Mushkin 2GB XP4000 Redline CE-5 – Test Results

Hey guys, I recently got a set of the “new” Redline XP4000 (part #991493) to test. This stuff uses Infineon CE-5 chips which seem to clock better than CE-6 and as far as I can see have no 3d stability issues. The RAM came from Mushkin and is a retail set, not hand picked; I’ve been told the recent batch of CE-5 is clocking quite well. I am going to share my results, suggested timings/tweeks, and impressions of the mem. First off a pic of the RAM:

http://members.shaw.ca/zryan/ce5pic-1.JPG


Test bed:
DFI LP NF4 Ultra D (BIOS 0704-2BTA)
FX-57
BFG GeForce 7800GTX @ 540/1385
OCZ PowerStream 520W
Windows Server 2003
Cooling - Custom H20

I started at DDR510 and 2.65v and once I got the drive strength and secondaries nailed down moving up from there was easy. Unlike XP4000 UCCC chips, CE-5 scales somewhat with voltage. 2.67v got DDR560, 2.77v got DDR566.

Secondary timings and drive strength are important w/ this RAM, the wrong drive strength or 1 secondary off and you cant boot. See the settings below or my tweeker screenie to see what settings worked for me. *Note these settings are the TIGHTEST possible that still maintain stability at the speeds I am running the RAM. Some can be relaxed a bit, YRMV. Also what settings work for me may differ slightly for you because of system differences.*

BIOS Settings:
Command Per Clock Enable
TCL-3.0
TRCD-03
TRAS-06
TRP-2
TRC-7
tRFC-13
TRRD-02
tWR-02
tWTR-01
TRWT-03
tREF-4708
tWCL-01
DRAM Bank Interleave-enabled
DQS Skew Control - auto
DQS Skew Value- 0
DRAM Drive Strength - Level 7
DRAM Data Drive Strength - Level 3
Max Async Latency- 8ns
DRAM Response Time - Normal
Read preamble Time- 5ns
Idle Cycle Limit- 00clks
Dyanamic Counter - Enable
r/w Queue bypass: 16X
Bypass Max - 07X
32 Byte Granularity - Auto
Vdimm-2.6v + 0.03v in BIOS

OCing summary

3-3-2
2.69v – 280Mhz
2.80v – 283Mhz
2.88v – 287Mhz

2.5-3-2
2.95v - 260Mhz

*Notes*

• Voltages are actual as measured by DMM
• Results are 32M and 3DMark 05 stable
• See next post for screenshots

280Mhz – 2.67v (2.69v DMM)

http://members.shaw.ca/Eldonko1/ce5-6.JPG

280Mhz 3D

http://members.shaw.ca/zryan/ce5-5.JPG

280 Mhz Prime

http://members.shaw.ca/Eldonko1/ce5-7.JPG

283Mhz – 2.77v (2.80v DMM)

http://members.shaw.ca/Eldonko1/ce5-8.JPG

283Mhz 3D

http://members.shaw.ca/Eldonko1/ce5-9.JPG

283Mhz Prime Blend 13 hours

http://members.shaw.ca/Eldonko1/ce5-12.JPG

Bandwidth

http://members.shaw.ca/Eldonko1/ce5-10.JPG

287 Mhz - 2.85v (2.88v DMM) 32M & 3D stable

http://members.shaw.ca/Eldonko1/ce5-15.JPG

CAS2.5. 260 Mhz - 2.93v (2.95v DMM) 32M & 3D stable

http://members.shaw.ca/Eldonko1/ce5-17.JPG

Hell yeah, I got the same ram. Haven't oc'd yet but this thread is just what I was looking for.

I'll try and contribute any way I can.

Wow, those are some sweet sticks, 3d stable too.

Quote:

---

Originally Posted by Vincentvega18

Wow, those are some sweet sticks, 3d stable too.

---

really good ce5....would be nice to see if it's average or only lucky sticks !!

Yeah, good stuff! How do they do at Cas2.5? Mine only gain ~4-5MHz by going from cas2.5 -> 3.. maybe yours will too :woot:

Also did setting ICL to 16 make a difference to you? That's the 1 setting I haven't tweaked yet, wanting to know if it will get me some extra mhz

Wow, only ~2.8v for 283MHz while keeping your tRP at 2? Damn, nice sticks! I'm also interested in seeing what they could do at CAS2.5.

donko, this is run of a IC's that clock really well from What Tom tells me. While they will do their best to keep clocks high, I think it would be prudent to say that this is one kit, and other weeks of production may or may not clock as well.

sweeeeet...... I have those sticks in my office rig, luv em!!!

Damn nice...good sticks man. :)

Yeah JNav89GT, Brian told me the latest batch of CE-5 all clocked very well, but I do agree this doesnt represent all CE-5. However, I think its pretty safe to say that newer modules will probably clock a little better on average. Of course its not as simple as saying CE-5 is all going to clock great, like in all OCing it all depends on the week and batch.

I will test CAS 2.5 and more voltage as soon as I can and keep this thread updated. :)

Quote:

---

Originally Posted by Repoman

Yeah, good stuff! How do they do at Cas2.5? Mine only gain ~4-5MHz by going from cas2.5 -> 3.. maybe yours will too :woot:

Also did setting ICL to 16 make a difference to you? That's the 1 setting I haven't tweaked yet, wanting to know if it will get me some extra mhz

---

ICL is a type of setting that gives better performance the tighter you run it. I use 16 because it is as tight as it gets without causing instability.

I wonder what IC's are in my ZX.. CE-5/6, Aeneon or what. They only like 2.58v, no more or less, and they don't like CAS3. strange stuff :stick: but I'm happy

Looking forward to CAS2.5 tests though, if they're anything like mine you'll get 275+ :eek: That's be awesome

Nice results! :toast:

I'm testing some other CE-5 based memory here, and it runs superpi etc very easy. But running 3dmark 2005 is no go.
It could be something with the ATI drivers... Techpowerup had the same problems with their 2GBHV set from Gskill (ce-5).

I've been told by Mushkin that the top-end of these ICs is more dependent on the date code than the ns rating, thus if its a good clocking batch you will high 3d stable clocks no matter if its ce-5 or ce-6. :)

rgr
I talked to Brain again today and he gave me some good news!
All of their current batch is clocking really well. Now he's not going to confirm each kit will do these clocks u have, I also got a kit and am doing 280 which I think will be tuneable to 24/7 stable. Very dependent on timing and I just got my Expert DFI tune complete last night after running 280 on MSI Diamond SLI 16x board for a few days. Anywho, Newegg has one kit of old batch left, and the next batch they get in will be the new batch. I asked about different revision for the better clocking modules, but he said it would be a logistical nightmare to do, b/c he's not sure if this batch is a sign of things to come, or just a lucky buy.

I don't see this stuff on newegg.com? They only have 4 2GB kits on there. And this isn't one of them :stick: I assume it's because the last set sold out and they're gonna put these on.

Very tempting, how much do they cost?

Kits are being shipped out to ABS/ Newegg today. They just ordered it this afternoon.

Tom

Very impressive RAM, isn't it Eldonko? :up:

results seem promising, as you say it scales with volts has anyone tryed say 3.0v+ on this ram?

I havent even tried 2.85 yet lol, but I will! Work been killer the past week =/

Yeah Bachus, its killer mem! :woot:

Quote:

---

Originally Posted by Eldonko

ICL is a type of setting that gives better performance the tighter you run it. I use 16 because it is as tight as it gets without causing instability.

