AMD "Vishera" FX-Series CPU Specifications Confirmed

[Super XP]

Calling bullshit on the core starvation, single threaded clock for clock it performs worse than the older generation. Perhaps that is caused by L2 bandwidth, but that can be checked with some math, that is unless they had to introduce artificial latency to keep the core clock up or for other reasons.

http://www.hardocp.com/article/2011/10/11/amd_bulldozer_fx8150_desktop_performance_review/4

The history of AMD's Bulldozer architecture is a painful one. The old management lead by Hector Jesus Ruiz and Dirk Meyer deliberately delayed the arrival of K10 "Barcelona" and got usurped by Intel's Core architecture. The mad rush to the original Bulldozer caused a lot of sacrifices in design and ultimately, lead AMD engineering team to cancel the architecture in 2008. Instead of creating a master-core, AMD's engineers envisioned a multi-step approach to increase the performance of its architecture.

Major Improvements. Steamroller should have been the original Bulldozer. Didn't want to underline the whole thing, so the link is below.

Comparing Steamroller to Bulldozer makes much more sense, since the two architectures are starting to differ in greater detail. First and foremost, AMD finally addressed the core starvation. Originally, Bulldozer had a single Fetch and single Decode unit, which were in charge of feeding both Integer and Float schedulers. It turns out that the size of those units were too small and quite often you'd waste precious cycles with either ALU or FPU pipelines not doing a thing. Steamroller goes back to square one and keeps the Fetch unit as a single entity, but the Decode part is now doubled. Each Decode unit feeds one INT unit (4 pipelines) and the FP Scheduler, which has three dedicated units (two 128-bit FMAC units which can act as a single 256-bit unit when you need 256-bit AVX. For legacy code, the MMX Unit is now a single separate entity (instead of multiple side half-units in Bulldozer design). Also, one of major improvements is the increase in the instruction cache size. Up until Bulldozer, AMD featured the largest L1 cache in the field - both L1 Instruction (I-Cache) and Data (D-Cache) were the same size (64KB). With 128KB of L1 cache, AMD easily compensated for the size deficit in L2 and L3 cache versus Intel Nehalem and Sandy Bridge architectures. Bulldozer sliced down L1 cache to "better than Pentium 4, but still crap", as one of our sources put it bluntly (16KB L1 D-Cache and 64KB L1 I-Cache). Steamroller increases the size of Instruction cache beyond K7/K8/K10/K10.5/BD, but L1 D-Cache won't remain the same either.

According to Mark Papermaster, the improvements should yield up to 30% performance increase, but our sources inside the company beg to differ.

"Steamroller is not Bulldozer Enhanced. F*** no. The layout might look the same but our LEGO blocks are completely different. When all is said and done we should get 45% improvement and this goes to show how the Bulldozer was f***** design. This is all what Bulldozer was supposed to be."

Read more: http://vr-zone.com/articles/amd-pus...erformance-increases/17088.html#ixzz26DXDxHDW

In response to Steevo's post below, I fully agree AMD has major cache issues. For example 8MB of say L2 or L3 would bring a significant performance boost for Intel, but not so much AMD.

 

[Steevo]

And again, single threaded so one scheduler feeding one core has WORSE performance per clock than a Phenom II.

So for the most part throw that scheduler bullshit out the window.

Next up the cache, yeah, it is crap. Increasing frequency to compensate for nothing to do is still a waste of time, Intel proved that with P4.

The ONLY way to get away from more cache is to have near perfect branch prediction, and AMD has missed it by a lot. Then they decided to compensate by using less cache, whoever made that decision I'm sure has been fired.

Last, http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested/6
since you don't get hints.

AMD has a huge issue with latency in bulldozer, either due to the size of the chip, or soe other issues.

 

[AvonX]

I also have to give thumbs up for AMD trying something way out of the box. Talk about having Balls of Steel.

That's not balls of steel, that is stupidity. Lets leave the performance aside because obviously it did not deliver. Windows did not utilize those cores properly or this arc at all.
Going for something completely different to Intel was silly to begin with. At least do it right but they did not do that also. They should have stayed with true cores and not this crap.
I am sure Jim Keller wants to throw this crap out of the window, but he will do what they tell him to do and he has no choice but to stick with it. What bulldozer needs is a complete redesign to fix all the faults of this arc and this will take some time. If you think that the only fault is the "L2" then you are mistaken.

