Vega Rumors

[stealthballer123]

I want that monitor for 60 bucks.

 

[N4CR]

Navi wait for Navi!!!!!!!

V-v-volta! Wait for volta giuyz!

 

[Magic Hate Ball]

Not all FreeSync is "cheap".

See C32HG70, costs the same as XB271HU.

You're comparing a 27" flat G-Sync display that's a year old to a brand new 32" curved Freesync display?

By the way, I got my C32HG70 for $575, so yeah, it can be cheaper already.

 

[JustReason]

Ut oh... we need details! But he isn't really allowed to post any. Can he even post what he did without breaking NDA?

I am hoping for 28Mhs and a break of NDA to make sure all miners know it.

 

[Deleted member 93354]

. doom can run flawlessly in a potato

I take offense to that (you know you said that in GLaDOS voice)

 

[sawe]

I hope Hardocp does some VR testing also, because VR Leaderboard is quite sad to look at if you think of AMD GPU's.

 

[Dayaks]

V-v-volta! Wait for volta giuyz!

That can be a serious suggestion even if you are hard core AMD. Vega prices should drop like a rock when Volta hits.

 

[Presbytier]

I hope Hardocp does some VR testing also, because VR Leaderboard is quite sad to look at if you think of AMD GPU's.

Those leaderboards won't change if this gets bought up by all miners.

 

[DooKey]

V-v-volta! Wait for volta giuyz!

That's actually good advice if someone wants something better than a 1080Ti.

 

[Presbytier]

That's actually good advice if someone wants something better than a 1080Ti.

At this point that is only a real concern for 4k gaming. We are at an inflection point in the technology where you can get by with even lower end mid range cards at high settings unless you want 4k.

 

[Dayaks]

At this point that is only a real concern for 4k gaming. We are at an inflection point in the technology where you can get by with even lower end mid range cards at high settings unless you want 4k.

Or high Hz or VR.

 

[SighTurtle]

https://linustechtips.com/main/topi...ng-performance-is-false-according-to-io-tech/

Tweaktown in the comments states that the mining rumor is false at least according to his testing. A finnish reviewer is also saying the same thing in the above link.

Also, Austin Evans has backed up the August 28th release date of RX Vega 56 during his Vega unboxing video.

 

[N4CR]

That can be a serious suggestion even if you are hard core AMD. Vega prices should drop like a rock when Volta hits.

That's actually good advice if someone wants something better than a 1080Ti.

But the same old argument applies, if you don't have a Ti or 1080/Vega 64, you're going to wait at least 4-5 months? I want something better than a Ti too but I'm not going to wait for god damn navi lol.
We all want 'the' 4k solution but as that darn scientist in Half life said; 'you'll just have to wait'.
I don't think Volta will be that solution either. Maybe for 4k60Hz but not 120Hz+ and heavy VR games maxed.

 

[DooKey]

But the same old argument applies, if you don't have a Ti or 1080/Vega 64, you're going to wait at least 4-5 months? I want something better than a Ti too but I'm not going to wait for god damn navi lol.
We all want 'the' 4k solution but as that darn scientist in Half life said; 'you'll just have to wait'.
I don't think Volta will be that solution either. Maybe for 4k60Hz but not 120Hz+ and heavy VR games maxed.

I believe we are two generations out from a card that can handle 4K 120Hz+

 

[Dayaks]

But the same old argument applies, if you don't have a Ti or 1080/Vega 64, you're going to wait at least 4-5 months? I want something better than a Ti too but I'm not going to wait for god damn navi lol.
We all want 'the' 4k solution but as that darn scientist in Half life said; 'you'll just have to wait'.
I don't think Volta will be that solution either. Maybe for 4k60Hz but not 120Hz+ and heavy VR games maxed.

4-5 months for Volta to lauch and Vega prices to be halved might be worth it to the AMD folks. Hell they've waited years already...

I bought a 1080 on launch so I've been enjoying Vega performance at about the same cost for quite some time now bahahah.

 

[cpuspeed]

https://linustechtips.com/main/topi...ng-performance-is-false-according-to-io-tech/

Tweaktown in the comments states that the mining rumor is false at least according to his testing. A finnish reviewer is also saying the same thing in the above link.

Also, Austin Evans has backed up the August 28th release date of RX Vega 56 during his Vega unboxing video.

not false just misunderstood. those rates are what the large mining farms will get with drivers they develop and not with what AMD releases.

