Wednesday, December 14th 2011

AMD Gives Bulldozer 6-core a Speed-Bump with FX-6200

AMD launched its AMD FX processor family with two eight-core parts (FX-8150, FX-8120), a six-core part (FX-6100), and a quad-core one (FX-4100), apparently a newer, slightly faster six-core FX processor is just around the corner, the FX-6200. Since all AMD FX processors are unlocked out of the box, the FX-6200 is essentially a speed-bump. Out of the box, it is clocked at 3.80 GHz, with 4.10 GHz maximum TurboCore speed. It features six cores, 6 MB total L2 cache, and 8 MB total L3 cache. Its TDP is rated at 125W. In a presentation to retailers sourced by DonanimHaber, AMD pitched the FX-6200 to have about 10% higher performance at Mainconcept HD to Flash conversion, than the FX-6100 (3.30 GHz nominal, 3.90 GHz max. turbo).
Source: DonanimHaber
Add your own comment

79 Comments on AMD Gives Bulldozer 6-core a Speed-Bump with FX-6200

#26
laszlo
can we call this chip "sexdozer" ?
Posted on Reply
#27
Assimilator
I wonder if this is a harvested FX-8xxx, would explain the TDP increase over the 6100.
Posted on Reply
#28
Fx
CDdude55
A 500MHz increase over the FX 6100 and 4GHz+ turbo is great, but still not significant enough for me to consider going BD just yet. A lot needs to improve.
I agree. I got the AM3+ mobo ready but I want more refinements before I pull the trigger
Posted on Reply
#29
pantherx12
Assimilator
I wonder if this is a harvested FX-8xxx, would explain the TDP increase over the 6100.
The TDP increase would be from the overclocking of the exact same CPU : ]

Could be wrong, but the whole point of bulldozer is that is is a modular design so there is no more need to have disabled cores, they can simply cut them away.
Posted on Reply
#30
blibba
pantherx12
Well depends how in to over-clocking you are.

Like I said by IPC matches a phenom @ 4ghz ( cept in some older software)

If I ran 1.45 volts through this chip I could probably hit 5ghz a good phenom can maybe get to 4.5 so again single core performance ends up the same but with 2 extra cores.
I fear you're missing the boat here to some degree. The thing is that even if you equalised per-core performance between Phenom and FX - and the FX would be drawing a lot more power, even power per core, to do so - the Phenom would still outperform it many multithreaded applications, because of the architectural bottlenecks of the FX.

pantherx12
But if you name what software is effected I can try and see if it really doesn't get an extra performance from those extra cores.
I've not had the chance to experiment myself, but one example of a piece of software that seems not to be held back by the architecture is this:



Notice the near excellent scaling vs. the (faster clock for clock) Phenom II.

pantherx12
I think people forget that two extra cores( over a phenomx6) doesn't necessarily mean 33% extra performance.

Like going from single to dual didn't give us the 100% boost people would of expected.

It seems the hype killed these chips more then anything else.
It's not just that two extra cores failed to mean 33% extra performance, it's that it actually meant less performance in many areas. This, as I've repeated above, is because Bulldozer "cores" are not cores in the same sense as Phenom cores. You could just as easily call it a quad core in which each core has two of some things.

The hype definitely was a big mistake by AMD - this should have been promoted as a budget multitasking chip. I think what really killed it, though, was the single-threaded performance and single-threaded performance per watt. Too much of what we do is still dependent on this.

There's also an issue (fixed in Windows 8 iirc) with how Windows assigns tasks to cores. If you have a dual-threaded task, for example, Windows may well send it to cores 0 and 1 of a Bulldozer CPU - because it sees it as a regular 8 core. But depending on the application, performance could sometimes be almost doubled by sending that task to cores on different modules, such as 0 and 2. Tests in Windows 8 show BD close the gap on but not catch up with SB.
Posted on Reply
#31
Casecutter
blibba
Yes, Bulldozer has 8 "cores", but it shares a lot of resources between them. So, in workloads reliant on those shared resources, it'll perform like a quad.
That's where the Windows 8 scheduler would add benefits, it suppose to break those packets and provide them to the modules in improved sequencing, the module has only to point the packet to the next "core" that will be ready.
Posted on Reply
#32
Super XP
Zubasa
Might as well grab the 4100, those overclock very well and is quite cheap.
I am half tempted to grab one to play around with it if there they release a FX-4200.
Don't quote me but within my Crosshair V bios there is a setting where you can enable "Core-Unlocking". The thing is can you unlock extra cores within the FX-4100? Some already claim they've unlocked cores with the ASUS Sabertooth mobo that has almost the same bios as the Crosshair V.

