AMD FX-8350 - "Piledriver" for AMD Socket AM3+

[Durvelle27]

I think Piledriver will be the last AM3+ CPU support. And the Steamroller may be on a new socket with DDR4. But we talk in mid to late 2013.

No Steamroller will be on AM3+

"AMD has stated that the FX-series will receive a Socket AM3+ version of a Steamroller based CPU in 2014"






Edit:

http://www.theinquirer.net/inquirer/news/2208525/amd-sticks-with-socket-am3-for-steamroller

 

[TheLaughingMan]

I think Piledriver will be the last AM3+ CPU support. And the Steamroller may be on a new socket with DDR4. But we talk in mid to late 2013.

Late 2013 and the only thing I can be sure about for Steamroller is AMD is currently working on it being a 28nm chip.

 

[os2wiz]

Late 2013 and the only thing I can be sure about for Steamroller is AMD is currently working on it being a 28nm chip.

I think the story about 2014 and AM3+ has never been confirmed by an AMD spokesperson.So
I think it is still an open book about many steamroller details like what motherboard features, memory, etc.

 

[tacosRcool]

No Steamroller will be on AM3+

"AMD has stated that the FX-series will receive a Socket AM3+ version of a Steamroller based CPU in 2014"

Edit:

http://www.theinquirer.net/inquirer/news/2208525/amd-sticks-with-socket-am3-for-steamroller

That's a long life for a socket

 

[Norton]

Late 2013 and the only thing I can be sure about for Steamroller is AMD is currently working on it being a 28nm chip.

I wouldn't be surprised if a Piledriver upgrade pops up earlier next year at 28nm- I believe AMD is trying to get their GPU's and CPU's down to the same node (should be more cost effective that way anyway).

My $0.02....

 

[Durvelle27]

That's a long life for a socket

well AMD is know to have long life sockets like AM2/AM2+

 

[EarthDog]

But we are talking at 1866 Mhz and 2133 Mhz which I use. I was told on the ROG web site by an Asus technical staffer that at 2133 even with loosened timings it would be near-impossible to get my HT link speed up to 2600. He was right. Even raising cpu-nb voltage and NB voltage couldn't get a stable run at 2400 or 2600. So it sits at 2200. So memory speed definitely impacts HT speed. I should lower it to 1600 and I bet the HT link speed could be raised to 2600 .

Thats probably true... but was solely talking memory speeds and bandwidth here, not the HT link speed and how higher memory speeds may force you to lower it. I see the correlation and what you are getting at though. There isnt anything like that in the Intel world to mess with. But again standing alone, memory bandwidth is not close to being saturated with dual channel.

 

[cadaveca]

Thats probably true... but was solely talking memory speeds and bandwidth here, not the HT link speed and how higher memory speeds may force you to lower it. I see the correlation and what you are getting at though.

For me, memory bandwidth plays a large role in Multi-GPU performance, and in a very perceptible way. I agree with your sentiment about memory perforamnce in a big way for everything else though.

 

[ChristTheGreat]

Can't see how you arrive at that.
Even if Haswell arrives Q2 2013, that doesn't make Ivy Bridge (or Sandy Bridge for that matter) redundant overnight, so come Q2 2013 the situation is still pretty much the same as it is now.
Steamroller is either going to backwards compatible with existing AM3+ (great if you want to upgrade the board but not the CPU, not so great if it means that a better memory subsystem, an integrated PCH etc. fall by the wayside)

The way I see it, going into Q2-3, 2013, the buyer now will be sitting on the present tech

AMD: Cheaper overall*, ~2-4 more SATA 6GB ports, higher power consumption, no SSD caching, no native USB3.0, no PCI-E 3.0**, (and no SAS if comparing to X79 or selected Z77), ECC memory support, native tri/quad single-GPU support.

Intel: More expensive in comparison to 8320,6 and 4 core AMD (* 3570K + Z75 isn't far away from 8350 + 990X/FX in pricing), fewer native SATA 6GB ports, iGP (handy for troubleshooting or placeholder between "proper" GPU's unless you have a 1440p/1600p screen)

Unless you're in the market for a better IGP I'd venture that an IB system is still going to be pretty competitive in 2013 versus Haswell...there's also no guarantee that AMD's Steamroller timetable incurs no slippage...not as if AMD don't have prior form.

** Thinking ahead, while PCI-E 3.0 might be a non-event ATM, it would still likley be a selling point iin the resell market if you were looking at upgrading, and of course a higher bandwidth intensive card of the next gen allied with a CPU physics game might make the difference more than academic.

AMD had native USB3 before Intel.. my P67 doesn't have native USB 3...

 

[Nirutbs]

always nice for the review thx...
nice chip but it cannot make me from intel....ie...ie

 

[EarthDog]

For me, memory bandwidth plays a large role in Multi-GPU performance, and in a very perceptible way. I agree with your sentiment about memory perforamnce in a big way for everything else though.

I dont run multi GPU setups, so I never experienced that. Thanks for the heads up! I have CrossfireX on the test bench as we speak.. I will see what happens when I cut back from 2666 to 1600Mhz tonight.

That still fits in to the 99% thing though (ok, 97% according to steam, LOL?) as most arent using multiple cards anyway.

 

[cdawall]

Can't see how you arrive at that.
Even if Haswell arrives Q2 2013, that doesn't make Ivy Bridge (or Sandy Bridge for that matter) redundant overnight, so come Q2 2013 the situation is still pretty much the same as it is now.

How about Intel hasn't given more than two generations on the step down chipset in forever? LGA1156 lasted what 3 years (late 09-12)?, 1155 is on schedule for maybe 3 as well. AM3 and later AM3+ have been out a while and all boards 890 series and up support BD and PD. AM3+ will be supported a while longer it is already on AMD's roadmaps to stay. So which is the better buy to you? The one you are out another $150+ for a mobo and $200+ for a CPU or drop in steamroller? I am sticking with AMD for that alone.

Steamroller is either going to backwards compatible with existing AM3+ (great if you want to upgrade the board but not the CPU, not so great if it means that a better memory subsystem, an integrated PCH etc. fall by the wayside)

Memory subsystem meaning what the IMC is integrated into the CPU? Even 790FX can give high ram clocks with subsystem timings changed. PCH? it isn't Intel the North and Southbridge on AMD boards support plenty of PCI-e lanes manufactures just dump standalones onboard. I have a several year old 790FX Gigabyte board with USB3.0, 2 16x slots, 10 SATA ports that supports through Thuban. The 890FX model is the same way with BD and PD support. What is the downside to keeping the old board because it sure isn't a performance or feature issue.

The way I see it, going into Q2-3, 2013, the buyer now will be sitting on the present tech

AM3+ is present tech as is Piledriver. With all of the rumors of every company complaining about multithreading which is a better performing chip for the future? The one with a monolithic die fighting for single core IPC or something that multithreads like a champ?

AMD: Cheaper overall*, ~2-4 more SATA 6GB ports, higher power consumption, no SSD caching, no native USB3.0, no PCI-E 3.0**, (and no SAS if comparing to X79 or selected Z77), ECC memory support, native tri/quad single-GPU support.

Intel: More expensive in comparison to 8320,6 and 4 core AMD (* 3570K + Z75 isn't far away from 8350 + 990X/FX in pricing), fewer native SATA 6GB ports, iGP (handy for troubleshooting or placeholder between "proper" GPU's unless you have a 1440p/1600p screen)

Try and be correct on this one AMD 970 allows for almost all of the same options as a Z77 let alone Z75. For $69.99. So for $289.98 you have a CF/SLi supporting dual 8x motherboard and FX 8350. The cheapest dual 8x board I could find for Intel was a Z77 so that breaks out to $334.98. so 15% more expensive for the same features. Oh and AMD has better performance in multithreading, mildly worse in games if any.

Unless you're in the market for a better IGP I'd venture that an IB system is still going to be pretty competitive in 2013 versus Haswell...there's also no guarantee that AMD's Steamroller timetable incurs no slippage...not as if AMD don't have prior form.

There is no guarantee Intel's timetable will not slip it has happened before and can happen again.

** Thinking ahead, while PCI-E 3.0 might be a non-event ATM, it would still likley be a selling point iin the resell market if you were looking at upgrading, and of course a higher bandwidth intensive card of the next gen allied with a CPU physics game might make the difference more than academic.

Current cards don't use the bandwidth available in PCI-E 2.0 let alone what is available for 3.0. Future cards are not likely to change that, but either way dual 8x slots will hamper the Intel side of things as it lacks the lanes themselves to transfer the data between the cards. Hence why a 4x slot can handle a single card, but when used for SLi or crossfire it sucks.

 

[ensabrenoir]

Credit where credit is due....

My gosh.......its....NOT ...a....delorean 2points amd welcome to the 90's no I mean ....?.....20's dont sound cool

 

[os2wiz]

There is no guarantee Intel's timetable will not slip it has happened before and can happen again.



I am predicting right now Haswell will not come out in the 2nd quarter of 2013 as Intel predicts. That is fluff that Intel has produced to freeze the cpu market in their favor and keep any potential customers from jumping to AMD. Going from .22 nm process to .18 nm is a very big jump. Each move down in process size gets that much more difficult to execute.
I am predicting that Haswell will not be out before the 4th quarter. I give any of you permission to wipe the egg off my face if I am wrong.

