Overclocking Questions

[Quarters]

 

Hell everyone, after RMAing an old crappy gpu that wouldn't even be stable at stock, I have tried overclocking my new card. 

Here is everything in Unigine Heaven.

Graphics : 2100 Mhz  (Overclocked)

Memory: 5610 mhz      (Not yet overclocked)

around 65 C

 

Are these numbers good or up to par?

 

Here is the unigine score.

https://imgur.com/a/4Y4Z0uj

[TrigrH]

2100 is unlikely to be stable in gaming

[Quarters]

Yes now that I have further tested this, I've noticed around 2030 - 2050

 

[TrigrH]

40 minutes ago, Quarters said:

Yes now that I have further tested this, I've noticed around 2030 - 2050

if you can stay over 2000 while gaming you have done good

 

As for memory, start with +300, note with memory going higher with lower performance if you are getting memory errors.

[TheDankKoosh]

1 hour ago, Quarters said:

 

Hell everyone, after RMAing an old crappy gpu that wouldn't even be stable at stock, I have tried overclocking my new card. 

Here is everything in Unigine Heaven.

Graphics : 2100 Mhz  (Overclocked)

Memory: 5610 mhz      (Not yet overclocked)

around 65 C

 

Are these numbers good or up to par?

 

Here is the unigine score.

https://imgur.com/a/4Y4Z0uj

 

1 hour ago, Quarters said:

Yes now that I have further tested this, I've noticed around 2030 - 2050

 

You might be able to get 2100 stable if you raise voltage and set up a slightly more aggressive fan curve, though overall it wouldn't be worth it for the extra heat and very little performance gain you'll see. The G5X in your card should do at least +500, though it depends on manufacturer of the memory. My 1070 ti can do a bit over 2100 stable, and can get +550 on the memory for reference.

[X_X]

1 hour ago, TrigrH said:

As for memory, start with +300, note with memory going higher with lower performance if you are getting memory errors.

What does that mean?

[TrigrH]

2 minutes ago, X_X said:

What does that mean?

memory clock offset?

 

You can be stable at +600 but it will be less performance the +400 due to memory error correction.

[X_X]

With Pascal there isn't any error correction, there is error detection which can request data to be sent again but I have never seen this happen but instead just a jump in memory tables. Let me see if I can find an old pic of it.

[TheDankKoosh]

8 minutes ago, TrigrH said:

memory clock offset?

 

You can be stable at +600 but it will be less performance the +400 due to memory error correction.

Possibly, but depending on the manufacturer, that memory will probably perform better at +600 over +400, my memory definitely sees an improvement up to +550 and mine uses micron, probably one of the worse of the 2 that primarily produce vram for nvidia. 

[TrigrH]

3 minutes ago, TheDankKoosh said:

Possibly, but depending on the manufacturer, that memory will probably perform better at +600 over +400, my memory definitely sees an improvement up to +550 and mine uses micron, probably one of the worse of the 2 that primarily produce vram for nvidia. 

all 1080ti's use gddr5x which are all micron, and they overclock differently to gddr5.

[THraShArD]

I have lots of trouble believing you say you last GPU was crappy at Stock speed and now new issues with something new.

 

Please don't overclock,

[TheDankKoosh]

8 minutes ago, TrigrH said:

all 1080ti's use gddr5x which are all micron, and they overclock differently to gddr5.

Yes, it overclocks better compared to regular micron gddr5 since they need a higher threshold for speeds, look at how well gddr6 overclocks.

 

6 minutes ago, THraShArD said:

I have lots of trouble believing you say you last GPU was crappy at Stock speed and now new issues with something new.

 

Please don't overclock,

I have lots of trouble believing that you are some sort of silicon mastermind since you never had the first card in question. I've encountered cards that had defects such as my friend's 970 ftw that would crash at stock unless you upped power limits.  

[X_X]

@TrigrHRunning a simple bandwidth test while programmatically increasing the memory clock and recording results for GTX1050Ti

The dips align with the VBIOS Memory Table entries. Same happens for GTX 1080 but different clocks of course.

 

[TrigrH]

27 minutes ago, X_X said:

@TrigrHRunning a simple bandwidth test while programmatically increasing the memory clock and recording results for GTX1050Ti

The dips align with the VBIOS Memory Table entries. Same happens for GTX 1080 but different clocks of course.

Yes and that large DIP is the point where you are getting too many memory errors and resulting in reduced performance.

[X_X]

It's nothing to do with errors, it's a change in Memory Table and timings.

[THraShArD]

1 hour ago, TheDankKoosh said:

Yes, it overclocks better compared to regular micron gddr5 since they need a higher threshold for speeds, look at how well gddr6 overclocks.

 

I have lots of trouble believing that you are some sort of silicon mastermind since you never had the first card in question. I've encountered cards that had defects such as my friend's 970 ftw that would crash at stock unless you upped power limits.  

 

I've built tons of systems since the Voodoo 2 days and never had to RMA a part.

[TrigrH]

4 minutes ago, THraShArD said:

 

I've built tons of systems since the Voodoo 2 days and never had to RMA a part.

looks like you just jinxed your 2080ti, have fun with those memory errors

[TheDankKoosh]

14 hours ago, THraShArD said:

 

I've built tons of systems since the Voodoo 2 days and never had to RMA a part.

Well that's you, and you seem to only use some of the most expensive parts on the market, which probably have lower failure rates compared to mainstream parts.