DM6467T DDR2

Hi,

Thanks for your fast answer, actually we haven't still tried, since we didn't know if it was possible. If we change the ram for a slower one, do we have to change any configuration?

If so, what should we change?

Regards,

Xavi

Thanks,

I'll try it, but I'm concerned with the timings, the 7-7-7 configuration will be compatible with the 5-5-5 configuration (CL-tRCD-tRP)?

Will it also work with the 667MHZ (5-5-5) and 533Mhz (4-4-4)?

Regards,

Xavi

Thanks again for your response.

When you say  max latency 5, so latency of 5 or less is fine, you mean 7-7-7 is fine? Because the dm6467t has a latency of 7.

I guess i have some concepts wrong.

Thanks,

Xavi

Thanks a lot,

We'll try micron 800MHz, I'll post if it works.

Regards,

Xavi

Hi,

I met a same problem, we use DM6467T with Hynix HY5PS1G1631CFR s6, which is a 6-6-6 chip. Does DM6467T support this chip?

As I mentioned before, DM6467 or DM6467T supports CAS Latency of 2, 3, 4 or 5. CAS latencies 6 and 7 are not supported.

Regards, Srirami.

Hi, thanks for everyng first of all.

At las we mounted the MT47H64M16HR-25E:H and works fine for us, as srirami said 6=6=6 is not supported, but note that yours also supports 4 and 5 timmings, as ours, so it should be fine.

Regards,

Xavi

I've finished my circuit diagram, and now I'm searching for DDR2 before starting the PCB design.  I want to make sure I can get the darned things.  

= Elpida EDE1116A* These are all End-Of-Life

Please comment on what you believe to be the suitability of the parts below.  Thanks very much.

= Micron MT47H64M16HR-25E:H is in production with stock available http://www.alldatasheet.com/datasheet-pdf/pdf/349745/MICRON/MT47H64M16HR-25EAT.html

= Winbond W971GG6JB-25 is in production with stock available http://www.winbond.com.tw/NR/rdonlyres/3B871307-A5FB-48A6-9A4F-1BC6B03464AB/0/W971GG6JB.pdf

= Hynix H5PS1G63EFR (S5 or Y5)  http://www.hynix.com/datasheet/pdf/dram/H5PS1G63EFR(Rev1.2).pdf

= Hynix HY5PS1G1631CFP http://www.hynix.com/datasheet/pdf/dram/HY5PS1G4(8.16)31C(L)FP(Rev0.6).pdf

From what I can tell, they all

- same 84 ball FBGA configuration

- support DDR2-800 with 5-5-5

Thanks,

Helmut.

P.S. I see reference earlier in this thread that MT47H64M16HR-25E:H should work.  Did it pan out OK since last month?

By now he still havn't got any issues with the RAM from micron. As far I know if they support 5-5-5 they should be fine. Again, for us the Micron Ram works fine for us.

Best Regards and good luck,

Xavi

Xavi,

Thanks for the info.

On a related note, for you and all community members, how do you fully confirm proper operation of your new board with the FBGA part on it?

CONTEXT: I've been doing microprocessors for 30 years, but this is my first SoC and first use of BGA's.  I'm accustomed to single-stepping my processor through exercise code that twiddles all the lines between processor and memory, just to make sure I have a good prototype (often hand soldered) board prior to getting started on firmware development.  I do the same kind of thing for each subsystem on the board.  With the BGA type parts, I won't be able to probe the pins with a scope unless I make special arrangements on the board.  Of course, I also should be doing very little firmware development when it comes to low level hardware drivers.  (Note that I'm still torn between Linux, BIOS, and no OS at all for my special purpose video analytics app.)  What do YOU do?  Just turn it on, load it up, and hope for the best?  Just the thought of that sends chills up and down my spine!  I do realize that one could write a memory test routine to access all the data and address lines, checking for opens and shorts of various kinds.  It's just been faster for me, in the beginning, to just check those with a scope.  Eventually, I do write a memory check routine.  But this is all in the uP domain, not SoC.  While I've been doing PC programming, including assy language, for 30 years as well, I've never before stepped so squarely in the void between uP and PC, which is where this SoC lies. (I have built a PC of sorts from scratch, even, in 1980.  It was based on a Motorola 6809 using OS9.  Still not an SoC, however.  It executed directly from EEPROM, with no "booting".  And then I've done "booting" with the PIC18F8680 and some other micros, but they're all still uP's.  So I've danced all around this void, but never stepped into it before.)

Thanks,

Helmut

(I don't know what's up with the date sorting on this two-page thread.  This response is to Xavi's 11/11/2010 response about using Micron DDR2, even though this post is getting sorted after some August posts on the second page.)

