I need to go look at the Nvidia gui again. I need to understand exactly what an OC of the graphics clock of say 50 means in terms of the binning you described.
Prior experience seems to indicate you can't push the graphics clock much past 170 without it locking up.
What I can't determine is the highest memory transfers rate to set OC to without losing the ability to transfer data synchronously (nearer synchronicity. I think I am hearing it is asynchronous).
Let me scratch my head and examine that gui some more.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Even on my gpus which are almost exclusively water-cooled, which means their temperature deltas are only a few degrees, ever, I only boost the graphics clock a max of 60-80Mhz on any gpu.
Trying to go to +100Mhz just bounces the graphics clock back and forth between two bins with the higher bin rarely winning out.
For the memory clock I normally only get back to video processing P0 clock state and adding 400-500 Mhz on top of that to keep from causing too many error correction events. I want every data transfer to succeed on the first try which is the most efficient.
For the memory clock I normally only get back to video processing P0 clock state and adding 400-500 Mhz on top of that to keep from causing too many error correction events. I want every data transfer to succeed on the first try which is the most efficient.
Took a look at the "top clock" with "0" overclocked memory transfer. Added +500 back in to take it up to what appears to be the default clock. So its +60/+500 with "auto". In spite of the apparent very fast processing speeds there was basically no real movement on the RAC #'s.
So I said the heck with it.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
For the memory clock I normally only get back to video processing P0 clock state and adding 400-500 Mhz on top of that to keep from causing too many error correction events. I want every data transfer to succeed on the first try which is the most efficient.
Took a look at the "top clock" with "0" overclocked memory transfer. Added +500 back in to take it up to what appears to be the default clock. So its +60/+500 with "auto". In spite of the apparent very fast processing speeds there was basically no real movement on the RAC #'s.
So I said the heck with it.
Remember... you need to wait upwards of a week or so to see any movement in your RAC numbers.
What is says is be careful about what ram you buy as there IS a difference
Yes, I can see that being a case to study... for some, if not many.
If I were to add more RAM, or get a new motherboard, I would be leaning towards G.Skill because they seem to be faster and better than most other brands. Plus, I also DO look at the motherboard's QVL for a listing from G.Skill, and then I DO search for the Samsung B dies which are double sided.
But, that's just me... Others may have their preferences, but I thought I'd share mine since you brought it up.
That post by the user just demonstrates his lack of knowledge and understanding of how memory access works.
The salient point he attempted to make was simply the concept of interleaving memory access.
Two channels versus four channels. Single rank interleaving versus dual rank interleaving.
It all has to do with being efficient with memory access so that you can achieve two different actions on the same clock cycle, reading from a memory line and refreshing a memory line.
DDR5 is different from previous RAM generations as each module has two sub-channels, in effect achieving dual rank with only 8 chips on a single stick or DRAM.
If you want to know how DDR5 memory works, I can recomment Patrick's YT vidoe at ServetheHome.
It mostly focuses on ECC but it explains well the construction of the two sub-channels of DDR5.
With the DDR5 generation, Samsung DDR5 B-die chips are the slowest of the competition. Better memory is built with Micron A-Dies or SKHynix A or M-dies.
Thank you, Keith. I need
)
Thank you, Keith.
I need to go look at the Nvidia gui again. I need to understand exactly what an OC of the graphics clock of say 50 means in terms of the binning you described.
Prior experience seems to indicate you can't push the graphics clock much past 170 without it locking up.
What I can't determine is the highest memory transfers rate to set OC to without losing the ability to transfer data synchronously (nearer synchronicity. I think I am hearing it is asynchronous).
Let me scratch my head and examine that gui some more.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Even on my gpus which are
)
Even on my gpus which are almost exclusively water-cooled, which means their temperature deltas are only a few degrees, ever, I only boost the graphics clock a max of 60-80Mhz on any gpu.
Trying to go to +100Mhz just bounces the graphics clock back and forth between two bins with the higher bin rarely winning out.
For the memory clock I normally only get back to video processing P0 clock state and adding 400-500 Mhz on top of that to keep from causing too many error correction events. I want every data transfer to succeed on the first try which is the most efficient.
Keith Myers wrote: For the
)
Took a look at the "top clock" with "0" overclocked memory transfer. Added +500 back in to take it up to what appears to be the default clock. So its +60/+500 with "auto". In spite of the apparent very fast processing speeds there was basically no real movement on the RAC #'s.
So I said the heck with it.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Tom M wrote: Keith Myers
)
Remember... you need to wait upwards of a week or so to see any movement in your RAC numbers.
Proud member of the Old Farts Association
GWGeorge007
)
+1000
I couldn't find a RAM forum
)
I couldn't find a RAM forum so am posting this here because you guys are talking about tweaking things to the nth degree:
http://www.primegrid.com/forum_thread.php?id=10161
What is says is be careful about what ram you buy as there IS a difference
mikey wrote: I couldn't find
)
Yes, I can see that being a case to study... for some, if not many.
If I were to add more RAM, or get a new motherboard, I would be leaning towards G.Skill because they seem to be faster and better than most other brands. Plus, I also DO look at the motherboard's QVL for a listing from G.Skill, and then I DO search for the Samsung B dies which are double sided.
But, that's just me... Others may have their preferences, but I thought I'd share mine since you brought it up.
Proud member of the Old Farts Association
That post by the user just
)
That post by the user just demonstrates his lack of knowledge and understanding of how memory access works.
The salient point he attempted to make was simply the concept of interleaving memory access.
Two channels versus four channels. Single rank interleaving versus dual rank interleaving.
It all has to do with being efficient with memory access so that you can achieve two different actions on the same clock cycle, reading from a memory line and refreshing a memory line.
DDR5 is different from previous RAM generations as each module has two sub-channels, in effect achieving dual rank with only 8 chips on a single stick or DRAM.
If you want to know how DDR5 memory works, I can recomment Patrick's YT vidoe at ServetheHome.
Server DDR5 RDIMM and ECC UDIMM the Video
It mostly focuses on ECC but it explains well the construction of the two sub-channels of DDR5.
With the DDR5 generation, Samsung DDR5 B-die chips are the slowest of the competition. Better memory is built with Micron A-Dies or SKHynix A or M-dies.
mikey wrote: I couldn't find
)
There are lots of forums where memory overclocking is the main focus. Suggest Overclock.net to start with.
Use a custom search filter and limit the search to the last six months.
Keith Myers wrote:mikey
)
I created a new forum topic and posted it in there, all new posts on this topic should be posted in here:
https://einsteinathome.org/content/why-buying-new-ram-isnt-easy