Anyone know if we can convert MB/s of SHA256 to MH/s?
If MB is megabytes then 1 MB/s of SHA256 would be 1*2^20/32/2 = 0.016384 MH/s because there are 32 bytes in one SHA-256 and Bitcoin uses two of those. But the SHA256 algorithm that "SiSoftware Sandra 2012" uses is probably a lot slower than the SHA256 OpenCL kernel that the Bitcoin community has. So these figures aren't really comparable to the figures we usually use. 
Topic: AMD A10 the infernal hashing Hellraiser? (Read 8764 times)
I know it sucks but FM1 was a disposable socket platform by architecture. The reason isn't the CPU so much but the GPU elements change so darned fast ...
Let's face it, the GPU portion of these chips is AMD's only claim to fame, they will change the graphics components to the last gen of video cards as they release new designs and not every socket is going to be capable of handling a new GPU design no matter how careful you are with pre planning.
The iteration beyond this using GCN likely won't be FM2 compliant, but there are chances that you'll be able to buy an FM2+ board shortly to buy yourself one more generation of CPU compatibility at least. So while FM1 is locked and forgotten, a board featuring the FM2+ design will manage this round and the next round of APUs.
I have an FM1 with a 3870K Oc'd to 3.5Ghz CPU cores and 900Mhz GPU cores at stock volts with DDR3 1866 sitting on my desk in an open frame and it's my favourite test rig ever. Responsive, lots of features and idle power draw is great considering it's got all that functionality crammed on one chip. I have no regrets about buying it and it will make some kid a happy camper when I upgrade the test bed to FM2+. These APU's are definitely my new go to rigs for low cost home computing. Unreal functionality for a very low price.
At this low end of computing, I'm pretty brand agnostic. I just want the most bang for the buck and I think these are it.
Let's face it, the GPU portion of these chips is AMD's only claim to fame, they will change the graphics components to the last gen of video cards as they release new designs and not every socket is going to be capable of handling a new GPU design no matter how careful you are with pre planning.
The iteration beyond this using GCN likely won't be FM2 compliant, but there are chances that you'll be able to buy an FM2+ board shortly to buy yourself one more generation of CPU compatibility at least. So while FM1 is locked and forgotten, a board featuring the FM2+ design will manage this round and the next round of APUs.
I have an FM1 with a 3870K Oc'd to 3.5Ghz CPU cores and 900Mhz GPU cores at stock volts with DDR3 1866 sitting on my desk in an open frame and it's my favourite test rig ever. Responsive, lots of features and idle power draw is great considering it's got all that functionality crammed on one chip. I have no regrets about buying it and it will make some kid a happy camper when I upgrade the test bed to FM2+. These APU's are definitely my new go to rigs for low cost home computing. Unreal functionality for a very low price.
At this low end of computing, I'm pretty brand agnostic. I just want the most bang for the buck and I think these are it.
Well, they were due for some socket flipping. The AM2->AM2+->AM3->AM3+ progression has been pretty good to piecemeal upgraders.
Nope, FM2 socket.
Are you serious?Wow that must have been the fastest obsolescence of a socket ever.
Thanks AMD. Really.
Those were my thoughts.
Nope, FM2 socket.
Are you serious?Wow that must have been the fastest obsolescence of a socket ever.
Thanks AMD. Really.
I'm sorry, but did you even watch that video? I've never cared less for a CPU comparison while listening to her... TH's plan backfired...
Nope, FM2 socket.
Will A10 use FM1 socket like A8?
As mentioned it's a 69xx series GPU architecture, not GCN. I totally expect it to perform WITH overclock at about 1/2 of a 7770 in keeping with AMD's targets for these devices from a performance perspective.
Since we're still talking Mining and not gaming , I don't expect it to do any better at all than it's llano counterpart from the previous gen. We already know more or less per shader how fast a 6900 series performs at given frequencies and we already know mining isn't contingent of high RAM speeds.... so it should be easy to extrapolate by number of shaders at any given frequency. It is exactly the same 69xx architecture but with lower memory access speeds and lower base frequency. A 6950 has 1408 stream processors at 800Mhz so take it from there.
For gaming it might clock higher and perform better than a Llano if you use faster RAM but really, if you want to game even medium well you can pick up a used 6870/5850 for a 100 bucks and actually play with some eye candy on at 1080.
Still can't argue with the value proposition it offers though for light gaming at lower than HD resolutions (like laptop 720 /900panels).
Since we're still talking Mining and not gaming , I don't expect it to do any better at all than it's llano counterpart from the previous gen. We already know more or less per shader how fast a 6900 series performs at given frequencies and we already know mining isn't contingent of high RAM speeds.... so it should be easy to extrapolate by number of shaders at any given frequency. It is exactly the same 69xx architecture but with lower memory access speeds and lower base frequency. A 6950 has 1408 stream processors at 800Mhz so take it from there.
For gaming it might clock higher and perform better than a Llano if you use faster RAM but really, if you want to game even medium well you can pick up a used 6870/5850 for a 100 bucks and actually play with some eye candy on at 1080.
Still can't argue with the value proposition it offers though for light gaming at lower than HD resolutions (like laptop 720 /900panels).
I stand corrected, it's not GCN. However, here are the changes that do happen:
A8-3870K - 400 VLIW5 cores @ 600MHz.
| l | |
V V V V
A10-4600 - 384 VLIW4 cores @ 686Mhz.
A 6970 - 1536 VLIW4 cores @ 950MHz = 415Mh/s.
Converting from 1536 -> 384, and we get 103MH/s @ 950Mhz (which is almost a 40% OC). Still not looking that great...
As mentioned it's a 69xx series GPU architecture, not GCN. I totally expect it to perform WITH overclock at about 1/2 of a 7770 in keeping with AMD's targets for these devices from a performance perspective.
Since we're still talking Mining and not gaming , I don't expect it to do any better at all than it's llano counterpart from the previous gen. We already know more or less per shader how fast a 6900 series performs at given frequencies and we already know mining isn't contingent of high RAM speeds.... so it should be easy to extrapolate by number of shaders at any given frequency. It is exactly the same 69xx architecture but with lower memory access speeds and lower base frequency. A 6950 has 1408 stream processors at 800Mhz so take it from there.
For gaming it might clock higher and perform better than a Llano if you use faster RAM but really, if you want to game even medium well you can pick up a used 6870/5850 for a 100 bucks and actually play with some eye candy on at 1080.
Still can't argue with the value proposition it offers though for light gaming at lower than HD resolutions (like laptop 720 /900panels).
Since we're still talking Mining and not gaming , I don't expect it to do any better at all than it's llano counterpart from the previous gen. We already know more or less per shader how fast a 6900 series performs at given frequencies and we already know mining isn't contingent of high RAM speeds.... so it should be easy to extrapolate by number of shaders at any given frequency. It is exactly the same 69xx architecture but with lower memory access speeds and lower base frequency. A 6950 has 1408 stream processors at 800Mhz so take it from there.
For gaming it might clock higher and perform better than a Llano if you use faster RAM but really, if you want to game even medium well you can pick up a used 6870/5850 for a 100 bucks and actually play with some eye candy on at 1080.
Still can't argue with the value proposition it offers though for light gaming at lower than HD resolutions (like laptop 720 /900panels).
There is a lot of modern games, that I play on this laptop (Llano based). I admit I've also overclocked the laptop to perform at over 3.0Ghz (standard clock is 1.6), but that is besides the point, the GPU is really what allows it to actually do so much in modern games today that are so GPU based.
With the Trinity, I look forward to seeing how far I can push this one to.
With the Trinity, I look forward to seeing how far I can push this one to.
I've tried playing SWTOR on an A8-3850, which has the same 6550D as the A8-3870. It's actually not that impressive. Granted this was at 1080 with most settings on Med, but you start running into sever bandwidth issues. APUs are great, but you really need dedicated GDDR5 if you wanna play any games.
At 1080P, yes. Can't really expect to jump from 800x600 low details integrated performance to 1080P in only a couple of generations. Especially since the DDR4 needed to really increase bandwidth won't be in volume until end of 2013 or beginning of 2014.
Edit: It's a good thing bitcoin hashing doesn't use much bandwidth.
The A10-4600M has a Radeon HD 7660G (384 cores @ 686MHz) (I'm assuming since it starts with a 7 its GCN)
A 7750 has 512 cores, and at 900MHz gets 136MH/s.
Doing the math from 512 cores -> 384, and from 900MHz -> 686MHz gives us ~77MH/s.
Assuming a 900MHz overclock (just like Ilikeham's A8-3870), that gives us ~102MH/s.
As far as MHs/W, we don't know. But, those GCN are pretty power effecient, so it's bound to be pretty decent. I'm gonna throw it out there and *guess* at under 20W for ~77Mh/s, and under 30W for ~102MH/s.
