Steamroller on time after all?

says opteron which is server grade, desktop will most likely NOT be seen till next year, so it is possible, however, I would prefer they wait a tad till they optimize the best they can, give devs MS etc time enough to get the most out of it etc.
 
It's nice it's coming out this year than later. The other thing that should be coming out by the end of this year are Kaveri (Steamroller-based) APUs. That would push desktop Steamroller FX series to 2014.

So that means we get Steamroller-based Opterons and APUs in 2013, then FX in 2014.
 
Not trolling, but has AMD ever been on time with new CPUs? And now xbit labs is saying they're going to be ahead of schedule?

Sounds like an April Fool's joke to me.
 
drescherjm said:
It says server processors later this year. Desktop processors next year.
Click to expand...

Yeah i saw that, but the reason i posted was hey, it is back on schedule and also the fact that a few months ago the talk around here was it had been delayed until 2015 or canceled all together.but we all know that when we see the opty that Athlon\phenom\fx is just months
maybe a quarter behind.(check 03)
 
Here's a link to the PDF.

It's an interesting read.

Excavator is still on the roadmap but it's a big blank spot on that map. The one thing that looks interesting is on Page/Slide 14.

By Q4 2013, 20% of their revenue will be from embedded and semi-custom products (i.e.- PS4 APU). In the future, that'll account for 40-50%.

Traditional computing? Well, if the market is any indication, then it seems AMD is making a smart move shifting resources towards a 50-50 mix of traditional and mobile/embedded/semi-custom/server solutions. I would not be surprised that for BOTH Intel and AMD, the mobile, server, embedded, and semi-custom market makes up more than 50% of their revenue in the near and immediate future.
 
CLC?

if you mean like a h80 or something that would be a AIO unless you mean custom liquid cooling in which case, umm no lol.

And it should be 8400 series if anything as it is the next one up and excavator follows.
 
dragonstongue said:
And it should be 8400 series if anything as it is the next one up and excavator follows.
Click to expand...

Yeah, seems logical the next ones will be FX-8450, 8430, 6450, 6430, 4450, 4430. It's an odd naming scheme. Lol. I'll admit that.

Why not just jump to FX 9000, 7000, and 5000? :D
 
Kaveri is the only thing with Steamroller cores coming out this year :) If we are lucky. And still not sure if that would require another mainboard , supposedly there would be graphics ram to get skip some bandwidth problems .

But I'm guessing that if you really want a Steamroller this year a server mainboard to go along with it is required ;) .
 
I like the numbering scheme they are using, makes it easy enough to know the generation, the "cores" count as well as the series.

IE FX 8xxx is core count(well kinds bs marketing spin) FX x4xx would be generation(why they skipped the x2xx I am not quite sure) FX xx5x is the model so all together FX 8450 would be 8 core, 3rd gen, 50 is the performance model of the lineup and has been the flagship for the moment(with70 and 90 unused as far as I know, at least not "launched" for any series so far.

It makes perfect sence to my mind, what does not make sence and will clean up a bit with steamroller is FX4-6-8 will be closer to a "true" core count but yeh skipping a numbering sequence did not seem to help much as 2 chips threw off the numbers/performance figures those being FX 6200 which was just lower wattage then FX 6120 and FX 4200 was actually slower then FX 4170 but had more cache.

Anyways, FX x1xx was Bulldozer First gen(actual architecture) Zambezi was the actual segment for desktop use
FX x3xx is Piledriver which is more or less an optimization of the base Bulldozer or now second gen microarch, also known as Vishera for the desktop use.
FX x4xx would make sence for steamroller or 3rd genHERE for some back info as it "should" also launch in 2014 so that itself also adds up :p
FX x5xx should be excavator launch in 2015 or so now 4th gen and very likely will need new motherboard to support ddr4 maybe just maybe even using GDDR5 embedded instead, as kaveri/steamroller will have the links able to use it and would give Sony enough time to have the tech in their hands to make $ off of it.
 
Wasn't AMD quoted saying that the Steamroller cores, due to their digitally-designed compact arch, will actually clock LOWER than current piledriver cores? So we can't expect huge improvements, especially in desktop/overclocking use...
 
It's fitting that AMD is naming their chips after pieces of slow moving machinery. :D

Actually I'd really like to see them become more competitive again. We'll see what they'll have to offer I guess.
 
Red Squirrel said:
It's fitting that AMD is naming their chips after pieces of slow moving machinery. :D

Actually I'd really like to see them become more competitive again. We'll see what they'll have to offer I guess.
Click to expand...