 

[EarthDog]

There is no guarantee Intel's timetable will not slip it has happened before and can happen again.



I am predicting right now Haswell will not come out in the 2nd quarter of 2013 as Intel predicts. That is fluff that Intel has produced to freeze the cpu market in their favor and keep any potential customers from jumping to AMD. Going from .22 nm process to .18 nm is a very big jump. Each move down in process size gets that much more difficult to execute.
I am predicting that Haswell will not be out before the 4th quarter. I give any of you permission to wipe the egg off my face if I am wrong.

QFPermanence.

Production for these are supposed to be starting in December. If that holds true and yields are acceptable, you should have a rag handy. They already showed ONE at IDF I thought.

Last I heard, Haswell was still 22nm...I bet you wan to change that statement now since there isnt a die shrink...too late, you were quoted.

That said, 2Q 2013 if they get ramped up in dec/jan, isnt off the charts... especially since they are staying on 22nm process.

 

[os2wiz]

QFPermanence.

Production for these are supposed to be starting in December. If that holds true and yields are acceptable, you should have a rag handy. They already showed ONE at IDF I thought.

Last I heard, Haswell was still 22nm...I bet you wan to change that statement now since there isnt a die shrink...too late, you were quoted.

That said, 2Q 2013 if they get ramped up in dec/jan, isnt off the charts... especially since they are staying on 22nm process.

I read that Haswell did in fact have a die shrink to .18 nm. Can anyone else corroborate the die size for Haswell? My statement of delay was solely predicated on a die shrink. If I am wrong about the die shrink all bets are off.

 

[EarthDog]

If I am wrong about the die shrink all bets are off.

:shadedshu


Intel (leaked)roadmaps say 22nm, Intel SAID 22nm..........just google. Not a lot comes up with 18nm but when you google 22nm.. its all over from reputable sources. Below is one... from an intel inerview

http://www.bit-tech.net/news/hardware/2012/09/12/intel-haswell-core/1

EDIT: Broadwell I believe may be the next shrink.

 

[cdawall]

I read that Haswell did in fact have a die shrink to .18 nm. Can anyone else corroborate the die size for Haswell? My statement of delay was solely predicated on a die shrink. If I am wrong about the die shrink all bets are off.

Wiki still lists 22 nm, the stopping block I see is the swap to LGA 1150. The PCH is shrinking from 65nm to 32nm (finally).

 

[cadaveca]

Man, you guys.






Tick tock, then tick tock,

That's the plan of Intel's process clock.



:shadedshu

 

[EarthDog]

Man, os2wiz.






Tick tock, then tick tock,

That's the plan of Intel's process clock.






:shadedshu

Fixed. I didnt botch up that fact and (eventually/potentially)lose a bet because of it!!! But then again, he did renege...

And that rhymed!

 

[os2wiz]

Fixed. I didnt botch up that fact and (eventually/potentially)lose a bet because of it!!! But then again, he did renege...

And that rhymed!

Come on. that is bit unfair. If my whole argument about thmissing their projection is based on
on a process shrink that is not happening, How did I reneg. I was wrong about the process shrink, yes. Give me a break???

 

[EarthDog]

You made a bet..and pulled it back = renege. It doesnt matter what stipulations/reasons you came up with in your head to have the fortitude to post that comment... that was your REASON for making the bet...you should have your facts straight before opening yourself up to that (Well aware its a simple mistake, just giving you the business!).

You are lucky this isnt a bookie and it WAS a gentleman's bet... Vegas/bookie wouldnt tolerate you attempting to renege becuase you thought Eli Manning was coming back off injury but found out after you made the bet he wasnt. Never make a bet without knowing for sure what you are actually betting on! But I'll send you a rag regardless. Im not wiping off the egg Im chucking at your grill for making a bet based off of misinformation!

(Eli isnt injured, just picked someone you may know being a NYker).

Anyway, you are off the hook, just giving you shhhhhhhhhhhhhhhh.

 

[nt300]

There is no guarantee Intel's timetable will not slip it has happened before and can happen again.

I am predicting right now Haswell will not come out in the 2nd quarter of 2013 as Intel predicts. That is fluff that Intel has produced to freeze the cpu market in their favor and keep any potential customers from jumping to AMD. Going from .22 nm process to .18 nm is a very big jump. Each move down in process size gets that much more difficult to execute.
I am predicting that Haswell will not be out before the 4th quarter. I give any of you permission to wipe the egg off my face if I am wrong.

Steamroller should easily be the CPu to directly compete with Haswell

 

[brandonwh64]

Steamroller should easily be the CPu to directly compete with Haswell

Proof?

 

[cdawall]

Proof?

Proof it wont?