[Simmons572]

Hey everyone, it looks like this thread is seems to be the go-to for suggesting changes to the TC, so I will attempt to keep this up to date as proposals are made.


March 2019

Adding GPU's to the TC:

GPU-O

• RTX 2080 (already in place)
• RTX 2070
• Vega VII

GPU-L

• RTX 2060
• GTX 1660 Ti

AMD

• RX 590

The 2080 Ti, Titan V and Titan RTX seem to be a bit too big for the time being.



Last Ruling, Enacted Fall 2018:

Spoiler

Sup folks. After getting caught up on everything that has transpired over the past several weeks, I would like to propose a re-balance of the TC Categories.

My proposed changes are as follows:

I7/R7

• 16C/32T - Increased from 6C/12T Cap

GPU-O

• Titan Xp
• Titan X(P)
• 1080 TI
• 2080
• Any GPU-L GPU

GPU-L

• 1080
• 1070 TI
• RX Vega 64/56
• 1070
• Titan X (Maxell)
• 980 TI
• Any nVidia GPU
• Any AMD GPU

nVidia

• 980
• 1060
• 970
• 780 TI
• Any nVidia GPU-W GPU

AMD

• Fury X
• Fury
• RX 580
• RX 480
• R9 390X
• Any AMD GPU-W GPU

GPU-W

• GTX 960
• GTX 1050 TI
• GTX 780
• R9 380X
• R9 280X
• RX 560
• Any Other GPU, ask Editor for Verification

These lists above are just my personal suggestions, and I am willing to make adjustments based on community input.
Please post your comments, suggestions or opinions below :thumb:

 

[Simmons572]

As for some justification for my category choices:

The top portion here is the current category limits, the average PPD based on the PPD database, and the highest grossing WU's based on the database.
The bottom portion is my proposed change, based on numbers from the database.

By increasing the Upper limit of GPU-O and GPU-L, there will not be as significant of a gap between the 2 categories. I expect there to be an overlap until we can get proper hardware slotted into GPU-O.
AMD remains unchanged, until we can get more info about the Vega 56/64/FE GPUs.
nVidia's upper limit has been raised to match the AMD category. This way teams can remain a bit more competitive cross categories.
GPU-W's upper limit has been raised to match the former nVidia category. This should put PPD close to what 16C/32T CPU's can put out these days.

 

[Simmons572]

I can get some Vega 64 numbers posted later this week. I'm retiring a Vega 64 from my testing server, and moving it into my main machine later in the week. Most of the numbers will be in Linux, but I could run it in Windows for a bit before I swap it into my main rig.


On the subject of the new category limits, what you have there looks good so far. I especially like the idea of bumping up the CPU categories. Even though CPU folding is not what it once was, we may be able to snag a few more people with cores to spare, now that high thread count CPUs are becoming more prevalent.

Thanks for the input bud. And Vega 64 numbers would be great to have, that should give us a good idea about where Vega 56 performs as well.

1050/1050Ti cards are common. I'm guessing GPU-W? Might want to just find a spot for the last 2 generations of AMD/NV cards. Vega/Polaris and Pascal/Maxwell.

The 1050 ti looks like it sits at 193k PPD on average. That would be just above the 780. I think it makes sense, so I will add it to the list

I believe we have covered the spectrum of Pascal here. Once we get some info about the Vega GPUs, I was considering putting Vega at the top end of the AMD category, which would mean making the 980ti the top end of the nVidia category. I think that could definitely work, but we will need more info first.

Thanks for the input folks!

 

[Simmons572]

In the past it has usually been OK to use higher core count CPUs, as long as the the folding was limited to the maximum number of allowed threads with the client.

Basically this. By setting the limit at 16C/32T, people will be able to use basically whatever CPU they want in this day and age, and if they have a CPU that has more cores, just limit it to the 16C/32T requirement.

 

[Simmons572]

I think the biggest pain for all of this is getting hardware shuffled around, so once we get some more input, we can decide on a date to put forth the change, and I can work with the editors to ensure a smoov transition

 

[Simmons572]

Is that Titan X Pascal... Or Titan Xp? I'll join the party again if one of my Titan Xp can come and play.

My apologies, that is the titan X(P). My concern with the Xp is that we don't have a lot of info in the database right now to determine performance. Between the 2 WU's that were submitted, it looks like it averages about 1.8m PPD, but I am not sure if that is accurate.

 

[Simmons572]

That would be greatly appreciated. If the numbers are close to the X(P), then I see no reason in omitting it from the category.

 

[Simmons572]

The PPD database was just udpated last night. Thanks for submitting WU's folks :thumb:

Looks like numbers are staying roughly the same, with the Titan Xp's PPD still sitting only slightly above the X(P). I would like to see a few more WU's submitted, but with these numbers, we should be able to add it to the GPU-O category.

Also, I am still looking forward to seeing some Vega WU's submitted

Keep on keeping on folks :thumb:

 

[Simmons572]

Upon reviewing everything that has been submitted as of this morning, I feel that the current list of proposed hardware looks like it's fine.

The only significant changes I have noticed is that there have been a few high PPD WU's submitted for the Titan Xp that are higher than 1.7m PPD. However, that appears to only be 100k higher than the highest batch of 1080 ti WU's.
Depending on how many WUs get submitted moving forward, we may add the 2080 to the GPU-O mix.