It's going to be interesting what "30% improvement in ops per cycle" performance translates to in real world performance in Steamroller FX/Opteron CPUs.

That and the very first HSA extensions in Kaveri APUs. I'd like to see how AMD is utilizing the GPU. Accelerated FPU?
 
KazeoHin said:
Wasn't AMD quoted saying that the Steamroller cores, due to their digitally-designed compact arch, will actually clock LOWER than current piledriver cores? So we can't expect huge improvements, especially in desktop/overclocking use...
Click to expand...

That is a feature that's not going to make it into Steamroller, but Excavator. And what it will do is limit max overclocks, so you won't see as much 8 ghz LN2 overclocks. However, you will see higher clocks (or same clocks) at normal operating temperatures due to the lower heat output and lowered power consumption.
 
CaptNumbNutz said:
Will 30% even catch them up to the now 2 year old Sandy Bridge architecture?
Click to expand...

I guess we have to wait and see.

If it matches Sandy Bridge in single-threaded workloads, that's good but way past overdue.

If it matches Ivy Bridge in single-threads, that's a little better but slightly late.

If it matches Haswell released this June, then I'd say they're almost there but that'd be a miracle in and of itself.

Stuff around the internet (yep, let's believe them for hypothetical purposes for now) believe AMD won't be one-to-one parity in both single and multi-threaded performance until Excavator which should be a new architecture.
 
dragonstongue said:
IE FX 8xxx is core count(well kinds bs marketing spin)
Click to expand...

It depends on how you view this people tend to keep looking at Intel and discover that their 4 cores + hyperthreading is "faster" so the 8 cores is BS according to people.They are just not very aware of what hyperthreading is about but the simple math tells you 4 is better then 8.

You can't tell people that it is 4 modules and the cores (all 8 of them) are real because of this fairytale story that Intel has been telling. Lately there is a video on youtube going around explaining how 8 AMD cores outperform 4 HT cores with games when you are using fraps. What is unfair is that most if not all websites are using stuff that is singlethread to the core to test AMD cpu this will show easily that AMD can't keep up. Also it requires a new compiled executable and so on.

The problem is not so much under Linux where you can compile your benchmarks to reflect the changes.
 
From the Tweaktown article:

"X-Bit Labs is reporting that AMD has pushed forward its plans for Steamroller and the new chips will be compatible with current sockets and that the new server microprocessors will support PCI-E 3.0. Desktop chip's featuring Steamroller technology will most likely not be seen until next year though."

How is PCIe 3.0 goes to work on current sockets (or more to the point, on the slow chip interconnect used in current sockets)?
 
Also, from the investor report:

"Single-chip custom processor features:
–
CPU : x86-64 AMD "Jaguar", 8 cores
–
GPU : 1.84 TFLOPS, AMD next-generation Radeon™ based graphics engine"

That rather scotches the notion that the PSU uses separate Jaguar APU's combined with a semi-custom 7790 GPU.
 
octoberasian said:
I guess we have to wait and see.

If it matches Sandy Bridge in single-threaded workloads, that's good but way past overdue.

If it matches Ivy Bridge in single-threads, that's a little better but slightly late.

If it matches Haswell released this June, then I'd say they're almost there but that'd be a miracle in and of itself.

Stuff around the internet (yep, let's believe them for hypothetical purposes for now) believe AMD won't be one-to-one parity in both single and multi-threaded performance until Excavator which should be a new architecture.
Click to expand...

I might be way off base here, but i thought haswell was just an improvement in wattage and ivy bridge was only 4% increase in IPC over sandy bridge? With that said, i would be really surprised if it matched sandybridge in IPC.
 
I might be way off base here, but i thought haswell was just an improvement in wattage and ivy bridge was only 4% increase in IPC over sandy bridge?
Click to expand...

Haswell is not generally an improvement in power usage at least from the TDP. It looks like the other direction. Also the gain from SandyBridge to IvyBridge was more than 4%.
 
There has not been set a hard release date, making it hard to delay the chip.

+ AMD has been ok at launching products since Bulldozer.
 
R3MF said:
From the Tweaktown article:

"X-Bit Labs is reporting that AMD has pushed forward its plans for Steamroller and the new chips will be compatible with current sockets and that the new server microprocessors will support PCI-E 3.0. Desktop chip's featuring Steamroller technology will most likely not be seen until next year though."

How is PCIe 3.0 goes to work on current sockets (or more to the point, on the slow chip interconnect used in current sockets)?
Click to expand...

Currently, the only processors with integrated PCI-e are Trinity and most definitely Richland later on. Kabini and Temash should have integrated PCI-e on their processors as well.

However, they're all PCI-e 2.0. The current 800- and 900-series AMD chipsets support PCI-e 2.0 through them. The same applies with the upcoming 1000-series chipsets. All of those are still connected on a HyperTransport bus to-and-from the AMD FX/Opteron CPU at 51.2 GB/s max. Trinity and most likely future APUs are directly connected to a PCI-e 2.0 bus from CPU to I/O components instead of HyperTransport.

What I think we'll see is this:
  • Steamroller FX (for example) supported on Socket AM3+ and 800-, 900- and 1000-series chipsets but PCI-e 2.0 ONLY and HyperTransport 3.1.
  • Steamroller FX supported on Socket AM3+ will support PCI-e 3.0 ONLY through 2000-series chipsets (hypothetical 2090, 2070, 2050) and HyperTransport 4 (hypothetical) running at double or quadruple the bus speed of HyperTransport 3.
The newer 2000-series boards will probably not see light of day until late 2014, but the 1000-series should be early 2014. My best guess. It's quite odd though AMD hasn't moved the PCI-e controller yet to the processor like they have done on the Trinity APUs.

So, in other words, if you want PCI-e 3.0 on Steamroller, would have to wait for the 2000-series chipsets that support newer HyperTransport speeds. If not, Steamroller FX will back-peddle to HT 3 and PCI-e 2.0 on older chipsets.

Perhaps when Excavator comes around we'll get on-die PCI-e 3.0? HyperTransport is still used by AMD in multi-socket processors and boards for CPU-to-CPU communication.

buttons said:
I might be way off base here, but i thought haswell was just an improvement in wattage and ivy bridge was only 4% increase in IPC over sandy bridge? With that said, i would be really surprised if it matched sandybridge in IPC.
Click to expand...

drescherjm said:
Haswell is not generally an improvement in power usage at least from the TDP. It looks like the other direction. Also the gain from SandyBridge to IvyBridge was more than 4%.
Click to expand...

If the preliminary benchmarks are any indication, Haswell not only improves power consumption and regulation, but has a slight increase in IPC and performance as well. But, it again depends on the application being used and will likely be a repeat of what happened between SB to IVB processors.
 
the whole thing with what AMD calls a "core" and what Intel shows in performance are really not central to my opinion. Intel does not call its Quad with HT an 8 core last I checked, though for some odd reasons AMD decided to muck everything up with wording, explanations in essence do a quad with HT(from a design side) and call it an 8 core, this I have a problem with, it is a BS marketing spin, it is more then a "standard" core, but also not near a "standard" dual per core sort of speak.

You cannot call anything but peanuts, peanuts after all, and this unfortunate side effect means that individuals expected performance levels at least as good if not better then previous gens and well in a lot of ways they simply were not, most certainly not from a power usage perspective, as well this most certainly was lead astray by them in the first place by those of us following their development, let alone consumers that did not..Could not imagine replacing an older chip with a modern one to find out that it really is not better for the most part, this is probably why they can price them lower :p

PCI-e 3.0 means more or less nothing to what we are currently using, well unless you are running wicked speed SSD on it, but for graphics cards benchmarks and stuff show it means next to nothing currently vs PCI-e 2.0
 
dragonstongue said:
the whole thing with what AMD calls a "core" and what Intel shows in performance are really not central to my opinion. Intel does not call its Quad with HT an 8 core last I checked, though for some odd reasons AMD decided to muck everything up with wording, explanations in essence do a quad with HT(from a design side) and call it an 8 core, this I have a problem with, it is a BS marketing spin, it is more then a "standard" core, but also not near a "standard" dual per core sort of speak.

You cannot call anything but peanuts, peanuts after all, and this unfortunate side effect means that individuals expected performance levels at least as good if not better then previous gens and well in a lot of ways they simply were not, most certainly not from a power usage perspective, as well this most certainly was lead astray by them in the first place by those of us following their development, let alone consumers that did not..Could not imagine replacing an older chip with a modern one to find out that it really is not better for the most part, this is probably why they can price them lower :p

PCI-e 3.0 means more or less nothing to what we are currently using, well unless you are running wicked speed SSD on it, but for graphics cards benchmarks and stuff show it means next to nothing currently vs PCI-e 2.0
Click to expand...

It has 8 physical integer cores, and 4 256-bit FPUs that can also act as 8 128-bit FPUs. That's why it can legitimately be called a 8-core processor.

Intel's hyperthreading is a circuit path designed to shove more instructions into a core when it's not fully utilized, which happens often in coding even if it's at 100%. There's a big difference between Intel's hyperthreading (getting more work out of one core) and AMD's module design (sharing of front end to reduce power consumption, heat, and die size).

The reason why Bulldozer and Piledriver don't scale 2x with integer core performance and 128-bit FPU performance when the 2nd core is used in the module is the front end decoder bottleneck not feeding the cores fast enough. That decoder bottleneck will be removed in Steamroller due to the implementation of two larger decoders per module.
 
Doesn't the 8350 shine in multi-threaded tasks regardless of what they call a core?

And they have made advances. Isn't 6300 performance similar to the 1100T?
 
Doesn't the 8350 shine in multi-threaded tasks regardless of what they call a core?
Click to expand...

Only in applications that make good usage of 6 or more threads when you compare it to i7 CPUs. Or applications that make good usage of 5 or more threads when you compare it to i5s. So the multithreaded advantage is limited at best being that there are few applications in this category.
 
R3MF said:
How is PCIe 3.0 goes to work on current sockets (or more to the point, on the slow chip interconnect used in current sockets)?
Click to expand...

I'd like an answer to this as well. It doesn't seem plausible to do while maintaining compatibility with older CPUs.
 
ok so 8 integer parts but only 4 fpu that "may be able to" act as 8 which is best case scenario, which very rarely transpires or so it seems.

I still do not like the spin they gave it, while I think the design is real neat, it obviously(at least for now) cannot meet up with its expectations. They would have saved themselves alot of ticked off people or even folks like me wanting to stick with AMD but unimpressed with their decision to call something what it is not.

It is a marketing spin in my books, if a standard core has 1 integer part and 1 fpu then it is a core(at least for simple folks like me) if they take 2 and 1 to make a "dual" and it gets bottlenecked cause it does not have enough resources or whatever, then I ask why they even bothered? they save space but seemingly are not saving power and as of yet are not getting the performance desired/required by design.

IDK, I just was not amused considering all the hoopla they gave to their design cause it sounded by far a more interesting choice compared to what Intel is doing, but its apparent it is really hard to thread it as needed to see its real benefit. They are right I supposed, folks do not buy modules, they buy cores, though this certainly is not helping them if its not behaving the same as a standard core would.

But yeh, the programs/apps designed to take advantage of FX multi-threaded design do seem to work very well, multi-gpu also seems to do well enough, though the power advantage did not seem to pan out as much as they wanted, though this will more then likely change for SR and excavator, by optimizing and compressing while reducing latencies and such.

for PCI-e 3.0 does asus or whomever not make a controller for this, as in AMD can make the specs the way they want and just "tack on" the pci-e 3.0 chip to allow it? USB3 and Sata 6 were/are like this.
 
Dan_D said:
I'd like an answer to this as well. It doesn't seem plausible to do while maintaining compatibility with older CPUs.
Click to expand...

dragonstongue said:
for PCI-e 3.0 does asus or whomever not make a controller for this, as in AMD can make the specs the way they want and just "tack on" the pci-e 3.0 chip to allow it? USB3 and Sata 6 were/are like this.
Click to expand...

Well, when we see this board from ASUS:
http://event.asus.com/2011/mb/TUF/product_990FXGEN3R2.htm

... PCI-e 3.0 is added via a third party chip possibly PLX again.

AMD doesn't have native PCI-e 3.0 including their current chipsets.

I gave one theory in my last post above on how they might implement it. But, I have no idea if there will be HyperTransport 4 or if HyperTransport can even back-peddle itself on older chipsets for backwards compatibility, or if it can switch between HT 3 and another version on the fly depending on what chipset is installed on the board.

You have to note that the current setup is this:

[Memory] <-> [AMD FX Processor] <- HyperTransport 3 -> [900 Series Chipset (PCI-e 2.0)] <-> [Southbridge] <-> [Peripherals and I/O]

To get PCI-e 3.0 in there, it'd have to be an external chip like the PLX bridge.

I'd assume to get PCI-e 3.0 in there is either through a new chipset (2000-series?) but the HyperTransport bus has to be upgraded or it won't have enough bandwidth. Or, they add it to the processor itself like in Trinity and get rid of HyperTransport.
 
dragonstongue said:
ok so 8 integer parts but only 4 fpu that "may be able to" act as 8 which is best case scenario, which very rarely transpires or so it seems.

I still do not like the spin they gave it, while I think the design is real neat, it obviously(at least for now) cannot meet up with its expectations. They would have saved themselves alot of ticked off people or even folks like me wanting to stick with AMD but unimpressed with their decision to call something what it is not.

It is a marketing spin in my books, if a standard core has 1 integer part and 1 fpu then it is a core(at least for simple folks like me) if they take 2 and 1 to make a "dual" and it gets bottlenecked cause it does not have enough resources or whatever, then I ask why they even bothered? they save space but seemingly are not saving power and as of yet are not getting the performance desired/required by design.

IDK, I just was not amused considering all the hoopla they gave to their design cause it sounded by far a more interesting choice compared to what Intel is doing, but its apparent it is really hard to thread it as needed to see its real benefit. They are right I supposed, folks do not buy modules, they buy cores, though this certainly is not helping them if its not behaving the same as a standard core would.

But yeh, the programs/apps designed to take advantage of FX multi-threaded design do seem to work very well, multi-gpu also seems to do well enough, though the power advantage did not seem to pan out as much as they wanted, though this will more then likely change for SR and excavator, by optimizing and compressing while reducing latencies and such.

for PCI-e 3.0 does asus or whomever not make a controller for this, as in AMD can make the specs the way they want and just "tack on" the pci-e 3.0 chip to allow it? USB3 and Sata 6 were/are like this.
Click to expand...

It's called:

1. Brand new architecture that does not fit the paradigms of traditional architectures. They have every right to call it an 8-core part.
2. Brand new architecture means that there are flaws to be fixed, first implementations are never perfect, especially on something as complex as a processor.

The Bulldozer backlash occurred NOT because of the marketing as 8-cores, but because of these reasons:

1. Overhyped by marketing.
2. Overpriced compared to the competition (Sandy Bridge) and previous gen (Thuban).
3. Very late, which only contributed to the overhype.
4. Absurdly high power consumption compared to previous gen (Thuban) and competition (Sandy Bridge) for the performance (only comparable to Thuban and significantly less than Sandy Bridge).
5. Overhyped by people on forums as a direct result of the marketing.
 
Tsumi said:
It's called:

1. Brand new architecture that does not fit the paradigms of traditional architectures. They have every right to call it an 8-core part.
2. Brand new architecture means that there are flaws to be fixed, first implementations are never perfect, especially on something as complex as a processor.

The Bulldozer backlash occurred NOT because of the marketing as 8-cores, but because of these reasons:

1. Overhyped by marketing.
2. Overpriced compared to the competition (Sandy Bridge) and previous gen (Thuban).
3. Very late, which only contributed to the overhype.
4. Absurdly high power consumption compared to previous gen (Thuban) and competition (Sandy Bridge) for the performance (only comparable to Thuban and significantly less than Sandy Bridge).
5. Overhyped by people on forums as a direct result of the marketing.
Click to expand...

Good post and that came from a person with an Intel cpu in sig.:D
 
Does anyone know what socket the steamroller opterons will be using? If its G34/C32 will they work in current boards with a bios update?
 
octoberasian said:
Well, when we see this board from ASUS:
http://event.asus.com/2011/mb/TUF/product_990FXGEN3R2.htm

... PCI-e 3.0 is added via a third party chip possibly PLX again.

AMD doesn't have native PCI-e 3.0 including their current chipsets.

I gave one theory in my last post above on how they might implement it. But, I have no idea if there will be HyperTransport 4 or if HyperTransport can even back-peddle itself on older chipsets for backwards compatibility, or if it can switch between HT 3 and another version on the fly depending on what chipset is installed on the board.

You have to note that the current setup is this:

[Memory] <-> [AMD FX Processor] <- HyperTransport 3 -> [900 Series Chipset (PCI-e 2.0)] <-> [Southbridge] <-> [Peripherals and I/O]

To get PCI-e 3.0 in there, it'd have to be an external chip like the PLX bridge.

I'd assume to get PCI-e 3.0 in there is either through a new chipset (2000-series?) but the HyperTransport bus has to be upgraded or it won't have enough bandwidth. Or, they add it to the processor itself like in Trinity and get rid of HyperTransport.
Click to expand...

And going through a PLX chip has it's own set of problems.
 
You must log in or register to reply here.
Back
Top