Sandiego Core = loss of bandwidth with tccd?!?!?!

[brandinb]

well i got my new 3700+ sandiego and am having fun testing it. then i decide to see if the mem controller can take my gskill 4400le higher then my 3000+ winchester which maxed at 305mhz when i notice at the same settings im about 400mb/s slower in memtest 86!!! i double checked and flashed bioses around and reset the cmos still same result. i get about 3200mb/s with my winchester and 2800 mb/s with the sandiego everything being the same besides the sandy gots more cach. i did some sandra tests in windows and noticed the same thing. was getting 7700 in sandra now getting 7400 mb/s. anyone expereiencing this. i get the best bandwidth with my sandy with the 414-3 bios with the other bioses i was getting 2700 in memtest!!

if anyone is experencing this please tell and if anyone knows of a fix for me or some bios to help then thanks
________
STARCRAFT 2 REPLAYS HOCHLADEN

[STEvil]

stuck at 2T probably.

You can adjust this with A64 Tweaker (may need to load cpu-z first).

[brandinb]

ok i just tested this and i am running 1t timings. i clicked 2t to enable and it got even slower amazingly. this new cpu is dissapointing because with this low bandwidth it isnt too much faster then my 3000 accept in 2001se where it smokes it hehe. my sandra is lower then i thought i score 6700 with 300x8 cl 2.5-3-3-6 all tight timings and 1t. i remember a post of another who claimed to loose 300mb/s too ill try to find it. i dont think im doing anything wrong i think it might be a problem with my dfi board maybe and this cpu
________
Rehab Forum

[TSFroggy]

You want to disable 2T timing in order to get it to run faster; a 1T command rate is far superior to 2T.

[Geforce4ti4200]

maybe its just a bug? if 3dmark and games are faster this is what matters

[Dumo]

Clean install windows?

[brandinb]

no its not a clean install of windows but i know that doesnt matter in memtest in dos

games and apps are faster but not by the amount i expected and all synthetic memory benchmarks show much slower memory performance compared to my winchester.

so yeah i double checked everything and this cpu just gives lower bandwidth at the same settings as the winchester. maybe a bios can be made to fix this.

can anyone prove what i have found wrong?? i have not been able to find any memory benchmarks comparing the sandiego core to a winchester or fx 55 at the same settings. any chance someone can do this to verify or prove wrong my findings
________
VAPORIZER AFFILIATE PROGRAM

[JonEapples]

Where it says "2T Timing" in A64 tweaker, change it to "Disable". You will notice a big difference. I think I read somewhere about lower bandwidth though even at 1T, probably due to changes in the memory controller. It may have a little less bandwidth, but greater reliability.

[brandinb]

it is dissabled i just tested enable to verify i was already running it dissabled before

dont worry i know what im doing it is running 1t and yes it is still slower then the winny

i think its a dfi bios bug / incompatability or maybe the diego is just slower with memory

can you tell me what you get for bandwidth in memtest joneapples as you have tccd also
________
LovelyWendie99

[JonEapples]

I pulled out my OC notebook to look at tests I did when I had my 3500+ Winchester in with the same hardware in my sig. Looking at both the chips running at the same settings:

Winchester:230x11, Patriot RAM 1.1@ 2-3-3-6 1T
Sandra: 6799/6725 Efficiency:92/91%

San Diego:230x11, Patriot RAM 1.1@ 2-3-3-6 1T
Sandra: 6487/6420 Efficiency:88/87

But I don't think it has anything to do with TCCD, because I also have test results from when I had my Corsair XMS3200C2 v5.1 (crap Promos chips)

Winchester: 230x11, Corsair RAM 1.1@ 2.5-3-3-6 1T
Sandra:6762/6689 Efficiency:?

