NB/RAM speeds once I get this sorted out.
edit edit: It's been easy to stabilize the cores at 3.8 ghz isolated, but in a consolidated overclock with OCed NB and memory, it is not stable. Even 3.744 ghz has been a pain-in-the-ass to stabilize. At least I can boot past 3.7 ghz, but getting it prime stable is another matter altogether . . .
Nice try bios rev 1.10b2, but you can't fix Propus. Well, not entirely.
Not many people around here talk about Propus overclocking, and apparently that is for good reason. While all the AMD OCers are atwitter about C3-stepping 965BEs or unlockable x2/x3s, I went and got myself an Athlon II X4 635 and paired it with an MSI 790FX-GD70 and G.Skill Pi series DDR3-2200.
Yes, I know its a bit silly, but I had my reasons that I'll ramble on about sometime in the future once I finish playing with beta BIOSes and suchlike on this board.
Back to the chip.
First impressions:
1). The HSF that shipped with my retail box Athlon II x4 635 is physically identical to the HSF that came with my s754 Sempron 2800+ (Palermo core, 256k l2) from 2005. I used the poor little thing for a bit and the temps were so high that I underclocked the chip and undervolted almost everything I could to get temps down to a managable level. Ridiculous.
2). When lapping the 635 (yes, I lapped it), it became readily apparent that the copper IHS was horribly uneven. Even more interesting was that AMD managed to "level off" the IHS by laying down extra nickel plating on the recessed pits present on the copper IHS. So, the unlapped chip was (mostly) flat, but the nickel plating had variable levels of thickness. My x2-3600+ had both an uneven copper and nickel surface, and was much harder to lap.
I also cut myself on the corners of the CPU and bled on the sandpaper . . . it didn't seem to make lapping any easier though. Imagine that.
3). It's a C2-stepping chip. Here I was thinking that one of AMD's most recently-released quads would *have* to be a C3-stepping chip . . . but noooooooo. Grumble grumble. Had I done proper research I would have known this in advance.
Overclocking:
The chip runs pretty cool until you start playing with vcore. Lapped 635 + Shin Etsu x23-7783D (still going after 3 years in storage) + Noctua nh-d14 = 27-29C idle, 36-37C load. Ambient is about 26-27C. Got the same temps at 3.35 ghz @ stock vcore.
At maximum overclock (3.7 ghz, 1.43 vcore, some other voltages jacked up, 2.85 gz NB, DDR3-1520 6-7-5-15 1T), Prime gets it up to about 46C.
All that extra cooling doesn't seem to do much to compensate for the fact that a). it's a Propus and b). it's a C2-stepping chip. At stock vcore (1.3v), it tops out at around 3.35ghz. It takes 1.43v vcore to reach 3.7 ghz.
The Northbridge can be stabilized at speeds as high as 2.95 ghz, which sort-of proves my theory that Propus should have a NB that is more overclockable because it doesn't have any L3. Sadly, the lack of L3 means that NB overclocking does a lot less for Propus than it does for Deneb. It also has a maximum NB multiplier of 10x which certainly makes life interesting.
Then there's the RAM . . . good old Propus can't post at RAM speeds higher than DDR3-1700, which is fantastic when you've gone out and bought DDR3-2200. Timings go straight to hell past DDR3-1620 or so (9-9-9-28 or worse). I'm told that BIOS rev 1.10b2 for the GD70 helps with RAM speeds but I have yet to test anything other than the stock BIOS (1.7).
As an added bonus, my chip's IMC refuses to POST if you:
Use the 1:2 (DDR3-800) RAM multiplier
Use CAS 4, 5, or 12 memory timings
As you can imagine, this chip is just a barrel of monkeys. And to think, I BLED FOR THIS CHIP!
In reality, though, I shouldn't be too bitter. Compared to my x2-3600+ I get:
twice as many cores
twice the l2 cache per core
500 extra mhz per core
improved ipc
And I have a pretty good ready-for-Thuban(Bulldozer?) platform set up in case I decide to replace this balky little thing.
Maybe once I try some new BIOSes and play with suicidally-high voltage levels, I'll try benching it, but for now it's nothing particularly wonderful. Alas. I *did* beat Anand's 635 OC by 200 mhz though . . .
Last edited: