Why Doesn't AMD Release an AM3+ Update for Steamroller?

[flipped_bit]

So according to various sources, the Godavari Kaveri refresh is expected to add ~10% IPC over the last Kaveri generation.  According to cpubenchmark.net, the X4 860K has stock single-threaded performance of ~1600, and that's without L3 cache!  A 15% overclock - very possible for that chip - brings it to ~1800.  If you tack 10% on to that number from IPC improvements with Godavari, you're pushing 2,000, which is competitive with a lot of Haswell i3s and even some of the i5s. 

 

The big thing FM2+ lacks is L3 cache.  So, why doesn't AMD plunk these refreshed cores into an FX refresh with L3 cache?  If they were to do that and keep prices in the same ballpark or even a bit higher, that would be a pretty good deal IMO and more people would go AMD.  It would help tide them over until Zen arrives in 2016/2017, that's for sure.

 

The myth that Bulldozer single-threaded performance is on par with the Phenoms should be put to rest.  My main rig is an Athlon X3 445 unlocked to a Deneb Phenom II X4 B45.  The maximum stable single-threaded performance I can squeeze out of this chip is ~1450 at 3.9GHz.  The existing X4 860K beats that handily at ~1800. 

[Tedster]

Welcome to the Forum.

 

I think that this is just because AM3+ is dead.

[FruitBasketSilex]

So according to various sources, the Godavari Kaveri refresh is expected to add ~10% IPC over the last Kaveri generation. According to cpubenchmark.net, the X4 860K has stock single-threaded performance of ~1600, and that's without L3 cache! A 15% overclock - very possible for that chip - brings it to ~1800. If you tack 10% on to that number from IPC improvements with Godavari, you're pushing 2,000, which is competitive with a lot of Haswell i3s and even some of the i5s.

The big thing FM2+ lacks is L3 cache. So, why doesn't AMD plunk these refreshed cores into an FX refresh with L3 cache? If they were to do that and keep prices in the same ballpark or even a bit higher, that would be a pretty good deal IMO and more people would go AMD. It would help tide them over until Zen arrives in 2016/2017, that's for sure.

The myth that Bulldozer single-threaded performance is on par with the Phenoms should be put to rest. My main rig is an Athlon X3 445 unlocked to a Deneb Phenom II X4 B45. The maximum stable single-threaded performance I can squeeze out of this chip is ~1450 at 3.9GHz. The existing X4 860K beats that handily at ~1800.

The correct number for the supposed godaveri chip would be 2024. You take the original performance, times it by 1.1 to add the IPC improvement, then the overclock, x 1.15.

Giving you 2024.

[Ahaber]

@filpped_bit

 

I don't know that much about nano fabrication and systems architectures, but I think that the ability for the cpu to work correctly on a completely different chip would mean a complete redesign of the nano structure of the silicon, to essentially code in the hardware for the usage of L3 cache and have compatibility with the different chipset on AM3+ boards.

 

P.S. Don't forget to follow your own topics.

[Sakkura]

The myth that Bulldozer single-threaded performance is on par with the Phenoms should be put to rest.  My main rig is an Athlon X3 445 unlocked to a Deneb Phenom II X4 B45.  The maximum stable single-threaded performance I can squeeze out of this chip is ~1450 at 3.9GHz.  The existing X4 860K beats that handily at ~1800. 

 

The 860K is not Bulldozer.

 

Anyway, the newer APUs are built on a different process node that limits clocks. So the gains in IPC are partially offset by lower clocks. And GlobalFoundries does not have a newer process node to build new FX CPUs on, it would have to be the same old 32nm shit if not the 28nm APU process.

 

So since there's little to be gained, I guess AMD decided there was no point investing time and effort in it. Just let the old Piledriver FX CPUs limp along until Zen is ready.

[flipped_bit]

The correct number for the supposed godaveri chip would be 2024. You take the original performance, times it by 1.1 to add the IPC improvement, then the overclock, x 1.15.

Giving you 2024.

 

Yes, I said ~2,000 in the OP.

[flipped_bit]

Welcome to the Forum.

 

I think that this is just because AM3+ is dead.