Anyhow nice speed bump from FX-6100 to FX-6200. Though it does sound odd that they didn't name it the FX-6120/6150. I believe the FX-**70 and FX-**90 are reserved for clock speeds higher than 4GHz such as the upcoming FX-4170 @ 4.20GHz, FX-8170 @ 4.00GHz and FX-8190 @ 4.60 GHz at stock speeds.
Posted on Reply
#33
blibba
Casecutter
That's where the Windows 8 scheduler would add benefits, it suppose to break those packets and provide them to the modules in improved sequencing, the module has only to point the packet to the next "core" that will be ready.
You replied while I was typing, see above :P

Even with this problem resolved, SB still leaves BD in the dust.

Posted on Reply
#34
Super XP
blibba
You replied while I was typing, see above :P
Even with this problem resolved, SB still leaves BD in the dust.
http://media.bestofmicro.com/N/G/310588/original/win%208%20wow%202560.png
blibba
There's also an issue (fixed in Windows 8 iirc) with how Windows assigns tasks to cores. If you have a dual-threaded task, for example, Windows may well send it to cores 0 and 1 of a Bulldozer CPU - because it sees it as a regular 8 core. But depending on the application, performance could sometimes be almost doubled by sending that task to cores on different modules, such as 0 and 2. Tests in Windows 8 show BD close the gap on but not catch up with SB.
It's because todays Bulldozer requires a lot more refining and tuning. Hopefully the upcoming Piledriver cores also called Enhanced Bulldozer resolved these issues. For today's Bulldozer AMD will have to jack up the clock speed to compensate for now. But great points nevertheless.
Posted on Reply
#36
Casecutter
blibba
Even with this problem resolved, SB still leaves BD in the dust.
Well, that's about a 12% increase just from Win8 while not a big thing it Free! :toast:

Then another 20% get to SB... but yes they’re playing catch-up to IB. Although, I don't necessarily subscribe to the idea they have to beat Intel in every B-M to be taken seriously, as long as the CPU/Mobo are priced right and available they' stay in the game.
Posted on Reply
#38
blibba
Zubasa
Funny that you use a 7 year old game that only have 2 main threads on an 8-core CPU build for multitasking ;)
Colossal missing of the point in evidence.

I used that benchmark PRECISELY FOR THAT REASON. Using an 8-threaded application clearly would not demonstrate this effect at all, but that's not all that significant as so few applications are optimised for 8 cores. The 10% performance boost is a best case scenario. It's explained in my post above and in the TH article, I cannot be bothered to go over it again for your benefit.
Posted on Reply
#39
Dent1
blibba
You replied while I was typing, see above :P

Even with this problem resolved, SB still leaves BD in the dust.

http://media.bestofmicro.com/N/G/310588/original/win%208%20wow%202560.png
Yes but how many gamers whom are serious about high end hardware play World of Warcraft - Not a lot! Warcraft users are generally laptop users or have mainstream hardware i.e. integrated GPU and a Atom CPU. So this benchmark is moot really.

If you want to turn this into a benchmark contest, we can post images of multi threaded games and applications leaving "SB still leaves BD in the dust". - But that would be immature. You agree :)
Posted on Reply
#40
repman244
Zubasa
What you need to consider is that everything 96W and up must be marked with the 125W TDP envelope.
So we really don't know how much more power does it use with a ~15% clock increase.
So until a full review is up we won't know if there are any improvements.
You are correct I totally forgot about that.

What we need from Bulldozer or better yet, from Piledriver is at least 5% higher IPC and lower power consumption...for a start. If Piledriver doesn't deliver AMD will fall even further behind Intel, since IB is suposed to have ~8% higher IPC than SB.
Posted on Reply
#41
blibba
Dent1
Yes but how many gamers whom are serious about high end hardware play World of Warcraft - Not a lot! Warcraft users are generally laptop users or have mainstream hardware i.e. integrated GPU and a Atom CPU. So this benchmark is moot really.
Fail.

The point is that (insert any 2-6 core optimised workload here) can perform noticeably better on BD with Windows 8 than with Windows 7, because of how Windows 7 sees it as a "full" 8 core.

If you look at the TH article, or go google "Windows 8 Bulldozer benchmark" (without the quotation marks of course) you can see plenty of other applications showing similar effects.
Posted on Reply
#42
Dent1
blibba
Fail.

The point is that (insert any 2-6 core optimised workload here) can perform noticeably better on BD with Windows 8 than with Windows 7, because of how Windows 7 sees it as a "full" 8 core.

If you look at the TH article, or go google "Windows 8 Bulldozer benchmark" (without the quotation marks of course) you can see plenty of other applications showing similar effects.
Maybe so, not getting into Windows 7 vs Windows 8. I'm not talking about that.

I think you are reaching, I fail to see what Bulldozer's performance has to do with World of War Craft Cataclysm, and why WWCC is even cared about in the enthusiast community.
Posted on Reply
#43
suraswami
FX-6200 Rename to Pentium IV FX Extreme 6200?