 

[Chris_B]

Just noticed Roy's response to the amd\nvidia gif.

 

[razor1]

not false just misunderstood. those rates are what the large mining farms will get with drivers they develop and not with what AMD releases.

Even in Linux, open source AMD graphics drivers that we have seen thus far never match the closed source versions (some apps get close though). Its just AMD has more intimate knowledge of their hardware.

Not only this, many of the newer miners now work just as good under Windows as they do with Linux. So its not as simple as just drivers.

Drivers just expose what the miners can do. So as long as everything is exposed properly the work being done should be done app side. Granted you can get some more performance out of drivers if there are bugs in the drivers. But don't expect a 30% uplift in performance over Windows counter part. Things like that are mythical.

Easiest way to figure it out is, at least for mining, pure compute performance, how many Tflops ya got that is the increase we should see in mining results if done properly. That's why that 60mhs seems extremely likely to happen with Vega 64. That is when ram timings and latency etc. are similar to Polaris Any more kinda not going to happen.

 

[SighTurtle]

not false just misunderstood. those rates are what the large mining farms will get with drivers they develop and not with what AMD releases.

We don't have evidence that those mining farms have developed better rates. The rumor was based on AMD informing their partners that the cards would have very high hash rates as a result of enabling features on drivers needed for non-mining purposes. If reviewers are using the latest drivers (which they 100% will be) then the rumor is false as is. Unless AMD is holding it back, but I can't think of a reason why. Unless they don't influence the card's performance except in mining, which is a possibility i guess.

 

[razor1]

We don't have evidence that those mining farms have developed better rates. The rumor was based on AMD informing their partners that the cards would have very high hash rates as a result of enabling features on drivers needed for non-mining purposes. If reviewers are using the latest drivers (which they 100% will be) then the rumor is false as is. Unless AMD is holding it back, but I can't think of a reason why. Unless they don't influence the card's performance except in mining, which is a possibility i guess.

Interestingly enough big frames, same hash rates, but because of the luck in mining factor, they tend to hit more lulls but it depends on the size of the farm as well, mainly because they will have their own pool luck will be a bigger factor for them.

 

[cpuspeed]

We don't have evidence that those mining farms have developed better rates. The rumor was based on AMD informing their partners that the cards would have very high hash rates as a result of enabling features on drivers needed for non-mining purposes. If reviewers are using the latest drivers (which they 100% will be) then the rumor is false as is. Unless AMD is holding it back, but I can't think of a reason why. Unless they don't influence the card's performance except in mining, which is a possibility i guess.

You will never have "evidence" from the mining farms, they have every reason to keep things private. I'm surprised they shared what they can do with AMD as even knowing what is possible gives an advantage. The rumor was based on AMD presentation that shared what some of those private mining farm rates are and was misunderstood by the audience. The presenter even said those rates are not what AMD could do because they have have to enable all features. Talked about this in the vega rx pre-order thread in regards to tomshardware article.

 

[N4CR]

I believe we are two generations out from a card that can handle 4K 120Hz+

Probably right on the money there. Wondering if Navi could even do it before Nvidia replaces Volta, hope and pray they have got Navi to at least 4k60-70/Volta and it can scale like TR, then we might even see 4k120 a little quicker. Either way I want to see the competition, AMD might just have another Ryzen on their hands there, they already have 500gb/sec IF links in Vega.

4-5 months for Volta to lauch and Vega prices to be halved might be worth it to the AMD folks. Hell they've waited years already...

I bought a 1080 on launch so I've been enjoying Vega performance at about the same cost for quite some time now bahahah.

Lol fair point and yeah the 1080 in retrospect was/is a pretty good buy, I would've personally avoided the frontier edition tax but always prefer reference cards where possible...

 

[razor1]

You will never have "evidence" from the mining farms, they have every reason to keep things private. I'm surprised they shared what they can do with AMD as even knowing what is possible gives an advantage. The rumor was based on AMD presentation that shared what some of those private mining farm rates are and was misunderstood by the audience. The presenter even said those rates are not what AMD could do because they have have to enable all features. Talked about this in the vega rx pre-order thread in regards to tomshardware article.

The first time this rumor was circulated was not to miners or an AMD presentation that we know of, if you have link so we can see the dates that would be great.