---

The ideal ICL-value all depends on the application-specific memory access pattern. The idea behind high values is to "gamble" that the same row will be accessed again within the timing interval, thus eliminating the tRCD-delay on the next transfer from that same row. The worst-case scenario is when the requested row is in the same bank as the active row (page conflict). A high value would hurt performance in such cases, because the active row would have to be closed first, i.e. the commands would be tRP - tRCD - tCAS. If the row had already been precharged (low ICL, or autoprecharged read), the involved parameters would be tRCD and then tCAS.

Lowering the ICL will NOT lead to instability if everything works the way it's supposed to, on the contrary. Think about it: If the value is too high, it'll cause the data (charges) to expire from the sense amplifiers before they are written back to the memory cells. The sooner the data are written back to the row, the smaller the risk of data corruption. Autoprecharged reads equal a value of 0, and I've never heard of stability issues related to that technique.

Is that some thing we cay expect if we buy now or that will be just another time to regret???? Any one else can confirm these results?

Bachus and JNav89GT got similar results, but nothing is guaranteed of course.

_damien_, you are correct, I coulda sworn I got instability @ ICL=0 but now when I retest 3d and PI work fine.

http://members.shaw.ca/Eldonko1/ce5-13.JPG

My kits a bit different. I'm getting about 280-281 max, but taking 2.9v to do it. My kit just doesn't want to fly with low volts like Donks and Bacchus. Oh well still happy :)

Quote:

---

Originally Posted by Eldonko

_damien_, you are correct, I coulda sworn I got instability @ ICL=0 but now when I retest 3d and PI work fine.

---

Curious. I've seen overclockers point out that connection before, but I am confident that the theory doesn't hold water. Not that I'm accusing anybody of making it up, I'm just wondering what could be the cause. Of course, just because a low ICL is not *supposed* to cause instability, doesn't rule out the possibility of BIOS-bugs, etc.


Great result, no doubt about that! Is the last screenshot @ 2.85 vDIMM?

A suggestion: Loosening the tREF would free up the command bus, and in a sense give higher priority to the actual CPU-requests. One of the 7.8us-settings should suffice to maintain data integrity. A low value will hurt memory efficiency and possibly reduce stability (but of course, if set too high the data could expire). When a row is refreshed, the capacitor charges of thousands of memory cells are sensed and amplified, then restored to the row again. The amplifying implies rather massive power-surges, which can cause unwanted electrical side effects like noise and voltage droops, so the memory shouldn't be refreshed more often than necessary. When I changed from 133MHz 3.9us to 133 7.8us I gained around 200 MB/s in the Sandra BW-test, even though I'm on single channel.

Quote:

---

Originally Posted by JNav89GT

My kits a bit different. I'm getting about 280-281 max, but taking 2.9v to do it. My kit just doesn't want to fly with low volts like Donks and Bacchus. Oh well still happy :)

---

280-281 is very good, I would be happy too. :)

Quote:

---

Originally Posted by _damien_

Curious. I've seen overclockers point out that connection before, but I am confident that the theory doesn't hold water. Not that I'm accusing anybody of making it up, I'm just wondering what could be the cause. Of course, just because a low ICL is not *supposed* to cause instability, doesn't rule out the possibility of BIOS-bugs, etc.

Great result, no doubt about that! Is the last screenshot @ 2.85 vDIMM?

A suggestion: Loosening the tREF would free up the command bus, and in a sense give higher priority to the actual CPU-requests. One of the 7.8us-settings should suffice to maintain data integrity. A low value will hurt memory efficiency and possibly reduce stability (but of course, if set too high the data could expire). When a row is refreshed, the capacitor charges of thousands of memory cells are sensed and amplified, then restored to the row again. The amplifying implies rather massive power-surges, which can cause unwanted electrical side effects like noise and voltage droops, so the memory shouldn't be refreshed more often than necessary. When I changed from 133MHz 3.9us to 133 7.8us I gained around 200 MB/s in the Sandra BW-test, even though I'm on single channel.

---

Last SS is @ 2.7 + .03v in BIOS, about 2.82-2.83v w/ DMM.