 

[eidairaman1]

That's not balls of steel, that is stupidity. Lets leave the performance aside because obviously it did not deliver. Windows did not utilize those cores properly or this arc at all.
Going for something completely different to Intel was silly to begin with. At least do it right but they did not do that also. They should have stayed with true cores and not this crap.
I am sure Jim Keller wants to throw this crap out of the window, but he will do what they tell him to do and he has no choice but to stick with it. What bulldozer needs is a complete redesign to fix all the faults of this arc and this will take some time. If you think that the only fault is the "L2" then you are mistaken.

I can only guess Steamroller is the next major engineering sample of the Arch, for all we know Steamroller may go back to what Was Great about Athlon 64 FX and Phenom II X2-X6 architecturally.

 

[Covert_Death]

i hope so.... i really think PD is simply BD v2.0 and that PD is the true FX chip in the family

 

[eidairaman1]

i hope so.... i really think PD is simply BD v2.0 and that PD is the true FX chip in the family

reading that flow chart the PD is to have a IPC increase, dont know how much but reviews here should specify it.

If PD doesnt improve anything AMD really needs to kill bulldozer arch off.

 

[Covert_Death]

reading that flow chart the PD is to have a IPC increase, dont know how much but reviews here should specify it.

If PD doesnt improve anything AMD really needs to kill bulldozer arch off.

what i meant with my statement was i expect improvements with PD but marginal ones. with SteamRoller i expect dramatic improvement. i think PD will be a worthy upgrade from PII atleast, which may tempt me to buy it, but if its similar to my PII x4 @ 4.0 Ghz then ill wait for Steamroller

 

[eidairaman1]

what i meant with my statement was i expect improvements with PD but marginal ones. with SteamRoller i expect dramatic improvement. i think PD will be a worthy upgrade from PII atleast, which may tempt me to buy it, but if its similar to my PII x4 @ 4.0 Ghz then ill wait for Steamroller

Hope AMD learned a lesson about designs this time around since their K6, K7 and K8 archs were beating on Intel pretty hard in the day,

 

[AvonX]

reading that flow chart the PD is to have a IPC increase, dont know how much but reviews here should specify it.

If PD doesnt improve anything AMD really needs to kill bulldozer arch off.

+1 for that.
Fighting with something that takes a huge amount of effort and time is useless.
It will only take you back even further.
I am sure Keller has something in mind and lets hope they listen to him before its to late.
But the bad news is that this will come with Excavator and not with Steamroller.
I just hope they let him do his job how he wants it to be.

 

[Super XP]

reading that flow chart the PD is to have a IPC increase, dont know how much but reviews here should specify it.

If PD doesnt improve anything AMD really needs to kill bulldozer arch off.

Internal figures show Piledriver @ 15% gain and Steamroller @ 45% gain over Bulldozer. They won't shoot down this architecture, they already put way too many years and money in it.

Also did you read this:
The history of AMD's Bulldozer architecture is a painful one. The old management lead by Hector Jesus Ruiz and Dirk Meyer deliberately delayed the arrival of K10 "Barcelona" and got usurped by Intel's Core architecture. The mad rush to the original Bulldozer caused a lot of sacrifices in design and ultimately, lead AMD engineering team to cancel the architecture in 2008. Instead of creating a master-core, AMD's engineers envisioned a multi-step approach to increase the performance of its architecture.

+1 for that.
Fighting with something that takes a huge amount of effort and time is useless.
It will only take you back even further.
I am sure Keller has something in mind and lets hope they listen to him before its to late.
But the bad news is that this will come with Excavator and not with Steamroller.
I just hope they let him do his job how he wants it to be.

Steamroller is getting a architecture overhaul. Steamroller looks different. The major design change will be with Excavator such as you stated. If the rumored 45% Steamroller holds any water, I wonder how much Excavator will give us over BD? Perhaps when playing with the math, I can see approx: 75%.

 

[AvonX]

Steamroller is getting a architecture overhaul. Steamroller looks different. The major design change will be with Excavator such as you stated. If the rumored 45% Steamroller holds any water, I wonder how much Excavator will give us over BD? Perhaps when playing with the math, I can see approx: 75%.

By the time we reach to Excavator it will no longer be a Bulldozer.
And lets not create a hype over numbers.
I only hope steamroller can be an upgrade over my 1090T, that's all that i hope for.
And lets hope they are still in the dedicated desktop CPU market.
That is also another thing to consider.

 

[Super XP]

By the time we reach to Excavator it will no longer be a Bulldozer.
And lets not create a hype over numbers.
I only hope steamroller can be an upgrade over my 1090T, that's all that i hope for.
And lets hope they are still in the dedicated desktop CPU market.
That is also another thing to consider.

Agreed, the wost case would be AMD releasing high end APU's to replace the discrete desktop CPU's. But I don't see AMD leaving the discrete anytime soon, there is still a massive market for them.

But give me a Steamroller APU that has the same speed, OC and perform just as well if not better than a discrete desktop and I am all over Socket FM2/3. That is if AMD chooses that pathway. But I hope not, or they would give Intel the discrete market and that my friend is

 

[AvonX]

Agreed, the wost case would be AMD releasing high end APU's to replace the discrete desktop CPU's. But I don't see AMD leaving the discrete anytime soon, there is still a massive market for them.

But give me a Steamroller APU that has the same speed, OC and perform just as well if not better than a discrete desktop and I am all over Socket FM2/3. That is if AMD chooses that pathway. But I hope not, or they would give Intel the discrete market and that my friend is

I hope that is not the case but for data centers is not logical to compete with APU's.
You need a discrete CPU and you need all the power and juice you can get from the CPU.
I am just worried what they in mind for the desktop market.

 

[Super XP]

I hope that is not the case but for data centers is not logical to compete with APU's.
You need a discrete CPU and you need all the power and juice you can get from the CPU.
I am just worried what they in mind for the desktop market.

Well you know as well and I know as well as anybody knows, us crazy gaming enthusiastsare moving to Intel if AMD stops supporting discrete CPU's, which is why I said AMD would have to really be convincing for us to stay with say a high powered APU.

For the sake of the industry, AMD please keep feeding us Discrete high powered CPU's and Graphics Please

 

[AvonX]

Well you know as well and I know as well as anybody knows, us crazy gaming enthusiastsare moving to Intel if AMD stops supporting discrete CPU's, which is why I said AMD would have to really be convincing for us to stay with say a high powered APU.

For the sake of the industry, AMD please keep feeding us Discrete high powered CPU's and Graphics Please

APU's cant hold that much power in them. APU's are good for laptops and mobile devices.
And also the mid desktop market for some lite gaming, but that's about it.

 

[Super XP]

APU's cant hold that much power in them. APU's are good for laptops and mobile devices.
And also the mid desktop market for some lite gaming, but that's about it.

In that case, AMD better keep discrete coming or else it will be there competition which will stear us away.
Also Piledriver will have to be at least 15% to 20% faster than the fastest Phenom II x6 for me to upgrade my FX-8120 which is currently running at 4.40GHz with all 8-Cores.
For me at the time Bulldozer made sense, my current setup blew away my late PII x4 setup .

 

[AvonX]

Also Piledriver will have to be at least 15% to 20% faster than the fastest Phenom II x6

You must be dreaming. Faster in what?
Where did you get those numbers from?

 

[cadaveca]

You must be dreaming. Faster in what?
Where did you get those numbers from?

That's what HE wants?

Interesting to see you posting so much. However, I thought I would mentino that changing ram interfaces to DDR4 or DDR5 would nto benefit AMD at all. The real weakness of Bulldozer is cache perforamnce, and I have been saying this since BEFORE the launch.


Why L2 is so slow, I dunno. Perhaps the problem isn't really the L2, and it's thread maangement that is off...I dunno these things.

BUt what I do know is that cache is where the problem is, for most users needs.

 

[Super XP]

That's what HE wants?

