Steamroller on time after all?

[dragonstongue]

well HT is done through CPU and yes backwards compatible pin count accordingly, this is why AM3 as an example could use a lot of different cpu and AM3+ could as well, they just shift the speed according to cpu installed, they have an upper and lower limit on speed, so for example, if SR will be using the same pin count as Vishera and Zambezi as well as Phenom II, then no problems, AMD did say they will be back compatible, and more then likely they will not need to release new motherboards with a new chipset that is anything but a little more advanced as in sata port counts and such, very much like the 800 series and 900 series are not so vastly different.

from the sounds of it, yes, it will use same socket for opty just need bios update to support.

 

[Wildace]

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.

this is what im hoping next gen consoles change and we see alot more games that can fully make use of 6-8cores

if we start seeing more games using 6-8cores i might replace my 8120 with a Steamroller instead of doing a full MB and CPU swap going to intel.

really hope the PS4 and Xbox next push devs to multi thread alot more and better.

its pretty bad seeing your 8120 sit there at 24% utilized in 90% of games lol.

i have a long time to wait tho, prob decide what i am going to do sometime in late 2014.

 

[Mackintire]

I expect Steamroller to trade blows (performance wise) with Sandy Bridge.

I also expect Intel to clock higher due to their superior fab process and retain the overall performance crown on lower thread metrics again due to high clock scaling.

What I do expect is some real competition.... which should drive down prices.

I expect Intel to release a 12 core "24 hyperthreaded" (all on die) server CPU that they will have to price competitively against AMD's 16 core steamroller release.

I expect AMD will have chips at 45, 65, 80, 95 and 130 watt intervals

I expect AMD to offer very competitive performance at a given price point...and put Intel in the position that the only way Intel can maintain marketshare overall is if they are willing to lower profit margins.

 

[Tsumi]

The FX-8350 already competes with the 3770k in multithreaded applications, so Steamroller should more than match Haswell. The biggest question is how it will perform in terms of single-threaded to 4-thread applications, and how well it will overclock in comparison to Haswell.

 

[octoberasian]

The FX-8350 already competes with the 3770k in multithreaded applications, so Steamroller should more than match Haswell. The biggest question is how it will perform in terms of single-threaded to 4-thread applications, and how well it will overclock in comparison to Haswell.

What do you think that "30% improvement in OPs performance" translate to in single-threaded applications? Think AMD finally fixed it or improved it?

 

[Tsumi]

What do you think that "30% improvement in OPs performance" translate to in single-threaded applications? Think AMD finally fixed it or improved it?

It will be a significant improvement for sure, as the micro-op cache is one of the major contributing factors in giving Sandy Bridge its IPC boost over Nehalem. Improved branch prediction (an area where AMD has always lagged behind Intel) and better decoders should all translate to higher single threaded performance. The key question is how much higher, and what the resulting clock speeds are going to look like.

 

[runs2far]

What do you think that "30% improvement in OPs performance" translate to in single-threaded applications? Think AMD finally fixed it or improved it?

Oh, please give me a link to this info.

AMD writes a lot about big improvements in decoding, branch prediction, cache misses etc. But I have not seen any statements that claims any definite performance increase over Piledriver.

For all we know Steamroller may be 50% better per clock, but clocked 50% lower due to the added complexity not being able to clock very high.

 

[Tsumi]

Oh, please give me a link to this info.

AMD writes a lot about big improvements in decoding, branch prediction, cache misses etc. But I have not seen any statements that claims any definite performance increase over Piledriver.

For all we know Steamroller may be 50% better per clock, but clocked 50% lower due to the added complexity not being able to clock very high.

2 ghz flagship CPU with IPC equivalent to the i7 920... that's laughable.

There is no way AMD is going down in overall performance. This is not a new architecture like Bulldozer was, this is an improvement of Bulldozer, making it better. It will also be on 28nm, which means lower power consumption, heat, and therefore similar clocks.

I understand your skepticism, but you're not being rational about it.

 

[Pieter3dnow]

Oh, please give me a link to this info.

AMD writes a lot about big improvements in decoding, branch prediction, cache misses etc. But I have not seen any statements that claims any definite performance increase over Piledriver.

For all we know Steamroller may be 50% better per clock, but clocked 50% lower due to the added complexity not being able to clock very high.

It is in the roadmap before everything was changed around. The thing is that it is now heading for 28nm where it previously was still on 32nm.

http://www.xbitlabs.com/news/cpu/di...Steamroller_Third_Gen_Bulldozer_Incoming.html

I'm not to sure what people are on about when they are talking about single thread and the expectation that this will magically be better then anything Intel.

The whole Bulldozer family is made to optimize parallel computing single thread is just an afterthought. There is nothing AMD would do to change this around multi cores/threads are the future.

 

[Tsumi]

Not sure where it was mentioned in this thread that we expect single threaded performance to be higher than Haswell?

 

[R3MF]

I'd like an answer to this as well. It doesn't seem plausible to do while maintaining compatibility with older CPUs.

It has been said that Trinity was PCIe 3.0 capable, but they didn't have time to validate it as such so it shipped as PCIe 2.0.

This may change with Richland, and will certainly change for Kaveri later in the year, but the other half of the equation is the making the motherboard/chipset 'electrically' compatible with PCIe 3.0.

Intel faced much the same dilemma lasy year; it had PCIe 3.0 CPU's (ivy) that could be used in non-PCI-e 3.0 motherboard (z68) resulting in PCIe 2.0 operation, and PCIe 2.0 CPU's (sandy) that could be used in PCIe 3.0 motherboard/chipsets (z77), but to make it more complicated there were also z68 motherboards that were marketed as PCIe 3.0 compliant................ but only if you used them with a PCIe 3.0 compliant CPU!

The simplest way for AMD to bridge this gap is to release FM2+ motherboards with Kaveri.............. which mandate elictrical compliance with the PCIe 3.0 standard, and can be used as such with a PCIe 3.0 compliant CPU (Kaveri, and potentially Richland).

.............. And bridge this gap by releasing Steamroller on the FM2+ socket/platform.

You immediately get access to a very high speed dual channel DDR3 memory system, an onboard PCIe 3.0 bus through the CPU socket, and PCIe connection to single southbridge chip with native SATA3 and USB 3.0.

In return, you no longer have to:
1. Invent a faster hypertransport that doesn't exist, and won't exist, as AMD seems to have lost faith in it as a competitive technology.
2. Support the cost a second high-end platform when corporate strategy has basically given up competing with Intel at the high-end.

What are the problems?
Well, you are limited to 100W CPU's........................... but so what!
Intel is working to a 77W limit in the market that AMD is reducing itself to compete in, and if they can't provide performance parity with an extra 23W of headroom then they have bigger problems than little old Bulldozer.

Why has no-one seen ANYTHING of the 1000 series chipsets? Because it's a technological and support dead-end that AMD is unwilling to sink anymore cash into.

I'd be amazed if 8 core Steamroller's weren't released on FM2+, and there are certainly board manufacturers that are marketing FM2 products way too expensive to be justified for a £90 FM2 APU (Asrock FM2A85X Extreme6 / Asus F2A85-V Pro). Practice much..............

 

[Tsumi]

Kaveri uses Steamroller cores. Steamroller is just the core architecture, AMD has not revealed a code name for 8-core FX Steamroller parts yet.

AMD could do what Intel did, and keep the same sockets, except have a new chipset with mandated PCI-E 3.0 support for any board that uses that chipset. Not to mention many FM2 boards are not SLI/crossfire capable, so they are technically PCI-E 3.0 capable due to the fact that there are no switches on the board. The traces on the PCB aren't the problem for PCI-E 3.0 operation, it's the switches on the motherboard for SLI/crossfire operation. Intel also faced the slight problem of Z68 PCI-E 2.0 boards not supporting SLI/crossfire for IB processors.

There's also the fact that the FM2 socket only provides 16 PCI-E lanes, and a new completely different socket will most likely be required for additional PCI-E lanes. FM2 will remain AMD's low-end platform as they work on getting HSA working, and once HSA is ready, which it should be for Excavator, they'll incorporate it into the FX line with a new socket, and all FX processors and future AMD CPUs will be APUs.

I doubt we will see FM2+ until AMD decides to put DDR4 controllers on their CPUs, but we'll see.

 

[R3MF]

Good point on the steamroller name.

Correction - it was rev 2.0 P61 boards that were branded as PCIe 3.0 compliant, implying that the original rev 1.0 boards weren't.

I don't see 16 lanes of PCIe as an impediment in the markets AMD is going to compete with, i.e. haswell on socket 1150 (rather than IvybridgeE on socket 2011).

Yours is an interesting notion, but I'm betting on FM2+ for Kaveri because:
1. I think kaveri will be validated as PCIe 3.0 compliant given that Trinity was 'capable'
2. I think existing FM2 boards don't support PCIe 3.0 operation for the same reason reason early P61 boards didn't (the chipset was willing, but the electircal layout was not).

Given the above, I cannot see the point in creating an expensive new high-end platform to operate in tandem with FM2+ if they are not intending to compete in the £250+ CPU market.

The highest end Piledriver sells for £150, so a £110 Asus F2A85-V Pro would be a very good match for it. If Steamroller performance moves this up to £240 (i.e. ivy-bridge i7 3770), then its not beyond the bounds of reason that high-end x85 board could match it at £150, and even then it doesn't reach the lofty heights of competing with socket 2011. Intel services the 0-250 pounds sterling bracket with socket 1152, why can't AMD do likewise with FM2?

My money is on a PCIe 3.0 FM2+ with Kaveri, and FM3 with DDR4 and a PCIe 3.0 4x chipset link.

http://www.pcper.com/reviews/Shows-and-Expos/AMD-CES-2013-Temash-Kabini-and-Kaveri-side-Sea-Islands

FM2 and FM2+ will be the primary platform for AMD for quite some time. It was essentially confirmed that Trinity has the ability to run PCI-E 3.0 speeds, but the parts were never certified due to a mix of time, budget, and personnel constraints. FM1 did not support PCI-E 3.0 while FM2 does, hence the change in sockets. No timeline has been given for when PCI-E 3.0 will be implemented on the AMD platform, but we were assured that when it makes sense for them to do so we will see products supporting this technology.

AM3 is a big question internally for AMD. These parts are partitioned off of the server market and they are not interchangeable with FM2 (HT vs. PCI-E connections). There will be no updates for some time other than potentially a speed bump down the road with the current Vishera product line. Kaveri will be the first Steamroller part and it is yet unknown if Steamroller will be making it to AM3 or if FM2 will simply replace it in the enthusiast market.

http://www./forum/hardware-canucks-...ni-temash-amd-s-2013-apu-lineup-examined.html

For those wondering, it looks like AMD's FM2 socket will stick around for the time being with an FM2+ refresh possibly rolling out in time for Kaveri. Since all FM2 APUs natively support PCI-E 3.0 (the ability hasn't been enabled in Trinity and presumably Richland), there's a good possibility that FM2+ APUs could include this standard.

 

[octoberasian]

My money is on a PCIe 3.0 FM2+ with Kaveri, and FM3 with DDR4 and a PCIe 3.0 4x chipset link.

This would make more sense than my theory on the previous page.

If AMD kept Socket AM3+ for Steamroller FX, they will need two things for that to work:

1. NEW HyperTransport specification (HT 4.0?) to handle the extra bandwidth.
2. NEW AMD chipset because the current way uses HT to link it and the CPU together.

Unfortunately, there hasn't been any news of HT 4.0 except rumors of a 40-bit bidirectional link on some places, and the ONLY new AMD chipset around the corner is the 1000-series (1090, 1070, 1050). And that one does not have PCI-e 3.0. However, that's the only way I can see AMD maintaining backwards compatibility with the older 800- and 900-series chipsets, and the upcoming 1000-series. But, you will have two different boards with PCI-e 2.0 (old) and PCI-e 3.0 (new) support.

So, the best logical conclusion is drop HyperTransport like what they did starting with Trinity, add PCI-e on the processor die itself, and move to a new socket entirely (FM3?, FM2+?, AM4?) and start over with an entirely new chipset to support it. A lot of people though are thinking Excavator would be the new socket while Steamroller still supports AM3+ as a stop-gap solution.

 

[R3MF]

A lot of people though are thinking Excavator would be the new socket while Steamroller still supports AM3+ as a stop-gap solution.

So did I about about 9 months ago (edit - nearly a year and a half ago *), but the longer and longer the silence continues on both hypertransport and the already outdated 1000 series chipset, the more I am convinced that Excavator will be on a new socket (FM4), and Steamroller will end up on FM2.

The 1000 series chipset cannot be called high-end, and couldn't for exactly the same amount of time that the 900 series ceased to be considered high-end, and the only thing that AM3+ provides is 125W TDP's vs 100W on FM2.



* http://hardforum.com/showthread.php?t=1658980&highlight=

 

[dragonstongue]

if AMD used HT at its full speeds, it would not be a problem, but generally they only use it at 1/2 the available speed. AMD still very much backs the HT consortium.

AMD wants a unified socket after SR/AM3+ maybe to fix shortcomings, reduce costs etc who knows, but there is a likely chance reading all the last few weeks of news, they will be unified(maybe having an internal cap to allow use of FM2 OR AM4 chips or whatever they call them i.e keyed) they will also have use of GDDR5 memory which would solve the issue Intel-AMD-Nvidia are facing right now for the most part, lack of sufficient bandwidth for a lot of things.

Also Intel will not clock their chips super high, why? Well the smaller and smaller they get(which they lead the industry don't forget) the tighter leakage and temperatures have to be controlled, they build their chips differently then AMD builds thiers, so a high voltage in one would kill the other more or less outright, AMD will be transitioning to the same type of process very soon as Intel, and Intel very shortly will have no choice but to go with a similar process as AMD currently uses, so, Intel clocking up really high to make the difference, nope the space and power constraints are starting to get quite limited.
just an example

AMD is facing this problem by choice of design but are "tuning" it constantly, problem with SOI that AMD currently uses, higher leakage as well as generally higher power consumption to drive the circuits(transistors) the problem with Intels process is very tight control on leakage/temps but if on the upper end of the design(clock speeds, overdriven etc) they also take a beating in the power usage/temperature departments, at least all I have read/heard over the modern generations is this, Intel chose to tightly constrain virtually all aspects of their chip which overall means they are very stable known performance up to X, AMD is a little looser but can handle the extra abuse and are more linear in power/temps then their Intel counterparts.

AMD could just allow the SR chips to access the full speed of HT 3 specs all it takes is the "sence" pin in the chip or whatever to trigger the HT to be used at the higher speed in this case the full 32 width link, and if an older chip is used then just use the smaller 16 width link or whatever, not impossible and they probably will do this so the new chips get to use the full speed available(51GB/s vs current 25GB/s) HT is capable of being set like this, again, this is why you could use a Phenom chip but only have HT run at 1800 but if use a Phenom II chip could get HT at 2000 or better.

Either way, HT can go much faster then they are currently using it(usually for servers) I am sure there could be some benefits for using it as such-better CF scaling etc who knows, but simply put, its not DDR4 that they are overall lacking far to many post about that, its the bus width that is constrained more then the speed it operates at that is, for memory its not the overall speed desktop users lack its capacity, Intel and AMD both use memory in a different way, so they both see benefits/drawbacks differently as well.

 

[R3MF]

if AMD used HT at its full speeds, it would not be a problem, but generally they only use it at 1/2 the available speed. AMD still very much backs the HT consortium.

interesting to hear that there is an 'easy' technical solution available to boost the bandwidth of HT, but I am not yet convinced that AMD have any interest in advancing this technology as a core part of their own consumer platform. it feels like a dead-end regardless of whether it be for technical or corporate strategy reasons.

the silence is deafening.

 

[conscript]

Where's John Fruehe when you need him? /sarcasm