I usually test tref by running bandwidth benches. I find in general 166 -1.95us and 100 - 1.95us give some of the better bandwidths for most RAM but I can test out 133 7.8us this evening for comparo. :)

Question.. If I understand your ICL post correctly, you saying that all mem should be able to run 0clks without instability? I found from tests in the past that higher voltage RAM like utt or bh-5 could run 0 clks fine but mem with overall looser timings like TCCD and UCCC had issues w/ 0. Interesting to hear how ICL actually works, all I knew is PI likes lower heh. ;)

Quote:

---

Originally Posted by Eldonko

Question.. If I understand your ICL post correctly, you saying that all mem should be able to run 0clks without instability?

---

That's what I'm saying :) Delaying the precharging by a fixed number of cycles doesn't really change much, electrically. The precharges would occur just as frequently, the only difference between e.g. 0 and 64 is that the latter will allow a recently accessed row to remain active for up to 64 cycles (after having completed a transfer), before it is forcibly precharged. So if the CPU later requests data from the same row, there's no need to sense/amplify it again (tRCD). In that case the data will be output after the completion of tCL - the only delay involved.

Quote:

---

Originally Posted by Eldonko

Interesting to hear how ICL actually works, all I knew is PI likes lower heh. ;)

---

If an application prefers low ICL that's an indication of many random memory accesses, as opposed to sequential, well organized reads. Typical examples of applications with random patterns are games, or "unpredictable" tasks as a whole. SuperPi probably has a pretty straightforward access pattern, so I guess the lack of modern optimizations could be the explanation to your finding.

First post updated, scaled to 287Mhz 3d and 32m stable w/ 2.8v (BIOS) 2.88v (DMM). :)

Quote:

---

Originally Posted by Eldonko

First post updated, scaled to 287Mhz 3d and 32m stable w/ 2.8v (BIOS) 2.88v (DMM). :)

---

Awesome! :up: Can it survive looping of 3DMark2001 for any substantial amount of time? I can also bench mine in 3D @ same 287MHz but it's no good for some prologed gaming or looping - it will freeze very often.

Not sure yet, just finished testing to 287 a few mins ago, gunna bump .03v and see if I can get to 290. :p: Will loop 3dmark01 as soon as I can.

283 had no gaming issues at all, I let my brother hang at my house and play BF2 all day when I was at work. I dont have the time to play many games so my lil bro takes care of those tests for me heh heh.

Did some 2.5-3-2 testing tonite, cant touch your set w/ CAS2.5 bachus, best I can do 3d benches with is 260Mhz and it needs alotta volts. No way is cas 2.5 worth losing 25Mhz.

Updated 1st post. :)

ELdonko,

Some very nice mem there, i might have to see if I can get my 3500LLPro to perform at 287 like yours :)

SAD

Hey SuAsDu! Thanks and GL :)

Quote:

---

Originally Posted by Eldonko

Did some 2.5-3-2 testing tonite, cant touch your set w/ CAS2.5 bachus, best I can do 3d benches with is 260Mhz and it needs alotta volts. No way is cas 2.5 worth losing 25Mhz.

---

Indeed not. The real difference is when tightening from 2.5 to 2.0, but the performance difference between 2.5 and 3.0 is tiny.

Quote:

---

Originally Posted by Eldonko

Did some 2.5-3-2 testing tonite, cant touch your set w/ CAS2.5 bachus, best I can do 3d benches with is 260Mhz and it needs alotta volts. No way is cas 2.5 worth losing 25Mhz. Updated 1st post. :)

---

My set could do 260Mhz 5-2-3-2.5-1T at same 2.58v (DMM) it maxed out at CAS3.0 in 3D but 270Mhz required hell of a bump in voltage.
If it reaches high at CAS3.0, then CAS2.5 is "just for show". Gains, even at same speed are very small, even from bencher's point of view.

This price drop is worth a thread bump, get it while you can. Link