Stock speeds:

Winchester:200x11, Patriot RAM 1.1@ 2-2-2-5 1T
Sandra: 5950/5886 Efficiency:93/92%

San Diego:200x11, Patriot RAM 1.1@ 2-2-2-5 1T
Sandra: 5698/5627 Efficiency:89/88%

Winchester: 200x11, Corsair RAM 1.1@ 2.5-3-3-6 1T
Sandra:5892/5827 Efficiency:?

I guess you do lose bandwidth, maybe with a bios update this will change, my Neo2 still has no official support.

[The Mofo]

Well tha totally blows. 1 meg cache to boot. Hopefully it can be resolved with a BIOS update. Core updates arent supposed to make the CPU slower then the previous stepping.

[Geforce4ti4200]

looks like a compatability issue of cpu and mobo

[misteroadster]

no , the rev E A64 has been modified for better compatibility with memory.
I think venice has the same lack of bandwith when compared to winnies.
sad thing the sandiego chips are'nt issued from winnies architecture.

[mrlobber]

Core updates arent supposed to make the CPU slower then the previous stepping.

I don't think lower Sandra bandwidth necessarily indicates a slower CPU

[eva2000]

I don't think a lower Sandra bandwidth necessarily indicates a slower CPU

Yup...

[MaxxxRacer]

well its been tested and clock for clock the venice and san diego are faster than the counterparts they replaced so i think its safe to say this really Doesnt matter.

[[XC] leviathan18]

i think a bios update should fix this... but if you get better perfomance in real world that is what you want

[Rubescen]

my Venice shows 2000 point lower on Everest memory read test than the Winchester I had, didn't change much on Sandra, though I get almost 500 point more on PCMark04!! I would rather trust PCMark04's score though

[computersmsa]

Yes
SD and Venice = loose of 300-400 Mb/s with all memory

[wickedld9]

I'm with Rubescen, sandra and everest should be for tuning your memory, not to indicate a superior CPU because it's now obvious this isn't the case.
Rev. E has more efficient use of bandwidth.

[J-Mag]

has this been shown in Everest and Sandra as well?

[xxmartin]

I have already posted this in another topic.

My Venice shows the same behaviour. I loose around 300 MB/s in SiSoft Sandra compared to equally clocked Winchester and NewCastle (same clock, same cpu-multi, same ram-divider, same timings, 1T, same HTT, ...).

1:1 comparison with SuperPI (vanilla version without any tweaks) gives the same result. The 1M/2M/4M/8M are actually with equal performance, but 16M and 32M differ some seconds. Since the 16M and 32M depend on memory bandwith the missing seconds should be the same as -300MB/s in SiSoft.

Anyway, 300 MB/s are nothing to worry about when handling with 7-8 GB/s.

[brandinb]

anyone find any fix for this yet or is this something hardcoded in the cpu
anyone like opb know if oskar is aware of this thing maybe he can fix it in bios code??
________
The cigar boss

[xxmartin]

anyone find any fix for this yet or is this something hardcoded in the cpu
anyone like opb know if oskar is aware of this thing maybe he can fix it in bios code??

I do not really think that this can be fixed that easy.

Of course, it might be just a BIOS issue but I think the slight performance drop could be caused by the new E-stepping memory controller. The new iMC is much more compatible and allows more memory banks to be accessed at DDR400 speed compared to Winchester & NewCastle. Maybe they have "lowered some timings" for improved memory compatibility but less memory throughput.

[uwackme]

Umm, guys.... there are ALOT of memory timings in the A64 controller, all with "defaults" and it could be some "defaults" hve changed from the factory with the new chips..."to increase compatability". So with A64 tweaker and maybe additional parameters we dont have access to yet (A64 tweaker needs expansion to ALL parameters?) you can TWEAK the Venice/SanDiego back to Winnie speed.

EX: if parameter Tfk-u was 3 on Winnie and is now 4, setting it back to 3 will get the bandwidth back. lots of aprameters are left as default by the bios's... default = CPU powerup default and if noone changes the new chips "slower" default, it will yeild lower numbers compared to "faster" default of Winnie.

So... and always remember this... DONT PANIC