If features are disabled that stop from fully utilizing the shader array. Is pretty much what this guy is getting at. That would be dumb on AMD's part. The only thing I can think of that would even remotely give that much increase will be FP16 capabilities. Which pretty sure its enabled . There is nothing else that increase its calculations per sec. With that much increase in calcs/sec then you will need an increases in cache amounts and bandwidth proportional to that.

 

[cpuspeed]

The first time this rumor was circulated was not to miners or an AMD presentation that we know of, if you have link so we can see the dates that would be great.

If features are disabled that stop from fully utilizing the shader array. Is pretty much what this guy is getting at. That would be dumb on AMD's part. The only thing I can think of that would even remotely give that much increase will be FP16 capabilities. Which pretty sure its enabled . There is nothing else that increase its calculations per sec. With that much increase in calcs/sec then you will need an increases in cache amounts and bandwidth proportional to that.

as I mentioned, talked about it in the pre-order thread. relevant link from that thread.

http://www.tomshardware.com/news/vega-ethereum-mining-performance-rumor,35160.html

I'm sure it doesn't go into as much detail as you'd like and I don't think that is ever going to happen so it's down to whether you believe the AMD VP or not.

 

[razor1]

as I mentioned, talked about it in the pre-order thread. relevant link from that thread.

http://www.tomshardware.com/news/vega-ethereum-mining-performance-rumor,35160.html

I'm sure it doesn't go into as much detail as you'd like and I don't think that is ever going to happen so it's down to whether you believe the AMD VP or not.

he said nothing about drivers he stated changing the firmware, bios man, yeah we are talking about ram timings, functionality and what not. Drivers can't do that.

All of miners do that man, not only do we do it we also get 30% increase in hashrates by doing so. At no time was he talking about Vega specifically, these are just normal things if you are a miner would do to get the best mining performance out of your cards.

Look up hashrates for polaris that are put up on review website, you will see it gets 21-25 mhs depending on which memory it comes with, once you mod it though, (change its ram timings) it will go up to 29 mhs easily then you over clock it and what not to get more.

When I'm single mining I'm getting 32 mhs. So there ya go, big difference with default settings vs. And if I don't change ram timings, the overclocking does almost nothing for the hash rates.

 

[cpuspeed]

he said nothing about drivers he stated changing the firmware, bios man, yeah we are talking about ram timings, functionality and what not. Drivers can't do that.

All of miners do that man, not only do we do it we also get 30% increase in hashrates by doing so. At no time was he talking about Vega specifically, these are just normal things if you are a miner would do to get the best mining performance out of your cards.

Look up hashrates for polaris that are put up on review website, you will see it gets 21-25 mhs depending on which memory it comes with, once you mod it though, (change its ram timings) it will go up to 29 mhs easily then you over clock it and what not to get more.

When I'm single mining I'm getting 32 mhs. So there ya go, big difference with default settings vs. And if I don't change ram timings, the overclocking does almost nothing for the hash rates.

he mentioned OpenCL AND they go into firmware. Probably also use different algo's as well, since large miners solo mine they can view things differently than those who have to join a pool.

as I talked about it in the pre-order thread:

Well, he does go on about public and private hash rates and the private ones being "whoa". community mining forums must be considered public and going from ~25mh/s to ~30mh/s is not "whoa" to me and that's what you get with those memory strap mods. Going 2x, or 3x hash rate is "whoa" to me so I think the 70-100mh/s is a private rate that is not going to be shared.

I've read people getting 31-32mh/s with 1070 using recent update to claymore or etheminer (think it was the later that had update algorithm with input from Nvidia that speed up rates and then claymore followed). You should be able to get mid 30s with your 1080ti, assuming you mining ETH which might not be the most profitable to mine with that card anyways.

edit: reading the VP quote again, it seems clear that huge increases in hash rate are NOT going to come from an AMD driver update, assuming quote is genuine and true. He basically says in order to do that you have to break something that you're not going to be using and AMD can't do that as they have to enable all features to work. None of the public community mods that I know of required installation of or have special tweaked driver. There is a leaked driver that fixes DAG epoch issue but that's not huge hash increase and it doesn't break anything like gaming.

 

[razor1]

he mentioned OpenCL AND they go into firmware. Probably also use different algo's as well, since large miners solo mine they can view things differently than those who have to join a pool.

as I talked about it in the pre-order thread:

The algorithm is only part of it, do you know what can't be getting around in Eth mining? Its the Dag file, that is why cards with 8gb are needed to get the most performance out of them. And that is also why bandwidth and timings are so important to Eth mining. There is no way around that. The Dag file is extremely large and if there are certain amount of calculations going on it needs a certain amount of bandwidth to sustain performance.

So if you increase your calculations per second, past a certain point it WON'T matter.

You are taking what that VP stated and trying to make reasons for the 100mhs rumors. DOESN'T WORK that way. What the VP stated is what most serious miners do anyways. That Dag file is want is most important to ETH mining, doesn't matter how much calculation power you have if the bandwidth isn't there. You will get bottlenecked.

They were specifically talking about eth mining with that 100mhs rumor, That is why I can say its BS even for special groups that can do more then others. Theoretically its just not possible with the bandwidth Vega has. This is why clock speed of the GPU alone doesn't do much for Eth mining, you can get 10% increase with clocks if you already increased bandwidth amounts by that 30%. But clocks alone won't give you much.

And if you want to test this out. Just look at the 1060 vs the 1070 vs the 1080. Or even better take a 1070 down clock to to the same Tflops and bandwidth of 1060 and test out the hashrates. You will end up almost the same hashrates.

Then up the tflops of the 1070 and leave the bandwidth as it is. Hashrates won't increase much.

So if Vega can do 100mhs, with its current bandwidth , its out of theoretically specs. We already know HBM2 can't overclock much. Does that make any sense it can do that?

 

[cpuspeed]

I'm just trying to point out where the Vega 100 mh/s rumor came from and was misunderstood as coming from an AMD driver update.

 

[razor1]

I'm just trying to point out where the Vega 100 mh/s rumor came from and was misunderstood as coming from an AMD driver update.

Doesn't matter where it comes from to hit that hashrate Vega would need 1 Terabyte/sec of bandwidth.

All the special bandwidth savings tech, like compression don't work in compute work loads.

 

[cpuspeed]

Doesn't matter where it comes from to hit that hashrate Vega would need 1 Terabyte/sec of bandwidth.

All the special bandwidth savings tech, like compression don't work in compute work loads.

I'm not trying to rehash the arguments from when the rumors first came out a week or so ago. I'm not an expert in this area anyway and if I was and new how to frankly I wouldn't be sharing it with you or any open forum. I'd be shopping my resume to the large mining firms for the big bucks.

 

[NoNRG]

Apparently even the reviewers packaging says about a minimum 1000 watt psu, for the one that got the aio version at least.

I have a Corsair 750W power supply. Do I need to upgrade this if I want to run a Vega 64 Liquid? Do we know yet, or should I wait to find out on Monday?

 

[Chris_B]

I have a Corsair 750W power supply. Do I need to upgrade this if I want to run a Vega 64 Liquid? Do we know yet, or should I wait to find out on Monday?

That guy from the adored tv channel said he's running the liquid one on a corsair rm 750 no problems in the "turbo" mode so seems like it should be enough, really depends on the rest of your system.

 

[Araxie]

I have a Corsair 750W power supply. Do I need to upgrade this if I want to run a Vega 64 Liquid? Do we know yet, or should I wait to find out on Monday?

with the system in your sig, if your PSU is a high quality one you should be safe.. everything depend on the rest of the system specs and overclocks..

 

[razor1]

I'm not trying to rehash the arguments from when the rumors first came out a week or so ago. I'm not an expert in this area anyway and if I was and new how to frankly I wouldn't be sharing it with you or any open forum. I'd be shopping my resume to the large mining firms for the big bucks.

The problem with that is people like Claymore take 2% of your mining profits, anyone that can make mining software run that much more would make a KILLING going on and taking from pools, the biggest pools have close to 20000 workers, you know what that means, 2% of those 20000 workers 6 cards per worker, is how much money?

Yeah shopping around for private miners, they will pay you a flat rate + a small % maybe if you are that good. Guess what when we are talking about hundreds of thousands of graphics cards being used in pools that 2% will blow away anything one guy can give.

This is why solo mining is not recommended because the pool luck will get you all the time, you need to have a butt load of rigs to ensure pool luck for yourself, and even then it still hurts when you have a few hours of bad luck. It can go on for a day or two as well, not just hours.

Simple economics.

