Quote from: bensam1231 on November 18, 2016, 01:59:34 PM

Quote from: xeridea on November 18, 2016, 03:41:21 AM

Quote from: bensam1231 on November 18, 2016, 02:31:18 AM

Quote from: nerdralph on November 17, 2016, 08:09:59 PM

Quote from: bensam1231 on November 17, 2016, 04:26:18 PM

Do you also know if you want to check if a algo is memory limited, you can go into GPUZ and check out the MCU (memory controller unit) and see the load on it?

I think this is wrong. Although I primarily mine using Linux, I have a Windoze box that I use for testing cards. GPU-z appears to show only external bus bandwidth use (to the GDDR), and not the utilization of the bandwidth between the controller and core. In practical terms, a miner kernel may be using 200GB/s of memory bandwidth, but a significant percentage of it can be from the L2 cache. The collision counter tables in SA5 would be an example of this.

Do you have a source for this hypothesis? In all memory restricted algos that correlates to MCU usage. Pretty sure it pertains to any sort of memory overload, bandwidth or bus width...

Quote from: xeridea on November 17, 2016, 09:56:45 PM

480 and 1070 have similar TDP. Mining Zcash, their power usage would be similar. 1070 maybe slightly less if you could downclock it, but you can also undervolt the 480. Even if the 1070 is slightly more efficient with optimized Zcash, it doesn't matter much. I make 9x more on ZCash than I spend in power. So it isn't worth spending $400 on card that has same speed as $200 card.

38% wasn't from me. I was using similar method of extrapolation. I get 160S on 480, no overclocks. ~60% MCU on Claymore 6.0.

Their power usage would be similar if they were both being maxed out. Equihash is not a highly optimized algo yet, especially for Nvidia. That's the whole reason we're talking about this. You're trying to make a point of Nvidia not being that more efficient then a AMD with highly unoptimized code, not sure why you assume Nvidia with almost no one working on it is in the same shoes as AMD. Because MBK added Nvidia support he put just as much effort into Nvidia as his AMD endeavors?

What is a 'similiar method'? I was literally talking about MCU usage. Also calling BS on 60% MCU usage. Give me a screenshot, which you didn't provide for Equihash either.

I like how you base assumptions on loose logic. The whole reason I'm not believing Equihash limits are based purely on memory bus width like Dagger (not bus bandwidth). That's what the whole BCT talk thread was.

Screenshot, 55% average memory controller load, GPU-Z. http://prnt.sc/d8phr4
Doing 160S/s, Claymore 6.0. At the wall ~150w, but haven't tuned the voltage/core as much as I could. If you don't believe screenshot, fire up CM 6.0, new version tomorrow.

I know Equihash miners aren't fully optimized yet. But it is obvious it is memory limited, so even if it is fully optimized, the cards would perform similar. I was saying since 1070 has higher compute (ignoring architecture differences that could favor either card), you may be able to underclock some to reduce power, similarly you could undervolt the 480, but it'ts not worth paying extra $200 to save $2/month.

Card wasn't in the screenshot, but I'll believe you for shits and giggles since no one lies online, especially in a argument. The card is at a a reported 106w, so even if the MCU is at 55% you'll hit TDP before ever maxing out the MCU, unless the code becomes more efficient, but that can be done for Nvidia as well.

This goes to show you even more so that this algo isn't completely memory bound. If it were we wouldn't be hitting TDPs before MCU usage. If TDPs are the limiting factor, efficiency definitely becomes more important. Depending on how this algo will stress the cards when it's finally maxed out, based on what we're seeing right here, it's definitely not just memory bound. Lyra2v2 and NeoS also stress memory, but not enough for it to be the sole bottleneck.

And Wolf0 must google his name every day and BCT.

The first card. Others are 4GB, and different 480s. While compute is more of a factor than for Eth, memory is still a major factor, I highly doubt it would ever be worth getting a 1070 over a 480.