If AMD released a 30th Anniversary APU, what would the specs and price be?

[cbn]

We've seen great success with the 20th Anniversary Pentium G3258, and I feel the almost non-stop availability of bargain priced overclockable $75 to $120 G3258 /motherboard combos in the US has boosted this popularity even further.

So if AMD released a 30th Anniversary APU what would the specs and the price be?

Here are the specs I would want:

Unlocked Kaveri quad core with 2MB cache and 320 GCN stream processors (The same number of stream processors as the R7 240 video card)

Price: As low as they can make it. It should definitely be priced below the A8-7600. (The way I see things the lower the price, the higher the chance for US retailers to release bargain cpu/motherboard bundles like we see with the Pentium G3258.)

 

[Insert_Nickname]

Unlocked Kaveri quad core with 2MB cache and 320 GCN stream processors (The same number of stream processors as the R7 240 video card)

I agree. I think I mentioned in another thread this product would close a gap in AMDs line-up. Why can't we get a 2M/4T Kaveri with a low-powered IGP? Its kind of annoying that you have to sacrifice either a CPU module or the IGP at the low-end.

I don't think you can get 320SPs. Kaveri seems to use a 128SP "slice" minimum between IGP models, so you'd get 256SP. Still, not a bad idea. I'd even settle for a 128SP IGP, just give me a cheap 2M/4T chip with an IGP... () (oh, and a h265 decoder, but that seems to be taken care of using HSA)

 

[superunknown98]

Hmm, If it could be anything in AMD's portfolio maybe a die shrunk stars core but with some improvements, just to see what it would perform like.

If it has to be derived from an actual product, I would go for a much higher clocked Kabini. Say 3.5Ghz or more, power usage be damned! Mainly because I'm curious as to how that could perform against the current lineup.

 

[Abwx]

I agree. I think I mentioned in another thread this product would close a gap in AMDs line-up. Why can't we get a 2M/4T Kaveri with a low-powered IGP? Its kind of annoying that you have to sacrifice either a CPU module or the IGP at the low-end.

I don't think you can get 320SPs. Kaveri seems to use a 128SP "slice" minimum between IGP models, so you'd get 256SP. Still, not a bad idea. I'd even settle for a 128SP IGP, just give me a cheap 2M/4T chip with an IGP... () (oh, and a h265 decoder, but that seems to be taken care of using HSA)

There s not much to gain by cutting the GPU whose dedicated power comsumption out of the APU TDP is about 32W for the 7850K , that s about 8W/128SPs at the higher frequencies, measurement on my Kabini 2.05 point about 5W for the 600MHz clocked 128SPs, wich correlate well with the higher figure of the 7850K s 720MHz, those numbers suggest that the best is rather reducing the GPU frequency.

 

[Erenhardt]

So ill be first with it:

3M/6T Steamroller cores @4Ghz and 768 GCN cores @900MHz. Be it 140+Watts, what ever... Just stack some HBM on it!
All under $200 would be sweet deal.

 

[cbn]

I don't think you can get 320SPs. Kaveri seems to use a 128SP "slice" minimum between IGP models, so you'd get 256SP. Still, not a bad idea. I'd even settle for a 128SP IGP, just give me a cheap 2M/4T chip with an IGP... () (oh, and a h265 decoder, but that seems to be taken care of using HSA)

Each GCN compute unit has 64 stream processors. (eg, Kaveri A4-7300 has 192 GCN stream processors, so three GCN compute units)

So 320 stream processors would mean the chip would need 5 compute units enabled.

 

[Insert_Nickname]

There s not much to gain by cutting the GPU whose dedicated power comsumption out of the APU TDP is about 32W for the 7850K , that s about 8W/128SPs at the higher frequencies, measurement on my Kabini 2.05 point about 5W for the 600MHz clocked 128SPs, wich correlate well with the higher figure of the 7850K s 720MHz, those numbers suggest that the best is rather reducing the GPU frequency.

From a purely functional point of view, I agree with you. I just have a feeling that there can't be that many dies that qualify for either having a defective CPU module or a completely non-functional IGP. I'm concerned that some of the AMD SKUs are "made up", having either a functional CPU module or functional IGP fused off. That seems a waste of a perfectly good business opportunity to me.

 

[Insert_Nickname]

Each GCN compute unit has 64 stream processors. (eg, Kaveri A4-7300 has 192 GCN stream processors, so three GCN compute units)

So 320 stream processors would mean the chip would need 5 compute units enabled.

Oh, I missed that one. 320SP just seems too close to the 384SP 7600, hence the 256SP "compromise" as it would be similar to the 7400K.

 

[cbn]

If it has to be derived from an actual product, I would go for a much higher clocked Kabini. Say 3.5Ghz or more, power usage be damned! Mainly because I'm curious as to how that could perform against the current lineup.

Although I am skeptical the clocks would go that high, having increased cpu frequency should help reduce some of the cost impact that chip suffers from because it is SOC.

While having the pch integrated is great for mobile it increases the die size and reduces the amount of performance adding silicon that can be used on the desktop. So increasing clock speed as you have mentioned might help return some value.

 

[Abwx]

From a purely functional point of view, I agree with you. I just have a feeling that there can't be that many dies that qualify for either having a defective CPU module or a completely non-functional IGP. I'm concerned that some of the AMD SKUs are "made up", having either a functional CPU module or functional IGP fused off. That seems a waste of a perfectly good business opportunity to me.

Possible for the GPU but less likely for the CPU part, what i find the most annoying is that due to the modules they cant release X3 parts if a single core is faulty, it s all good to have some low end parts but a 2 ratio between the 4C and 2C is quite big and create a price gap between 40-50$ offerings and the 100$ one.

 

[AtenRa]

12 (4+6) Compute units APU with 1000MHz iGPU and Side Port 2GB GDDR-5 on FM2+ motherboard.

 

[Rvenger]

8 core excavator APU at 4.5ghz on 16nm @ 130w with 1000MHz iGPU :sneaky:

 

[cbn]

Possible for the GPU but less likely for the CPU part, what i find the most annoying is that due to the modules they cant release X3 parts if a single core is faulty, it s all good to have some low end parts but a 2 ratio between the 4C and 2C is quite big and create a price gap between 40-50$ offerings and the 100$ one.

Not only that, but all the dual core APU models have also had at least half of the GPU units disabled.

That increased the gap even further.

This makes me wonder how much of their binning process is done out of necessity vs. purposes of artifical product segmentation.

If it is done mostly for purposes of artifical product segmentation that is fine. That is AMD's business, but I would only hope they could find a way to get a little more aggressive.

 

[Abwx]

Not only that, but all the dual core APU models have also had at least half of the GPU units disabled.

That increased the gap even further.

This makes me wonder how much of their binning process is done out of necessity vs. purposes of artifical product segmentation.

If it is done mostly for purposes of artifical product segmentation that is fine. That is AMD's business, but I would only hope they could find a way to get a little more aggressive.

Binning is made such that you have the higher ASP possible out of a given dispersed output caracteristics wise, as such it make no sense to disable GPU parts of a fully functional chip cores wise but the contrary is not true as you point it and it s unlikely that we ll ever see a 2C part with 384 or 512 SPs even if they were actualy functional.

Other than that agressivity require some means, for the time i find their A8 7600 extremely agressive while their FX line sell very well thanks to agressive prices as well.

8 core excavator APU at 4.5ghz on 16nm @ 130w with 1000MHz iGPU :sneaky:

Would be a hit, numbers from 6 months ago point the FXs at 4-5% of the whole PC CPUs/APU market and about 10% of the DTs, i guess that AMD didnt expect such an evolution...

 

[Blue_Max]

Bigger. Just, plain bigger. More video speed, more processor power...

Moooorrrrrrrrreeeeee.... *drool*

 

[Centauri]

Something ARM-based, looking toward the company's future.

 

[cbn]

(snip) but the contrary is not true as you point it and it s unlikely that we ll ever see a 2C part with 384 or 512 SPs even if they were actualy functional.

Yes, as I understand it, the fear that enabling functional GPU units on their dual core models would be that those SKUs could eat into the sales of the higher priced quad core offerings.

But something just doesn't seem right about the AMD's situation regarding the following concept you mentioned:

Binning is made such that you have the higher ASP possible out of a given dispersed output caracteristics wise

I feel as if the AMD needs to lower prices on their top APU SKUs and improve the specs on their lowest end SKUs. This essentially would move the two extremes closer together lessening their spread.

 

[DeathReborn]

I expect it would be a: A10 7870K 3.8GHz/4.1GHz & 800Mhz GPU. Basically a slight OC on a 7850K. Priced just $10 or so above the 7850K.

I would like a AM3+ 6M12T Excavator FX @ 3.5-4.5GHz ~95W TDP @20nm.

 

[jcazes]

I expect it would be a: A10 7870K 3.8GHz/4.1GHz & 800Mhz GPU. Basically a slight OC on a 7850K. Priced just $10 or so above the 7850K.

I would like a AM3+ 6M12T Excavator FX @ 3.5-4.5GHz ~95W TDP @20nm.

I think you're off slightly. It will be $20 more! ;-)

AMD has a way of pricing...

 

[cbn]

Hmm, If it could be anything in AMD's portfolio maybe a die shrunk stars core but with some improvements, just to see what it would perform like.

I have often wondered how high Llano's stars cpu cores would have clocked if they shared the same process tech improvements as Richland. (Both APUS were on 32nm)

IIRC, A8-3870K had a base frequency of 3000 Mhz, but didn't overclock all too well.... Max OC on Air was something like 3.6 Or 3.7 GHz. Meanwhile, the star cores on E0 45nm silicon were going something like 4.0 to 4.2 GHz on air.

 

[burninatortech4]

I'd like a successor (Steamroller or future arch) to the 760k. A perfect budget, gaming overclocker. Whatever amount of GCN units that's realistic (I personally won't use them anyway).

 

[frozentundra123456]

Yea, seems like a full 4 core cpu with just a big enough igpu for basic use might be attractive at the right price. That way, OEMs could put it in cheap pre-builts without having to use a discrete card, and for gaming one could add a stronger dgpu without wasting more than half of the chip area, i.e. the igp. Until (or unless) they solve the bandwidth problem, a good portion of the igpu is just wasted.

 

[Insert_Nickname]

I'd like a successor (Steamroller or future arch) to the 760k. A perfect budget, gaming overclocker. Whatever amount of GCN units that's realistic (I personally won't use them anyway).

What you wish for is already out, its named the 860K...

 

[NTMBK]

A Phenom II shrunk to 16nm.

 

[mrmt]

64C excavator cores @ 5GHz
32CUs @ 1.5 GHz
16GB HBM
New chipset
Custom 2MW water loop