 

Thanks for the welcome.  I've been watching the videos for months and figured it was time to join the forum.

 

I know it's dead, I'm just curious why they never considered refreshing it given their IPC improvements since Bulldozer.  Realistically I don't expect Zen until 2017 at the earliest (hope I'm wrong), and another two years is a long way to go with just the current AM3+ and FM2+ platforms.  It's odd that single-threaded performance on the "entry-level" platform is now possibly going to be 10-20% faster than the on enthusiast platform.

[flipped_bit]

@filpped_bit

 

I don't know that much about nano fabrication and systems architectures, but I think that the ability for the cpu to work correctly on a completely different chip would mean a complete redesign of the nano structure of the silicon, to essentially code in the hardware for the usage of L3 cache and have compatibility with the different chipset on AM3+ boards.

 

P.S. Don't forget to follow your own topics.

 

Yeah I don't know a lot about it either, but my understanding is that the same cores are used across the AM3+ and FM2+ lines.  I'm sure there would additional R&D needed to add on the L3 cache, but surely not that much since the Steamroller cores are done and the FX Piledriver chips already have L3.  It just seems like low-hanging fruit to combine the two, but I guess I'm missing something.

[sgloux3470]

The money that they will make from it will be less than the money they will have to spend to make it happen.

 

To be honest, while it would make benchmarks look a little bit it's not like it would really affect anyone's purchasing decision.  

[FruitBasketSilex]

Yes, I said ~2,000 in the OP.

Its better to give exact numbers. And your arithmetic was flawed.

[flipped_bit]

Its better to give exact numbers. And your arithmetic was flawed.

 

Seriously?  I was within 24 points, or 1%, of your number, and you're making a big deal of it? 

 

The reason I say approximately is because each chip, as well all know, overclocks differently.  Some will overclock better, others less so.

 

And the arithmetic wasn't flawed.  I multiplied stock performance by overclock percentage by expected IPC percentage to get that ~2,000 number. 

[QueenDemetria]

Its better to give exact numbers. And your arithmetic was flawed.

For this its better to estimate, since the math is based off of estimations and there are always variances with PC components(like how you can run a benchmark 10 times and get slightly different results each time).

[Glenwing]

Thanks for the welcome.  I've been watching the videos for months and figured it was time to join the forum.

 

I know it's dead, I'm just curious why they never considered refreshing it given their IPC improvements since Bulldozer.  Realistically I don't expect Zen until 2017 at the earliest (hope I'm wrong), and another two years is a long way to go with just the current AM3+ and FM2+ platforms.  It's odd that single-threaded performance on the "entry-level" platform is now possibly going to be 10-20% faster than the on enthusiast platform.

 

To be completely honest, it's probably because AMD has very limited engineering resources. I think they'd rather put more effort into Zen to make it a great architecture, rather than pull engineers away to work on a stop-gap refresh at the cost of making their newer projects suffer, stuff that they really need to make a big splash with.

[FruitBasketSilex]

Seriously? I was within 24 points, or 1%, of your number, and you're making a big deal of it?

The reason I say approximately is because each chip, as well all know, overclocks differently. Some will overclock better, others less so.

And the arithmetic wasn't flawed. I multiplied stock performance by overclock percentage by expected IPC percentage to get that ~2,000 number.

You were off by ~24, because you gave an ~number.

In this case, both would work out to the same, but with slightly larger equations it all would've gone to shit.

Not trying to knitpick or make you mad man, sorry. I've had a pretty rough few days, so sorry if I came off as rude or mean.

[flipped_bit]

You were off by ~24, because you gave an ~number.

In this case, both would work out to the same, but with slightly larger equations it all would've gone to shit.

Not trying to knitpick or make you mad man, sorry. I've had a pretty rough few days, so sorry if I came off as rude or mean.

 

Hey no problem, we've all been there.  No offense taken.

[FruitBasketSilex]

Hey no problem, we've all been there. No offense taken.

I personally Dont think there's gonna be anything until zen though. It'll probably come out late this year or early next year. I posit that AMD will probably just hang on to their current CPU sales and then release a bombshell.