AMD Readies A10-7870K And A8-7670K APUs

[Opcode]

Source

 

New AMD A-Series microprocessor for socket FM2+ was spotted on a price comparison site. The new APU has a model number A10-7870K, and it was listed at a preorder price of 175 Euro. The specs provided were generic for Kaveri A10-Series products, and they mention 4 CPU cores, 28nm manufacturing process and 4 MB L3 cache. No CPU or GPU clocks were given.

 

The listing on the shopping comparison site confirmed some of the information, that we received earlier this month about two upcoming 7000 series APUs. According to that information, the A10-7870K will have 4 cores and 4 MB of L2 cache. The APU will be clocked at 3.9 GHz, and the "K" suffix suggests that it can be overclocked to achieve higher performance. The A10-7870K will be rated at 95 Watt TDP, and will be built on newer core stepping, identified by letters "JC" in the tray and box part numbers. Godavari processors will be built on the same core stepping.

 

The second "Godavari" APU is a quad-core A8-7670K, and it will operate at 3.6 GHz. The microprocessor will also have 4MB L2 cache, unlocked clock multiplier and 95 Watt TDP:

 

 

Seems like Godavari is clocked a little higher than expected. First teraflop APU with a new stepping wonder if it will provide full HSA support as it would greatly benefit AMD if it did rather than having full HSA support just for their new mobile platform. Leaving full HSA support in the dark for desktops until Zen.

[CoolaxGaming]

Yawn apus.

Why do we need high end ones, they make no sense...

[Deletive]

AMD should be focusing more on different aspect/markets for these APU's, consumers and vendors aren't going to be putting these in their computers because it'll cost more then give gain.
 

[Swndlr]

Yawn apus.

Why do we need high end ones, they make no sense...

if they could make one that is powerful enough for a good price, they would make sense

would be good for easy to run stuff like LoL and csgo and dota and yeh, stuff like that

but yeh rn theres no point in buying one unless its being sold for cheap

[CoolaxGaming]

if they could make one that is powerful enough for a good price, they would make sense

would be good for easy to run stuff like LoL and csgo and dota and yeh, stuff like that

but yeh rn theres no point in buying one unless its being sold for cheap

+1

[Opcode]

Yawn apus.

Why do we need high end ones, they make no sense...

Heterogeneous computing is a new realm of performance and power efficiency.

 

[CoolaxGaming]

Heterogeneous computing is a new realm of performance.

Ahh but you see, an i3+dgpu would be better. Only mid-lowend qpus make sense.

The higher end do beat competing processors in graphics. But a equivalent CPU performance CPU frommintel can be had with a cheap GPU and it will beat the apu

[patrickjp93]

Heterogeneous computing is a new realm of performance.

 

-snip-

To be fair, that is 512 AMD SPs vs. only 160 Intel SPs. We need a comparison against the Broadwell Iris chips when they land. It would certainly show whether or not AMD's HSA actually gains them anything. 384 SPs at 1 GHz vs. 512 at 750MHz is a much more reasonable comparison.

[patrickjp93]

Ahh but you see, an i3+dgpu would be better. Only mid-lowend qpus make sense.

The higher end do beat competing processors in graphics. But a equivalent CPU performance CPU frommintel can be had with a cheap GPU and it will beat the apu

Not entirely true. Leading industry experts expect CPU core counts to max out around 8 for consumer chips and 32 for servers, and all the rest of the space will be taken up by GPU with every revision and process shrink. Effectively 90+% of our "CPU" chips will be GPU before the close of the decade according to some estimations. And if you look at what that scaling means, you'd effectively have a flagship dGPU in your CPU slot.

[Opcode]

To be fair, that is 512 AMD SPs vs. only 160 Intel SPs. We need a comparison against the Broadwell chips when they land. It would certainly show whether or not AMD's HSA actually gains them anything.

According to benchmarks HSA has one over even on OpenCL which is quite interesting. There's a lot working behind the scenes with HSA that's efficiently optimizing what code gets executed where. I think the general idea is not writing a kernel for accelerating one specific task but rather having the compiler optimize anything in the software to get parallelized that it wants for overall performance.

[patrickjp93]

According to benchmarks HSA has one over even on OpenCL which is quite interesting. There's a lot working behind the scenes with HSA that's efficiently optimizing what code gets executed where. I think the general idea is not writing a kernel for accelerating one specific task but rather having the compiler optimize anything in the software to get parallelized that it wants for overall performance.

Except the compiler has no way to account for dynamic workloads of highly variable scope and size. Furthermore again with less than 1/3 the shaders and likely no heterogeneous implementation for Intel such that both the CPU and GPU are doing work simultaneously, I'm skeptical that this cherry-picked benchmark shows any advantage at all for HSA at this time.

[Opcode]

Except the compiler has no way to account for dynamic workloads of highly variable scope and size. Furthermore again with less than 1/3 the shaders and likely no heterogeneous implementation for Intel such that both the CPU and GPU are doing work simultaneously, I'm skeptical that this cherry-picked benchmark shows any advantage at all for HSA at this time.

