X16R - RVN - Miner head to head test log

[JackIT]

in the branch on the ravens actively promote a certain Silent Miner v1.1.0. I would like to hear here the answers to the question - what are the speeds on it, if someone has already tested it. so already there is a certain instruction on the network to get rid of the dev fee in the Enemy Miner 1.08

I tested 1.0.9 here https://ravenforum.org/topic/33/enemy-1-08-vs-silent-miner-v1-0-9
it lost by 8% to Enemy 1.0.8

Additional, Silent miner has the same 1% dev fee as Enemy


I'm happy to pay a 1% fee to any developer that dedicates their time and effort to helping me make more $$. It's motivation for them to continue development and make better miners.

[1714999178]

1714999178

[1714999178]

1714999178

[1714999178]

1714999178

[JackIT]

Even WITH variability is average.

It just takes time and a lot of patience.

This is why even if there is a huge variation in Algo pairing, over a finite period of time, mined many times over, a pattern and 'average' hashrate surfaces.

The amount of time is determined by the fluctuations of hashfunctions involved, and as such, can still be determined. As mentioned above though, time is the key factor here, regardless of what cards are being used. If there is a static source of testing, and a static method of testing, then comparisons can still be made over a static period of time.

#crysx

Agreed, but you need to determine the amount of time needed to establish an average with an acceptable level of confidence.
is 12hrs enough? that will get you through about 720 permutations of x16r, out of 1.84e+19 possible combinations.

[ztaz]

in the branch on the ravens actively promote a certain Silent Miner v1.1.0. I would like to hear here the answers to the question - what are the speeds on it, if someone has already tested it. so already there is a certain instruction on the network to get rid of the dev fee in the Enemy Miner 1.08

I tested 1.0.9 here https://ravenforum.org/topic/33/enemy-1-08-vs-silent-miner-v1-0-9
it lost by 8% to Enemy 1.0.8

Additional, Silent miner has the same 1% dev fee as Enemy


I'm happy to pay a 1% fee to any developer that dedicates their time and effort to helping me make more $$. It's motivation for them to continue development and make better miners.

thank you for checking out the miner, it's now clear that while still the best on ravens), and pay or not to the developer's interest, this is a strictly individual matter for everyone, although of course you need to thank the developer, and I think that soon Enemy will release a patched version with greater speed than the existing one, which can already remove the commission.

[JackIT]

Pool / Stratum Difficulty Testing is finished.

Results: https://bitcointalk.org/index.php?topic=3385643.msg35598147#msg35598147

[seriousirony]

would running with -500 mem +150 core be detrimental to hashrate vs 0 mem +150 core ie. does OCing mem even do anything in the X16r? 

[JackIT]

would running with -500 mem +150 core be detrimental to hashrate vs 0 mem +150 core ie. does OCing mem even do anything in the X16r? 

look at the first post.. there's a mem overclocking test.

unfortunately I can't overclock core on x16r, without periodically crashing, so I can't test that

[seriousirony]

would running with -500 mem +150 core be detrimental to hashrate vs 0 mem +150 core ie. does OCing mem even do anything in the X16r? 

look at the first post.. there's a mem overclocking test.

unfortunately I can't overclock core on x16r, without periodically crashing, so I can't test that

yeah, I saw test with/without +250 mem and results (+250 mem being slightly better: https://ravenforum.org/topic/34/enemy-1-08-100-tdp-no-oc-vs-250-mem) - I mean You didn't test with actually underclocking mem, does it have negative impact on a hashrate?

[WickedPigeon]

I ran 12hours at each setting and recorded the average hash rate at the end of each 12 hour period.

Unfortunately, there is no such thing as average hash rate for x16r.. you can quote a range of expected hash rate.. but average does not exists.
Why? because there are 16^16 different algo permutations. So x16r in reality is not one Algo, but rather it's 18,446,744,073,709,552,000 different algos

Hey Jack,
Thanks again for the tests you do. I know it takes time, effort, and focus from optimized mining to run your tests. Your analysis helps all us miners find the right tweaks to gain those all-important “last couple percentages.” They really do add up! So thank you.

I like your methodology and would use it if I could tolerate my rigs running that long with sub-optimal settings/miners – clearly this is my problem  .

I agree with you on using averages, and I disagree with you on using averages at the same time.  Yes, you are correct about the algo and how it changes causing there to be millions of algo combos that are always changing which cause all sorts of testing problems. This does make time studies and averages difficult, but not impossible.
 
Here is where I disagree. Having been mining RVN over long period of time, months now, I have an average hash rate that I expect to produce. So by watching/recording shorter periods, 12 hours at a time, I can record “Average Hash Rates” (AHR) for the trailing twelve hours, period after period, minute by minute, and if I do this, I get a bell curve, a normal bell curve.

With a normal bell curve I get all sorts of fun math stuff like, mean and standard deviation. So when I am testing, each 12 hour period may vary, but I can see how that period has performed compared to the mean and the standard deviation of a rig that is a "control" rig. And Voila! Confidence rating for the test.
 
Perfect no, but good. Actually – really good and faster than rotating through my rigs. You can see from the attached photo where the average is reported on Virtopia. On rig 1 thru 4, which are all 1070TI, I get expected averages, and over time, they do hold up. One standard deviation on each rig is about +-2 MH/s and on all 5 rigs combined the SD is +- 6 MH/s. I when I change settings or miners, I can see that easily in those averages and how that period stacks against the norm.
 
So it’s yes and no to averages. No in theory, but with a little data, yes to the bell curve.

https://imgur.com/a/XWWWC0y

[JackIT]

Enemy 1.08 Vs. a1min3r 1.4.2 Vs. Silent Miner 1.1.0:

For this test I used a single 12 GPU 1080Ti rig on Windows 10, split into three mining instances. All three mining instances were balanced to get as close to the same hash rate as possible.
Each mining instance is running at the same time on the same machine mining on the same pool.

All three instances have intensity set to 21, GPU target power set to 100%, no overclocking. Dynamic pool/stratum diff

The instances/rig are managed by Awesome Miner, which restarts the miners for various reasons, based on rules I set.

The plan is to do three rounds of testing, each round lasting approximately 10-12hrs, at the conclusion of each round I'll rotate the miner to a new instance, until each miner has had a chance to run on each instance.
I'll also normalize the results to averaged blocks found per round.

Miners tested:

Enemy 1.08 - 1% dev fee
a1min3r 1.42 32bit - no dev fee
Silent Miner 1.1.0 - 1% dev fee


Results:

Round 1 -  1658 Blocks Found - Duration: 731  minutes (a1min3r = 1 restart, accepted not increased in 4 mins)

• [Instance 1] 211.20 RVN - Enemy - Pool Link
• [Instance 2] 162.76 RVN - a1min3r - Pool Link
• [Instance 3] 183.68 RVN - Silent Miner - Pool Link

Round 2  - not needed

Round 3  - not needed



Average Results FINAL

#1	*	| Enemy 1.08	| 211.20 RVN
#2	-14.98%	| Silent Miner  1.1.0	| 183.68 RVN
#3	-29.76%	| a1min3r 1.42	| 162.76 RVN

Notes:

• a1min3r 1.42: is buggy. miner reported hash rate is way off on some algo(s), often reporting 5-10x expected hash. Looking forward to a revised version
• Silent Miner 1.1.0: I tested Enemy 1.08 Vs Silent Miner 1.0.9, on these same instances just four days ago. The difference in the test is that 4 days ago I used a static pool diff of 50, and TDP was set at 90% and the test ran about 80 minutes shorter. four days ago 1.0.9 lost by 7.27%, today 1.1.0 lost by 14.98%  It could also be the case that Enemy 1.0.8 scales better with more power (90% vs 100%) than Silent Miner LINK TO PREVIOUS TEST

[Max Likelihood]

This is good work JackIt, helps lots of people.

[chrysophylax]

Even WITH variability is average.

It just takes time and a lot of patience.

This is why even if there is a huge variation in Algo pairing, over a finite period of time, mined many times over, a pattern and 'average' hashrate surfaces.

The amount of time is determined by the fluctuations of hashfunctions involved, and as such, can still be determined. As mentioned above though, time is the key factor here, regardless of what cards are being used. If there is a static source of testing, and a static method of testing, then comparisons can still be made over a static period of time.

#crysx

Agreed, but you need to determine the amount of time needed to establish an average with an acceptable level of confidence.
is 12hrs enough? that will get you through about 720 permutations of x16r, out of 1.84e+19 possible combinations.

We must understand that the Algo does NOT go through ALL the permutations ALL the time, only some.

Once that is established, a reasonable amount of time needs to be set.

We at CWI have ALWAYS tested in the long term, even with 'stable' hashrate Algos, of 24hours to 72hours. This allows for any anomalies to enter and then the test is redone over and over again. This is the only way to be sure of the results, and have a more accurate estimation of hashrate.

If you think this is difficult, try Timetravel10 which has more permutations than you can poke a stick at, yet there was NEVER this sort of fuss about it, until RVN came olong. Seems weird how the community can pick and choose which Algo is an Algo to be looked at and picked at, and others to be left alone.

The base conditions for ALL algos is time and non-variable testing equipment. That is pretty much it.

We will not mess with this sort of situation because as with TimeTravel10, a LOT of time and effort is wasted over the smallest things- especially when we were building our miner - CWIgm. So to test this 'properly' make sure you have three weeks available to do so. We will not commit that time to this, we will commit that time to the real work involved in designing, development, and physical work we have been doing.

Have fun doing this though, as it is quite a learning curve if you have never done thins before.

#crysx

[JackIT]

Enemy 1.08 Vs. a1min3r 1.4.2 Vs. Silent Miner 1.1.0

Test Results are can be found: https://bitcointalk.org/index.php?topic=3385643.msg35754113#msg35754113

[WickedPigeon]

Enemy 1.08 Vs. a1min3r 1.4.2 Vs. Silent Miner 1.1.0

Test Results are can be found: https://bitcointalk.org/index.php?topic=3385643.msg35754113#msg35754113

Nice - and that is all she wrote folks.
Excellently done.

[WickedPigeon]

Inspired by JackIT and his 1080TI tests, I ran my own little experiment with 21X1070Ti.

Best ROI on 1070TI with Enemy 1.08, at Virtopia

I have 21 GPUs, all 1070TI, sitting on 4 different rigs, all slightly different builds, one rig is 6x MSI Tritium, another is 5x MSI Gaming, the last two are 5x a mix of EVGA, Gigabyte, and MSI.
4 rigs; 3x5 GPUs and 1x6 GPUs.

All were mining on Virtopia – each rig tracked separately but under one wallet address.  I ran 12hours at each setting and recorded the average hash rate at the end of each 12 hour period.
Perfect test? No. But it does confirm much of what JackIT found in his tests with the 1080TI.

After 12 hours at each setting, the average of all 21, Enemy 1.08, Auto Diff, auto i, Virtopia pool:
85%,+150,+200 – 11.90 MH/s per GPU - Baseline
80%, +150,+200 – 11.95Mh/s per GPU – increase of 0.4%
70%, +150, +200 – 11.58 95 Mh/s per GPU – decrease of 2.7% over baseline
70%,+0,+0 (no OC) – 11.29 Mh/s per GPU – decrease of 5.1% over baseline

The change between 80 and 85% showed that there was no significant difference. Clearly as the TDP was reduced, hashing fell off. As for OCing, there is a benefit to overclocking too. This, at least in my thoughts, confirm what JackIT is seeing in his tests and confirms the settings suggested in Enemy’s Readme file: 80% TDP and slight over clocking.

I know my testing isn't as thorough in methodology as Jack's - but:
I took off the Core over-clocking for 24 hours : 80%,0,+200
The hash rates under-preformed the benchmarks by 1-2%, just enough to be statistically relevant/noticeable.

Now I'm trying to "push" the 1070TI to it's limit. I'm bumping up the core and mem without crashing. I'll post the outcome once finished.

[JackIT]

After my most recent test run, the results were very lopsided. I started to question whether my instances were still well balanced.

So I immediately fired up a test of the instances, using the same settings (no static diff), all three using Enemy 1.08.

The results below. are validation of how balanced they really are, (kinda surprised myself) just 1.5% separates all three instances

Results: 692 Minutes - No restarts

• [Instance 1] 205.21 RVN - Pool Link
• [Instance 2] 205.11 RVN - Pool Link
• [Instance 3] 202.17 RVN - Pool Link

I also found that the pool reported hash rates and avg hasrate line, for the three instances are all over the place, yet produce the same results... I really don't have much faith in Yiimp pool reported hash rates

[carlos33193]

Have you done any testing using nevermore 0.2.2?

[JackIT]

I decided to do pool testing using my three balanced instances. The pools I'm testing are: Suprnova, Ravenminer and CryptoPool.party


I'm sure everyone has heard, "earnings from various pools should more or less even out over the long term" and while that's mostly true, there are several things that can impact earnings on one pool versus the next.
 
Here are some things that can comparatively impact pool earnings.

• Pool fees, can range from 0%-2% (some even higher)
• Stratum Stability, from DDOSing to overcrowding stratums, if the pool is down or struggling to stay up, your earnings will take a hit.
• Orphaned/Stale blocks, pools with better connectivity and infrastructure will have an edge
• Payment calculation methods, PPLNS, PPS, Prop, Yiimp etc (search google for moe info)  

Final Cumulative results after 5 days of testing:



Day 1 results:



Day 2 results:



Day 3 results:



Day 4 results:

[mercoinz]

thank you so much JackIT for your tests! I split test a lot of pools and suprnova always, always comes out on top by a landslide.

[sp_]

New free opensource x16r miner without a fee

https://github.com/sp-hash/suprminer/releases/tag/spmod-git1

will add some more speed later.

[ruplikminer]

New free opensource x16r miner without a fee

https://github.com/sp-hash/suprminer/releases/tag/spmod-git1

will add some more speed later.

Amazing!! when will you release a faster version? I tested this one and it's still slower than the Enemy miner