cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

Meteorhead
Challenger

S10000P?

I have seen on several forums that AMD has announced the S10000 dual-Tahiti FirePro card aimed at workstations. I wanted to ask whether there are plans for passive cooled (S10000P) variant of this card. Also, there are no indications on any of the forums, that this card would (or would not) feature ECC RAM. Passive cooled variants (with a lot less display outputs, max 1 DP or HDMI) would suit rack mounted workstations a lot better.

1 Solution
hsaigol
Adept III

yes it has ecc its a workstation tahiti (dual actually)

View solution in original post

0 Likes
10 Replies
binying
Challenger

As far as I know, AMD is unable to answer future roadmap related questions on this forum...

0 Likes
nou
Exemplar

passive cooled beast with over 300W TDP?

0 Likes

Wheres the Problem?

In a server it is not really passiv. You have very strong 80mm fans, than blow more air through your cooler than most most coolers on a custom design GPU.

It is more like 80-100% speedfan of a radial fan on a ref design GPU

0 Likes
hsaigol
Adept III

yes it has ecc its a workstation tahiti (dual actually)

0 Likes

Isnt it so that all of AMDs GPUs have EDC ( can correct 1 or 2 bit errors) which is probably better than ECC (can correct 1 bit error, unless SECDED) ?

http://www.anandtech.com/show/2841/12

Sorry, you don't understand what EDC and ECC means

EDC: ErrorDetectionCode

ECC: ErrorCorrectionCode

EDC can only say, that there is a error, and if, the data is read again (Errors during Ram<->GPU-Chip transfer)

ECC can find a error AND! correct it with the ECC bits. So you don't have to read it again. Also in the RAM, the wronge value can be corrected.

The retransmission is automatic in hardware. According to Elpida

http://www.elpida.com/pdfs/E1600E10.pdfhttp://www.elpida.com/pdfs/E1600E10.pdf

I wouldnt be worried about re-transmission. But sure, it probably cant detect if the bits suddenly change after writing to memory.

0 Likes

There is something I don't understand (among many):

2.2 and still no DMA suggests that all AMD cards have EDC RAM, while yurtesen's anandtech article suggests that EDC is a GDDR5 feature rather then the feature of the AMD memory controller.

Beyond gaining the ability to use GDDR5’s power saving abilities, AMD has also been working on implementing features to allow their cards to reach higher memory clock speeds. Chief among these is support for GDDR5’s error detection capabilities.

Which statement is true? Do GTX cards also feature EDC? I can see that S10000 uses ECC, that's cool, but I would also consider using cheaper GPUs, if they prove to make mistakes less often.

Second question: what errors can ECC correct? The anandtech article shows that EDC can correct up to 2 bits of error with 100% accuracy that the bus makes, however it cannot compensate for errors made by the memory itself. How often does that happen? And how is ECC better in this regard?

0 Likes

DRAM ECC may be part of GDDR standard, but that wouldn't cover on-chip protection (which takes a lot of transistors). In addition if I recall correctly it consumes a chunk of DRAM to manage the state, and of course you wouldn't want that enabled on a gaming card when you might want to use that memory for textures. That's why there is market segmentation for this sort of feature.

0 Likes

Yeah, that is right.

The Protection is done with paraty Information and so on.

This is the reason why you have less memory to usw, when you use ECC with the Fire pro. But that is not all!

You also loos VRAM-Bandwidth! because you also need to read/write this additional information. So really use ECC only if you really need it.

0 Likes