• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD could be preparing high-performance A10-7870K Black Edition and A8-7670K APUs.

Yes their chips with 128MB eDRAM are faster than Intel's other chips, but when they are not noticably faster than AMDs chips that don't have any on-package RAM, it's a bit of a hollow victory - and just proves how bad the rest of the intel onboard graphics still are.

I wouldn't read it that way. Compare the 4770K (HD 4600 graphics = 20 cores, no onboard memory) to the 4770R (Iris Pro 5200 graphics = 40 cores, onboard memory). The former is about 3.25x faster, but we'd expect 2x the speedup just from doubling the core count, so maybe +60% of the performance comes from the cache alone.

That's a pretty big jump, if it worked for the 7850K it would be posting about 40 FPS on Bioshock Infinite@1080p, i.e. 50% faster than the Iris Pro.

In other words, the Iris Pro GPU is way weaker than Kaveri, but the onboard cache makes it competitive.
 
Because you're missing the point.
Performance is a sum of parts.

So if the Iris Pro actually had some GPU performance to go with the eDRAM, it would be worth having


If kaveri wasn't bandwidth limited, the gpu performance would be up there with a 7750, it's no where near now.

I assume if it was easy to bolt some on die ram onto it now then they would - however I imagine they don't because:

1) They are stuck at 28nm and so have no on die space
2) The architecture is not designed for it
3) It would increase per unit cost to an unmarketable position
4) It would cannibalize their low-mid range gpu business at present

Intel don't have any of these problems.

The Zen refresh of this and other AMD platforms may be interesting as if they do come with on-die HBM memory, it will be interesting to see what happens regarding point 4 above, and also how they deal with cost in point 3 - e.g. 1GB or more of on-die probably isn't going to happen.
 
If it's got memory on die, then it's problem solved. But if you bolted any current gpu memory, the problem is resolved.

Again. The problem is not the memory, rather the lack of existence.

It's not, do you realise how much bandwidth and/or how big gddr5 memory chips are? 512MB is as big as they come currently and they offer 1/16th of the bandwidth you can find on a 290x, that is 320GBs/16= 20Gb/s bandwidth.

A gddr5 chip is almost the size of an apu already, to actually get a meaningful amount of memory or bandwidth you'd need to add 4 chips for 80GB/s of bandwidth.

When HBM 2 turns up you will be able to get 256GB/s of bandwidth with anything from 2-8Gb in one stack that is about 1/8th the size of a single gddr5 chip.

GDDR5 on package on an APU is entirely, 100% not feasible at all. Cost wise it would drastically increase power usage and package size, it would mean new sockets, expensive sockets, bigger motherboards in the key demographics that are laptops and the like. GDDR5 isn't included on package because there is no sensible way to do it. HBM has been designed(with AMD involved from day one) for this exact reasoning.
 
So if the Iris Pro actually had some GPU performance to go with the eDRAM, it would be worth having




I assume if it was easy to bolt some on die ram onto it now then they would - however I imagine they don't because:

1) They are stuck at 28nm and so have no on die space
2) The architecture is not designed for it
3) It would increase per unit cost to an unmarketable position
4) It would cannibalize their low-mid range gpu business at present

Intel don't have any of these problems.

Intel have generally the same problems. They are on a different process but also don't pack on die memory despite the fact it would increase performance drastically. Their architecture is as designed as AMD's for a 8GB stack of 256GB/s memory... not. You design an architecture with what it will be used with, that simple. They can't fit big high end memory on die either.

Intel's current on die cache solution makes those chips incredibly expensive. It's also very low capacity for the cost involved and why it's not used on every chip they sell and/or has increased the cost of those chips they sell that do have it.

Intel exactly like AMD has come up with a HBM like product precisely for the type of use being discussed here, HMC memory. They don't have it on their dies due to cost, die space and ability to manufacture it yet like AMD.

You also can't particularly cannibalize sale of low end gpus.... by selling a different low end gpu.. same thing either way frankly.

Fact is that the biggest markets with the least shrinking is mobile, where discrete gpus are far far less important to AMD than APU sales in general.
 
APUs for PCs don't seem to be much of an industry focus. Here is a document I happened upon yesterday discussing the various types of new memory and their applications. PC doesn't get much mention aside from high-end graphics.

http://www.yole.fr/MEMORY_ROADMAP.aspx

They all sound good but none are currently suitable or doable for APUs.
 
If we can assume the new A10 will be about £140? G3258 £60 + something used like a 670 or new like an R7 265 @ £80 would be faster, right? Not to mention better upgrade path for the CPU

I think they should stick to sub £90 APUs and call it that
 
Either way the tittle tattle arguments go, people on here with mid range or better desktops that are enthusiasts would not really be buying into this just yet.

They do have a place like someone above mentioned which is to the budget gamer. The next gen APU's will make this space interesting for sure.
 

If wccftech are right i don't get the point of this APU, according to them its exactly the same as the A10-7850K but with an 856Mhz iGPU instead of 720Mhz and +100Mhz on the CPU boost clock.

Same Steamrooler Cores
Same GCN 1.1 iGPU

Pointless.

http://wccftech.com/amd-godavari-refresh-apu-a10-7870k-a10-7670k/
 
^ 856 from 720 is a nice bump in GPU clock speed. With (very) fast main memory it could be a very good APU for gaming. Those benchmarks linked earlier were with 2133 memory and were "playable" at 1080p, a few more FPS on the 7970K would be welcome surely.
 
Pricing is the important factor here. If they price the A10 7870K much over £100 its not going to be that competitive. The A8 SKUs tend to be better value for money anyway.
 
This sounds like the apu equivalents of the fx 8370/8370e/8320e. More of the same but will clock higher with less voltage due to a more mature process.
 
If wccftech are right i don't get the point of this APU, according to them its exactly the same as the A10-7850K but with an 856Mhz iGPU instead of 720Mhz and +100Mhz on the CPU boost clock.

Same Steamrooler Cores
Same GCN 1.1 iGPU

Pointless.

http://wccftech.com/amd-godavari-refresh-apu-a10-7870k-a10-7670k/

Yeah it's just a refresh, re-package, inventory clear out :p It's better than AMD doing nothing for a year though I guess. AMD don't really have anything for desktop until 2016, so at least this generates some advertising for them on there current lines.

If they can bring the price down to £100 like previous APU's (7850K is around £115) then it might have appeal for small form factors, considering the small bump in clocks as well.

Pricing is the important factor here. If they price the A10 7870K much over £100 its not going to be that competitive. The A8 SKUs tend to be better value for money anyway.

This sounds like the apu equivalents of the fx 8370/8370e/8320e. More of the same but will clock higher with less voltage due to a more mature process.

+1
 
Back
Top Bottom