I think I know a bit about programming to know what can be done and what can't be done in reasonable aspects. Whom ever pulled that 100 mhs number pulled it out of their ass, its just not possible to do that much with one card right now.

And if you don't believe me about Eth being bandwidth bound, just look up Dagger Hashimoto, that is the algorithm Eth's blockchain is based off of.

They don't use multiple algorithms per coin, its one algorithm and that is it (some coins are combinations of previous algorithms, but the mining software must use all of those algorithms as specified by the coin otherwise it can't mine that coin).

here is the link about the Algo for Eth

These are the guys that made the damn thing they really should know what they are talking about right?

https://github.com/ethereum/wiki/wiki/Mining

Our algorithm, Ethash (previously known as Dagger-Hashimoto), is based around the provision of a large, transient, randomly generated dataset which forms a DAG (the Dagger-part), and attempting to solve a particular constraint on it, partly determined through a block's header-hash.

It is designed to hash a fast verifiability time within a slow CPU-only environment, yet provide vast speed-ups for mining when provided with a large amount of memory with high-bandwidth. The large memory requirements mean that large-scale miners get comparatively little super-linear benefit. The high bandwidth requirement means that a speed-up from piling on many super-fast processing units sharing the same memory gives little benefit over a single unit.

What does that mean, well pretty much ya need a butt load of bandwidth and memory more than processing power. That is why frequency more cores etc doen't do much for Eth mining if the bandwidth isn't there.

 

[Anarchist4000]

But the same old argument applies, if you don't have a Ti or 1080/Vega 64, you're going to wait at least 4-5 months? I want something better than a Ti too but I'm not going to wait for god damn navi lol.

At that time you're also looking at a refresh of Vega. With the typical 3-5 year upgrade cycle for most of the market, 6 months is nothing.

 

[Anarchist4000]

Which is actually worse than before, because their effectiveness is rather limited right now, ie: they don't offer any tangible performance increase in current games. Vega appears locked at 4 tri/cycle. At least with developer intervention hope is there for it to offer more with proper handling. Now it won't even do that.

You act like 4 tri/cycle is actually a problem. That's 4/cycle before you start tessellation and is based on binning geometry into four quadrants in screen space. For almost all titles, excluding CAD, that is more than sufficient. They will offer tangible gains in current games as Mantor explained. Just not in peak geometry rate without explicit programming. That wasn't a possibility before and something devs have requested.

In the case of VR and other upcoming titles with TBDR, the issue is even less as the scene gets reprojected. Best way to efficiently push 240Hz and even non-VR titles will likely adopt similar tech. Techniques like async spacewarp with improvements so devs can do more with less.

Did you read Vega's white paper?

No, it hasn't been released yet. You get an advance copy? All I've seen was the ISA paper.

There is no damn difference between older GCN to VEGA when it comes to instruction sets lol (outside of FP 16), You know what that means? Primitive shaders is nothing new They just are using a different name for what programmers have used in the past.

Beyond the 40ish new instructions that we're exposed? That's not necessarily all of them either, just what's released.

Primitive shaders get rid of the black box and allow a programmable pipeline. Beyond what used to be driver optimizations, they would be limited by the old pipeline structure without game specific testing. A lot is possible without those limitations and that's what devs have been presenting papers on.

And ya RYS, what he just stated is BS, he lied....... He didn't answer the question properly, It doesn't auto magically work, AMD has to do the work and guess what that is why you won't see more than Polaris's geometry through put most of the time, I have already confirmed this. So unless the application is changed to take advantage of primitive shader, we will not see any improvements in Vega when it comes to geometry through put.

It doesn't work because it just works? The drivers have always done that work, only difference being they are likely culling a bit more efficiently. The attribute fetching was the big one Mantor mentioned that should work with existing games. Would likely entail their intelligent workgroup distribution as well.

Eventually primitive shaders will have dynamic memory and scheduling capabilities. They seem to exist to control AMDs binning process in addition to optimizing the culling. Passing hints from prior frames to better predict distributions.

 

[razor1]

At that time you're also looking at a refresh of Vega. With the typical 3-5 year upgrade cycle for most of the market, 6 months is nothing.

Vega won't be refreshed that quickly, why would it be, Polaris refresh was one year and did we see anything extra from that? 5% more performance at a cost of 30% more power?