It can account for such workloads specifically by referencing the source. If it looks like it can benefit from being parallelized then just optimize it to do so. Keep in mind both the CPU and the GPU support the same HSAIL instruction set. So not every HSA workload needs to be piped to the GPU to be completed. This is a strong suit for HSA as it will determine what needs to be ran where at a given time. It seems like the TCU is only there for when workloads do scale up as it would be far more efficient. Where another strong suit comes in is that the TCU can queue itself workloads and doesn't need to wait on the kernel. Instead of your traditional frameworks like OpenCL, OpenMP, OpenACC and others the GPU and CPU work in tandem with each other as if the GPU was just another CPU. Compared to where in other frameworks you have to have the software queue up the workloads and all that. The HSA foundation wants the GPU to be just as native as the CPU when it comes to tasks.

[Speedbird]

Now, what is the GPU in the 7870K, and can it Crossfire with a 260X?

[Opcode]

Now, what is the GPU in the 7870K, and can it Crossfire with a 260X?

Still GCN 1.1 @ 512 SPU. It will likely be clocked higher due to AMD fixing the GPU bug with Kaveri. It's dual graphics capabilities are unlikely to change.

[Guest Strangerbob]

nice replacement for people who dont want anything more powerful than a 750 or 260

 

too bad freesync has a ghosting issue, otherwise this would be fucking perfect for some artifact free casual gaming

[Meladath]

nice replacement for people who dont want anything more powerful than a 750 or 260

 

too bad freesync has a ghosting issue, otherwise this would be fucking perfect for some artifact free casual gaming

Because a casual gamer with an APU will totally know what freesync is AND decide to buy one right?

[Timmy-P]

Because a casual gamer with an APU will totally know what freesync is AND decide to buy one right?

Coming from an APU user, yes. What's your point?

[patrickjp93]

It can account for such workloads specifically by referencing the source. If it looks like it can benefit from being parallelized then just optimize it to do so. Keep in mind both the CPU and the GPU support the same HSAIL instruction set. So not every HSA workload needs to be piped to the GPU to be completed. This is a strong suit for HSA as it will determine what needs to be ran where at a given time. It seems like the TCU is only there for when workloads do scale up as it would be far more efficient. Where another strong suit comes in is that the TCU can queue itself workloads and doesn't need to wait on the kernel. Instead of your traditional frameworks like OpenCL, OpenMP, OpenACC and others the GPU and CPU work in tandem with each other as if the GPU was just another CPU. Compared to where in other frameworks you have to have the software queue up the workloads and all that. The HSA foundation wants the GPU to be just as native as the CPU when it comes to tasks.

The compiler cannot do that. That's a choice which can only be made at runtime. OpenMP allows the same thing. That's why we have offload with nowait.

[Megahurt]

nice replacement for people who dont want anything more powerful than a 750 or 260

 

too bad freesync has a ghosting issue, otherwise this would be fucking perfect for some artifact free casual gaming

Ed from sapphire said the ghosting was a non issue, yes you can see it on a high contrast demo, but not in regular use, he also said some gsync monitors had the same thing with the same demo.

[ivan134]

nice replacement for people who dont want anything more powerful than a 750 or 260

 

too bad freesync has a ghosting issue, otherwise this would be fucking perfect for some artifact free casual gaming

.....I'm pretty sure that anyone who can't afford more than a 750 or 260, can't afford a FreeSync monitor, but hey, don't let that inconvenient fact get in your way. Let's also ignore the fact that AMD doesn't mandate what scalars manufacturers can use. Nah, it's AMD's fault.

[pixeldensity]

Yawn apus.

Why do we need high end ones, they make no sense...

You do realize that intel does apus too? Also why does it matter whether or not it is a apu, if it has good performance.

[CoolaxGaming]

You do realize that intel does apus too? Also why does it matter whether or not it is a apu, if it has good performance.

I know. But Intel's "apu"s are faster in CPU performance, making them a go to for gamers to pair a GPU with

[pixeldensity]

I know. But Intel's "apu"s are faster in CPU performance, making them a go to for gamers to pair a GPU with

and all i am saying is that we should not just automatically assume that the apu is going to preform badly because it is an apu.

[CoolaxGaming]

and all i am saying is that we should not just automatically assume that the apu is going to preform badly because it is an apu.

Well u see, if it is an apu, it is an AMD CPU, which are never good for games...

[Guest Strangerbob]

Well u see, if it is an apu, it is an AMD CPU, which are never good for games...

 

https://www.youtube.com/watch?v=Cv57qDXpEPU

 

Look at this fag playing battlefield4 on an apu...@ medium @ 1080p @ 50fps

 

.... how dare he.

 

Honestly Id say be that a freesync monitor and this is some artifact free, super cheap gaming rig for medium/low for the demanding todays games, and MAX setting for the other 99.5% of PC games on steam.

 

*artifact free once AMD or whoever fixes that ghosting bullshit freesync has

*and the drawback here is = no upgrade path; APUs arent ment for disrete gpus... unless you gona use them @ 4k or higher, where amd cpu cores are just as power as the best intel ones