The only future changes that I foresee happening is adding the Vega GPUs to the AMD category. If that happens, and the PPD falls where I am expecting, then we will probably shift the 980ti into the nVidia category.
I am also considering picking up a Vega 56 and Vega 64 GPU at some point, if I can snag pre-owned GPU's at a good price.

All of this being said, I would like to push these changes to the TC on 01November. This should give us enough time to figure out how we are adjusting our teams' hardware slots.
If anybody has any issues with passkeys, or need their information verified, please reach out to @lanofsong, @axipher or myself.

Also, if anyone has any further questions and concerns, please post them here, or we can discuss them in PM.

 

[Simmons572]

I honestly haven't been following the PPD numbers of everything so my opinion in terms of what hardware goes in what category is probably not the greatest.


I am definitely all about keeping the categories up to date though, so I support the changes and we probably should be keeping track of new hardware and driver changes.


From what I can recall, we are normally stuck waiting for 6 months after a new GPU launch to even start to really see what it can do and try to classify it in a category.

Thanks for the input bud. Historically, the biggest contingency to adding "big" GPU's to the fold has been seeing how much they outperform previous generations. And honestly, the step from the 9xx series to the 10xx series has not been all that significant. Hell, looking at the single WU that has been submitted for the 2080 under Ubuntu puts it right below the Titan X(P)/p's. Hopefully we get more submissions on the newer GPUs soon.

And on a side note, please AMD for the love of heck give us an affordable GPU that competes with nVidia stuff. I want some bigger GPUs in the AMD category lol

Looks like in Linux I finally got a few nice WUs... haven't seen many... but I did get a couple so far.

P11728 TPF= 45 sec for an estimated 1,703,843 ppd

Good to know. Like I said above, depending on more WUs submitted for the 2080, we may actually be able to get it added to GPU-O sooner rather than later

 

[Simmons572]

Yeah I know..

I can still dream at least

 

[Simmons572]

We submit completed WU information to the site's PPD database: https://www.overclock.net/forum/55-...eam/475163-gpu-projects-ppd-database-106.html

Not really for data validation, but it's a good source when collecting information about certain GPUs.

 

[Simmons572]

That change was made when the Ryzen CPUs initially came out. Then the 8700k was allowed in the fold.

I figured that by lifting the CPU limit to 16c/32t, we can get some old 2p systems or some modern beefy CPUs a workout

EDIT: Wait a second, you are right. The previous limit was 6c/12t, that is my bad. Good catch @notyettoday

I am tempted to reduce the thread count to 8c/16t, but at the same time, I feel a higher cap of 16c/32t would actually make the CPU category competitive. What do you all think?

 

[Simmons572]

I think opening up the CPU thread cap is a good idea. I feel like 16/32 is a good limit. Especially with First gen Threadrippers coming down so far in price.

If we do end up shifting the 980ti into NVIDIA, then would the 1070 be eligible for NVIDIA as well?


*Edit* I also agree with upping the core count in the CPU cat. :thumb:

Thanks for the input about the CPU adjustment. I would like to get a couple more opinions before we solidify this idea, but I am also in favor of the large boost.


@tictoc

Spoiler

It is definitely worth considering. Especially with how close the upper-limit WU's appear to be to each other. However, a lot of the WU's are pretty old at this point. It seems like the newest WU submission for the 1070 is from 11Nov2017.

I am not sure what the difference in PPD would look like with current WU's. My concern is breaching the 1Mil PPD mark in the nVidia Category. The average PPD for the 980ti currently looks like 657k PPD, but the upper limit of WUs is in the 800-900k PPD range. Granted, the 1070 average is 702k PPD, wit the upper limit also sitting in that 800-900k range..

I don't really want to make an opinion on this until we see what Vega is fully capable of.

I would also like to get some more WU's from the 1070 and 980ti to see what these devices are capable of in current year.
I have a 1070 waiting for a PCIe extension cable, and a 980ti that needs to be plugged in, so I am going to start crunching some of these WU's over the next couple weeks.
I am also keeping my eyes out for a "cheap" Vega 64 card, so I can have it staged for if we decide to add it to the AMD category.

 

[Simmons572]

Good to know. That is a lot more reasonably positioned than I expected.

And it looks my PCIe extention cable is coming in today, so I will be able to get that 1070 up and running sooner than I expected

 

[Simmons572]

Fantastic, I really appreciate it bud

I am now negotiating for a Vega 64 on ebay, so hopefully this works out.

 

[Simmons572]

I ordered a Vega 64 Powercolor GPU over the weekend for what I believe is a decent price. Once that shows up, I can start collecting WU information, and start tinkering with overclocks

 

[Simmons572]

@tictoc @mmonnin @CynicalUnicorn @WonderMutt @NBrock @msgclb @navjack27 @u3b3rg33k @CptAsian @axipher @PimpSkyline @notyettoday (I wish there was a more graceful way to mass-notify...)

I would like to get a tally on what you all think we should do with the CPU core cap. https://www.strawpoll.me/16657228

I can't figure out how to retroactively add a poll to the thread, so this will have to do.

 

[Simmons572]

Welp, the vote seems unanimous at this point. We will be sticking with the proposed 16C/32T limit

 

[Simmons572]

The last big hardcore testing that was done on optimizing CPU folding was when @navjack27 did a super deep dive into optimizing Ryzen folding.

But absolutely, I would love to see research done on this category.