 

[godihatework]

not false just misunderstood. those rates are what the large mining farms will get with drivers they develop and not with what AMD releases.

Moving goalposts.gif

I mean, Jesus Christ dude.

 

[razor1]

You act like 4 tri/cycle is actually a problem. That's 4/cycle before you start tessellation and is based on binning geometry into four quadrants in screen space. For almost all titles, excluding CAD, that is more than sufficient. They will offer tangible gains in current games as Mantor explained. Just not in peak geometry rate without explicit programming. That wasn't a possibility before and something devs have requested.

Its good for yesterdays games not tomorrows games. Effectively double the possible triangle counts of games but the thing is we already see Fiji hitting today's game limits on possible polygon counts, look at the games where Polaris creeps up to Fiji's frame rates.

No, it hasn't been released yet. You get an advance copy? All I've seen was the ISA paper.

Yeah not much difference, and if you looked at the ISA paper, you can see the pipeline is essenatially the same.

Beyond the 40ish new instructions that we're exposed? That's not necessarily all of them either, just what's released.

Almost all of them are for FP16. hence why I stated not much difference other than that in the shader array.

Primitive shaders get rid of the black box and allow a programmable pipeline. Beyond what used to be driver optimizations, they would be limited by the old pipeline structure without game specific testing. A lot is possible without those limitations and that's what devs have been presenting papers on.

But nV's architecture doesn't hit those limitations, only AMD's do...... That is because they only have so many geometry units.

It doesn't work because it just works? The drivers have always done that work, only difference being they are likely culling a bit more efficiently. The attribute fetching was the big one Mantor mentioned that should work with existing games. Would likely entail their intelligent workgroup distribution as well.

Look first off you thought Vega was going to be a 1080ti killer because of all the specs AMD was shouting out for close to a year now, Now that is not going to happen and its obvious. So you are going to tell me, what they just stated about primitive shaders are going to make any difference. If developers don't have that control over them, forget it, its polygon through put will be no better than Polaris for almost all titles. Past and near future, untill developers have access to it. AMD will not be able to do it through drivers as I stated unless initiation and propagation of vertices are done with FP16 there will be no way for primitive shaders to automatically done via drivers.

Eventually primitive shaders will have dynamic memory and scheduling capabilities. They seem to exist to control AMDs binning process in addition to optimizing the culling. Passing hints from prior frames to better predict distributions.

I do agree on that point it should, but it also depends on how its implemented on AMD hardware don't know that yet.

 

[cpuspeed]

The problem with that is people like Claymore take 2% of your mining profits, anyone that can make mining software run that much more would make a KILLING going on and taking from pools, the biggest pools have close to 20000 workers, you know what that means, 2% of those 20000 workers 6 cards per worker, is how much money?

Yeah shopping around for private miners, they will pay you a flat rate + a small % maybe if you are that good. Guess what when we are talking about hundreds of thousands of graphics cards being used in pools that 2% will blow away anything one guy can give.

This is why solo mining is not recommended because the pool luck will get you all the time, you need to have a butt load of rigs to ensure pool luck for yourself, and even then it still hurts when you have a few hours of bad luck. It can go on for a day or two as well, not just hours.

Simple economics.

I think I know a bit about programming to know what can be done and what can't be done in reasonable aspects. Whom ever pulled that 100 mhs number pulled it out of their ass, its just not possible to do that much with one card right now.

And if you don't believe me about Eth being bandwidth bound, just look up Dagger Hasimoto, that is the algorithm Eth's blockchain is based off of.

They don't use multiple algorithms per coin, its one algorithm and that is it (some coins are combinations of previous algorithms, but the mining software must use all of those algorithms as specified by the coin otherwise it can't mine that coin).

Maybe not so simple economics.

If someone released a new ETH miner software that did say 10x current rates and only asked for a 4% fee. You think he would make more than Claymore does now, right? However, he wouldn't make anymore than Claymore does now because difficulty rates would increase which would adjust the rewards rates down proportionally. He would probably make less, probably need to charge 20% fee to break even with Claymore. Only advantage is to keep it to yourself, relatively speaking.

Also any software with increase rates would get reverse engineered and copied which I think has happened between Claymore, ethminer and sgminer. They all have about the same rates but Claymore's easier to setup, especially for dual mining.

Also, you can cheat the Claymore devfee and have it sent to yourself, though he makes a lot to be sure.