Hi Helmut,

To our experience, what we did was aking the schewmatics from the demo board and modify them following as much as possible  the philosophy, just to minimize the efforts on porting everything to this board. Even though, we've found many issues, and we've finished going to a third party company, since deppending on what you want to do and the timmings you expect, it will be hard. Answering your question, what we did waas exactly this, design everything, implement it and build it, load everything and from there adapt every modification, but be sure to have something exactly as the demo so you can load everything and check that at least this part works, if not, bad news. I don't know if everybody does this way, maybe that's temerary, but that's what we did.

I'm not sure what you mean with "Note that I'm still torn between Linux, BIOS, and no OS at all for my special purpose video analytics app", but everything works with linux, and if you're not used to work with it, and used means knowing it at very low level, expect long time on it.

Feel fre to e-mail me to tisaire@wingsformedia.com.

Regards and good luck,

Xavi

Hello!

I just found this forum here and would like to inform you that Memphis has released a 2GBit DDR2 components last year which is available in 800MHz CL5 as well as 1066MHz CL7. We even provide industrial temperature (-40 to 95°C).

Part number MEM2G08D2DABG-25 for x8 configuration,  MEM2G16D2DABG-25 for x16 configuration, both DDR2-800.

For further information refer to the datasheet: http://www.memphis.ag/fileadmin/datasheets/2Gbit_1012.pdf

One question I have: The DaVinci allows a max CAS latency setting of 5-5-5. We have a customer who wants to run it 1066MHz, but it is very hard/impossible to reach 1066MHz @ CL5, since that would mean 5 x 1.875ns = tRCD of only 9.375ns. No DDR2 in the world can do that.

Is there really no way to set the DaVinci to CL7 or at least CL6 ? The CAS-latency is not really influencing the overall speed much, a high clock-rate is more important.

At DDR2-1066 ( 1 cycle is 1.875ns):

Reading 50000 data-words in burst-mode at CL7 means it will take 50007 cycles of 1.875ns  = total 93763.125ns

At DDR2-800 (1 cycle is 2.5ns):

Reading 50000 data-words in burst-mode at CL5 means it will take 50005 cycles of 2.5ns  = total 125012.5ns

As you can see, the overall speed is much better with 1066MHz CL7 parts than with 800MHz CL5 parts.

But to go for 1066MHz, we need to find a way to set the Davinci to CL7!

Regards,

Thorsten

Thorsten,

Thanks for letting us know about Memphis DDR2 parts that are compatible with DM646x (DDR2-800 with CL5). DM646x doesn't support 1066MHz DDR2. Which device are you referring to regarding DDR2-1066 with CL5?

Regards, Srirami.

Dear Srirami,

it is the TMS320DM6467T

The evaluation board running 1066MHz is the HD1080P DVM that you can find here: http://support.spectrumdigital.com/boards/evmdm6467t/revc/

This evaluation board uses a 1066MHz Elpida DDR2 IC  EDE1116ACBG-1J-E which is also just able to do CL7.

We could program the CL7 into the DaVinci, but officially this is not supported. Officially supported is max CL5.

The problem is that 1066MHz DDR2 memory components are only available in CL7. The evaluation board HD1080P DVM could theoretically not be produced/would not work, except if TI would allow the CL7 to be set in the CPU.

If TI made such evaluation board running DDR2-1066MHz ICs with Elpida CL7 ICs, then this would mean that you can allow CL7, is it right?

Regards,

Thorsten

PS: Would TI be interested in some free samples of the Memphis 2GBit ICs for testing-purposes? If yes, please contact me. You can have 800MHz CL5 as well as 1066MHz CL7.

FYI: Programming CL7 into the Davinci lets the board crash. We then tried setting CL5 and using 1066MHz CL6 ICs (very rare parts) and it worked, but this means we drive it "out of spec", so it is critical if we can not set the Davinci to CL6 or CL7.

Will there eventually be a new version of the Davinci soon that solves this issue?

Thorsten,

Sorry for the confusion.

Intially we planned to support 1066MHz speed on DM6467T. But, we endeded up with 800MHz. That was the reason for using 1066-MHz DDR2 part on the DM6467T EVM. The EVM is working fine with "“EDE1116ACBG-1J-E” part when configured for :800MHz & CL=5.

The in-consistency of DDR2 part (used on DM6467T EVM) with the DM6467T data sheet was observed by Spectrum Digital Inc . They will use “EDE1116ACBG-8E-E” in the next version of EVM, which meets DM6467T spec (800MHz with CL=5). They already updated DM6467T spec with this new part.

We don't have any plan for new version of Davinci for 1066MHz DDR2 speed.

Right now, we are not doing any testing with new DDR2 parts. We will definitely consider Memphis parts in future.

Regards, Srirami.

Hi,

I'm planning to mount the EDE1116AJBG-1J-F (CL7) in spite of the fact that U-boot and other software set the DDR2 interface registers to work with a 800MHz part. (CL5)

Based on the above comments, I understand that it will work just as fine as the original EDE1116ACBG-1J-E part since all changes refer to die revision and ROHS compliance. Is that correct?