Interesting to see you posting so much. However, I thought I would mentino that changing ram interfaces to DDR4 or DDR5 would nto benefit AMD at all. The real weakness of Bulldozer is cache perforamnce, and I have been saying this since BEFORE the launch.


Why L2 is so slow, I dunno. Perhaps the problem isn't really the L2, and it's thread maangement that is off...I dunno these things.

BUt what I do know is that cache is where the problem is, for most users needs.

Fully agree, AMD always had cache issues. Intel blows them out of the water. If AMD can fix its cache problem, then it would make a big difference on performance whether it be 1MB or 20MB. Right now it's not really benefitting from its cache.

 

[cadaveca]

Right now it's not really benefitting from its cache.

But let me re-iterate, because I want to see if I'm right about this. It's NOT L3.

It's L2 and lower caches that are the issue.

Perhaps L3 needs be a bit more flexible, but it seems to me that everyone is focused on L1 and L3 caches, and I guess it's possible those two are affecting L2.

What I do know for sure is that INtel uses an L3 Ringbus, so that data can be quickly passed form core to core. I don't think AMD has anything similar..they simply use L3 for core-tocore transfers.

I mean, I dunno...this is more of a question then a statement based on how I have seen performance differences in different workloads.

To me, this all is a very interesting subject, becuase there is much here that I do not understand, but want to.

 

[Steevo]

Ringbus, like on a X1xxx series card?

I believe the issues with cache are related to distance to and from, and since they do have a shared scheduler it has to wait for both "cores" to be ready thus creating extra artificial wait states in a lame attempt at keeping both pipelies full without having to dump from dependancies.

 

[AvonX]

That's what HE wants?

Interesting to see you posting so much. However, I thought I would mentino that changing ram interfaces to DDR4 or DDR5 would nto benefit AMD at all. The real weakness of Bulldozer is cache perforamnce, and I have been saying this since BEFORE the launch.


Why L2 is so slow, I dunno. Perhaps the problem isn't really the L2, and it's thread maangement that is off...I dunno these things.

BUt what I do know is that cache is where the problem is, for most users needs.

So if you don't know what is the point talking about L2?
I don't think that is the only problem in the architecture, there is more to it.
You can also see the changes they have done in Steamroller.
If L2 was the only problem they would not bother with all the rest.
Jim Keller is the most qualified to talk about this subject.
Why don't we invite him here at techpowerup?

 

[librin.so.1]

Jim Keller is the most qualified to talk about this subject.
Why don't we invite him here at techpowerup?

That would be epic.
You deserve $$$ for even coming up with such an idea!

Now, how do I/We contact him...

 

[AvonX]

That would be epic.
You deserve $$$ for even coming up with such an idea!

Now, how do I/We contact him...

Even if we knew how to contact him AMD does not even let the guy to talk publicly for some reason.
The guy is a star, i don't get it really.

 

[cadaveca]

If L2 was the only problem they would not bother with all the rest.

The "rest" are not real performance-affecting issues. They are natural optimizations in transistor design, eliminating extra space used by redundant transistors for yeild, or extra space left so if needed, they could tweak the design.


Bulldozer was jsut about getting the design functional, and at that, and for highly-parralellized workloads, Bulldozer excels. Hence them selling nearly every single chip.


Now, when it comes to home users and gmaers, and the workloads they have, L2 cache does seem to be a limiting factor. As I said, I do not understand why that appears to be the bottleneck specifically, but it most definitely IS one.


Also, there is taking their 64-bit memory controllers and maknig a real 128-bit one, like Intel has. the whole "linked/unlinked" memory control option is BIOS..that's what that does...sets teh controller to be used as dual independant 64-bit controllers on the same bus(unlinked), or a singular 128-bit one, formed by using both gether as a single unit.

DDR4 is not required..that change is possible on DDR3, assuming there are not patenting issues that prevent such.

Jim Keller is the most qualified to talk about this subject.
Why don't we invite him here at techpowerup?

I do not know who this guy is. I mean, I'm into hardware...like few others are...I've heard his name before, but it's not something that EVER seemed to be something I should remember.

I could be wrong though...but really, not someone on my radar, at all.

Personally, I'd rather speak to people that no longer work @ AMD in the GPU division, that no longer work there.