A 7750 has 512 cores, and at 900MHz gets 136MH/s.
Doing the math from 512 cores -> 384, and from 900MHz -> 686MHz gives us ~77MH/s.
Assuming a 900MHz overclock (just like Ilikeham's A8-3870), that gives us ~102MH/s.
Do we know enough about this proc to estimate mhash/w?
As far as MHs/W, we don't know. But, those GCN are pretty power effecient, so it's bound to be pretty decent. I'm gonna throw it out there and *guess* at under 20W for ~77Mh/s, and under 30W for ~102MH/s.
Wow, thats actually really beefy for an integrated GPU. you could easily play modern games on that, with some compromises on quality options of course.
I've tried playing SWTOR on an A8-3850, which has the same 6550D as the A8-3870. It's actually not that impressive. Granted this was at 1080 with most settings on Med, but you start running into severe bandwidth issues. APUs are great, but you really need dedicated GDDR5 if you wanna play any games.
It's actually VLIW4 cores, so it may not actually be much better than the highest end Llano given it's lower core count. The next APU after Trinity, Kaveri, is supposed to be GCN based.
Anyone know if we can convert MB/s of SHA256 to MH/s?
Anyone know if we can convert MB/s of SHA256 to MH/s?
The A10-4600M has a Radeon HD 7660G (384 cores @ 686MHz) (I'm assuming since it starts with a 7 its GCN)
A 7750 has 512 cores, and at 900MHz gets 136MH/s.
Doing the math from 512 cores -> 384, and from 900MHz -> 686MHz gives us ~77MH/s.
Assuming a 900MHz overclock (just like Ilikeham's A8-3870), that gives us ~102MH/s.
As far as MHs/W, we don't know. But, those GCN are pretty power effecient, so it's bound to be pretty decent. I'm gonna throw it out there and *guess* at under 20W for ~77Mh/s, and under 30W for ~102MH/s.
A 7750 has 512 cores, and at 900MHz gets 136MH/s.
Doing the math from 512 cores -> 384, and from 900MHz -> 686MHz gives us ~77MH/s.
Assuming a 900MHz overclock (just like Ilikeham's A8-3870), that gives us ~102MH/s.
Do we know enough about this proc to estimate mhash/w?
As far as MHs/W, we don't know. But, those GCN are pretty power effecient, so it's bound to be pretty decent. I'm gonna throw it out there and *guess* at under 20W for ~77Mh/s, and under 30W for ~102MH/s.
Wow, thats actually really beefy for an integrated GPU. you could easily play modern games on that, with some compromises on quality options of course.
The A10-4600M has a Radeon HD 7660G (384 cores @ 686MHz) (I'm assuming since it starts with a 7 its GCN)
A 7750 has 512 cores, and at 900MHz gets 136MH/s.
Doing the math from 512 cores -> 384, and from 900MHz -> 686MHz gives us ~77MH/s.
Assuming a 900MHz overclock (just like Ilikeham's A8-3870), that gives us ~102MH/s.
As far as MHs/W, we don't know. But, those GCN are pretty power effecient, so it's bound to be pretty decent. I'm gonna throw it out there and *guess* at under 20W for ~77Mh/s, and under 30W for ~102MH/s.
A 7750 has 512 cores, and at 900MHz gets 136MH/s.
Doing the math from 512 cores -> 384, and from 900MHz -> 686MHz gives us ~77MH/s.
Assuming a 900MHz overclock (just like Ilikeham's A8-3870), that gives us ~102MH/s.
Do we know enough about this proc to estimate mhash/w?
As far as MHs/W, we don't know. But, those GCN are pretty power effecient, so it's bound to be pretty decent. I'm gonna throw it out there and *guess* at under 20W for ~77Mh/s, and under 30W for ~102MH/s.
Do we know enough about this proc to estimate mhash/w?
As a Llano owner, it does not surprise me one bit that the Trinity act this way and proves it self quiet well.
I look forward to picking up one of these one day. Like the llano, I hope they come down to a very good price once they hit the laptop market.
I look forward to picking up one of these one day. Like the llano, I hope they come down to a very good price once they hit the laptop market.
Still they are fun little chips to play with... think about it, for 100 ish bucks you get a nicely capable processor and light game friendly GPU that gobbles up the HD 4000 for breakfast.
The sad thing is that these CPUs are still competing with mobile Clarksfield/Arrandale where the GPUs are horrid. You won't find HD4000 until you go into mid-to-high level i7-3xxx mobile processors.
Jump to: