eva2000

I got a 3200+ 0518 APDW and 3000+ 0516 GPAW Venice cpus just yesterday.

AMD64 3200+ 512k cache (Venice)
ADA32000DAA4BP
LBBLE 0518APDW

AMD64 3000+ 512k cache (Venice)
ADA30000DAA4BP
LBBLE 0516GPAW

Only started testing the 3200+ 0518 APDW and the memory controller ain't anywhere near as good as my 4000+ san diego (i.e. 2x512MB Gskill PC4400LE on San Diego can manage at least 300mhz 2.5-4-3-7 at 2.8v for memtest and benching, but with 3200+ Venice errors out immediately in memtest86+ v1.55.1 but still have to retest the ram properly over next few days.)

* The venice's cpu stability is even more closely tied in to memory controller/mem stability than previous winnies/clawhammer/sandy cpus I've had.

* Using my Gskill PC4400LE i had to throw out all known stable timing/settings on my other cpus and start testing from scratch with my Venice. Tref, Drive and Drive data strength can get very unstable at some settings so have to retest them all as well.

* Right now superpi 32M is fine at 2750mhz at 1.58v and stock vcore fine for up to 2550-2600mhz for pifast and 2663mhz at 1.53v.

* But memory timings play such an important part to stablize this Venice as i loosen timings further i get more stability at higher end! It's going to take alot more time to find stable mark as i have to retest from scratch Gskill PC4400LE settings then test cpu for stability.

* Different bioses gave me different stable points with cpu and ram - tried 510-2, 510-3, 414-3, 414-2 and now 310 official which seems best for cpu stability.

System
3200+ 0518 APDW water cooled
DFI NF4 SLI-D 310 official
2x eVGA 6800U PCI-E SLI 76.10
2x 512MB Gskill PC4400LE orange slots
120GB WD 1200JB
650W Silverstone SLI PSU

WinXP Pro SP2
NF4 6.53 remix chipset drivers

- note SLI mode will probably reduce cpu related performance slightly so take that into a/c with the benchmark results.

2x 512MB Gskill PC4400LE dual channel kit

@10x 266mhz 2.5-3-3-7 7-14-2-2-1-2 at 2.8v

chipset = 1.5v
LDT = 1.5v
LDT 3x

CPC = Enabled
CAS Latency(CL) = 2.5
RAS to CAS(Trcd) = 3
Min RAS# Active time(Tras) = 7
Row Precharge Time(Trp) = 3
Row Cycle Time(Trc) = 7
Row Refresh Cycle Time(Trfc) = 14
Row to Row Delay(Trrd) = 2
Write Recovery Time(Twr) = 2
Write to Read Delay(Twtr) = 1
Read to Write Delay(Trwt) = 2
Refresh Period (Tref) = 3072
Write CAS Latency(Twcl) = Auto
DRAM Bank Interleave = Enabled
Skew = AUTO
Skew Value = 0
DRAM Drive Strength = Level 3 (prior cpus 7 or 8 worked, now 1, 4, 5, 6, 7, 8 all either reboot when running pcmark04 memory test or BSOD)
DRAM Data Drive Strength = Level 1 or 2 (which allowed me to prime longer)
Max Async Latency = 7ns
Read Preamble Time = 5ns
Idle Cycle Limit = 256
Dynamic Counter = Enabled
R/W Queue Bypass = 16x
Bypass Max = 7x
32 byte Granularity = Disable(8burst)

Haven't found max yet, just trying to find a stable reference point for cpu and the ram.

Click image for larger pic



Not the best venice, but not the worst either

eva2000

reserved

ArcTan

hmm I thought the main reason there was a venice(besides SSE3) was to improve their memory controller?

btw
do you still think it's worth getting a venice?
iirc these seem to clock similar to your old 0441 or 0445 winnie

eva2000

Well improved means venice handles 4x 512MB at DDR400 2T instead of DDR333 2T, that's all i think.

Clock for clock venice edges out winchester and memory bandwidth is high on venice i.e. 10x200mhz 2-2-2-5 7-14-2-2-1-2 memtest86+ v1.55.1 reports something like 2680MB/s, compared to winnies at 2380-2466MB/s

But maybe my initial stance on the memory controller is because I'm forced to retest my Gskill ram from scratch. But I'll reserve final judgement till i fully test both 3200+ and 3000+ venices I have.

Problem is oc'ing range varies alot with venice, seen a few 3200+ not get past 2500-2600mhz stable while others clock to 2750-2850mhz.

But venices do run cool compared to san diego. At idle 10x200, venice according to smartguardian/MBM5 is at 25-27C and load 36-39C. Even at 10x266HTT at 1.57v load is only 38-41C. That's at least 10-13C cooler than san diego.

My 3200+ 0445 VPMW winchester did 2745mhz at 1.6v on Asus A8N-SLI Deluxe, but only 2700-2717mhz at 1.63-1.66v on DFI NF4 SLI-D. While this 3200+ venice can 3dmark at 2750mhz at 1.58-1.67v.

ArcTan

hehe probably could do what my winnie is doing
I'll probably wait for some good steppings to pop up before I buy
don't want to take a punt

groovetek

same ere... but if there aren't still any real decent results by the time i finish exams... then i know i'll be itching to get a chip and end up taking a punt... don't see why not though... coud be lucky and get a 2.8-2.9g on air chip.

eva2000

seems so far my findings are

1. both san diego and venice like BH-5 ram more than TCCD ram
2. san diego has much better mem controller than venice, both BH-5 and TCCD clock way higher on san diego than venice
3. san diego mem controller is better than clawhammer, newcastle, winchester and venice for BH-5/UTT BH-5 ram

This is with 2x512MB OCZ PC3500 Plat LE BH-5 on 166 divider and 3.3v jumper on DFI NF4 SLI-D







Got same 3d2k3 score with 4000+ san diego at stock 12x200HTT = 2400mhz heh

babyelf

woot... i think i'll be ordering the 3700+ then =/