Seems like Chief design Engineer from P IV got fired and got hired to design the FX line :laugh:
Posted on Reply
#44
xenocide
Dent1
I think you are reaching, I fail to see what Bulldozer's performance has to do with World of War Craft Cataclysm, and why WWCC is even cared about in the enthusiast community.
Cataclysm is a very CPU-bound game. The difference between a Athlon X2 and a Phenom II X4 is very noticeable as such, so people use it to compare CPU performance since you can get valid observable results. From my own personal experience, going from an overclocked Q6600 to a stock i5-2500k, with the same GPU resulted in double the frame rate.
Posted on Reply
#45
BeepBeep2
pantherx12
There's only a 10% ipc difference betwen phenom and fx.

So my fx 8120 @ 4.4 is like a phenom x 8 ( theoretically) @ 4ghz.

That might not be true across the board but certainly in all the apps I use.

Super pi is a lot slower though :laugh: ( super pi is ancient code though)

If you have the cooling a BD chip will got a lot higher than a thurban core though.


I had a 1055t before this by the way.
No, the IPC difference is 20-30% depending on what you app are running. The good thing is that you can clock it 10-15% higher and MT apps improve overall over the old architecture.
Your 8120 @ 4.4 is like a Phenom II X6 @ 4 Ghz when apps use all of its cores.
Posted on Reply
#46
antuk15
Speed bump and yet would still get hammered by a low-end SB powered CPU..
Posted on Reply
#47
blibba
Dent1
Maybe so, not getting into Windows 7 vs Windows 8. I'm not talking about that.
No, I'm not interested in a Windows 8 vs. Windows 7 debate either. I'm just using the example to explain and demonstrate one of Bulldozer's performance issues.

Dent1
I think you are reaching, I fail to see what Bulldozer's performance has to do with World of War Craft Cataclysm, and why WWCC is even cared about in the enthusiast community.
You don't think I am reaching? I don't understand.

It's not that Bulldozer's performance is somehow related to WoW:C. I don't see how that point is relevant or interesting, though.

I certainly don't care about WoW:C, but, as explained above, it's an excellent way of comparing and analysing how CPUs deal with applications optimised for two cores (which is a LOT of applications and games right now).
Posted on Reply
#48
seronx
blibba
The point is that (insert any 2-6 core optimised workload here) can perform noticeably better on BD with Windows 8 than with Windows 7, because of how Windows 7 sees it as a "full" 8 core.
Bulldozer will run the same as Windows 7 as it will in Windows 8

The flaw of Tom's Hardware guide is the implications not noted

Windows 7:
Both Cores in a module being used
Core A1 <-- 30-50 ns --> Core A2

Windows 8:
Two different cores in two differnt modules being used
Core A2 <-- 100-200 ns --> Core B2

The only reason Windows 8 is showing an increase is because World of Warcraft is optimized for Intel Architectures where the decoders are an odd number
Intel Sandy Bridge, Nehalem can have 5 macro-op decodes(3 simple, 1 complex)

While Bulldozer 1 module can decode 8 macro-ops, 4 per core... meaning for Bulldozer to have relatively the same performance it will need a uops cache and 6 decoders to have perfect alignment with code in World of Warcraft...because you will have a bleed of 2 macro-ops with World of Warcraft unless Blizzard recodes the game for AMD FX

What is Interesting is the benchmark that is on the front page is single threaded

4.1 GHz / ~17 = .2412 x 14 = 3.4GHz

i5 2400 = 3.4GHz single core turbo
FX-6200 = 4.1GHz single core turbo
meaning that Bulldozer has a 16-17 stage pipeline compared to Sandy Bridges 14 stage pipeline
Posted on Reply
#49
blibba
seronx
Bulldozer will run the same as Windows 7 as it will in Windows 8...

The only reason Windows 8 is showing an increase is because World of Warcraft is optimized for Intel Architectures ...
Your reasoning is flawed. World of Warcraft does not suddenly become optimised for AMD FX when you run it in Windows 8. The game is the same. As such, the game's being optimised for Intel does not explain the performance gap between Windows 7 with BD and Windows 8 with BD. Furthermore, you do nothing to refute the generally accepted explanation for this phenomenon. That said, if you can provide a valid argument, or better still a credible source boasting one, I'd be very interested.

Until then, I think we're better off with the generally accepted explanation, which is that in treating BD CPUs as "normal" 8-cores, Windows 7 isn't (yet) understanding the Bulldozer architecture properly.
Posted on Reply
#50
seronx
blibba
Your reasoning is flawed. World of Warcraft does not suddenly become optimised for AMD FX when you run it in Windows 8. The game is the same. As such, the game's being optimised for Intel does not explain the performance gap between Windows 7 with BD and Windows 8 with BD. Furthermore, you do nothing to refute the generally accepted explanation for this phenomenon. That said, if you can provide a valid argument, or better still a credible source boasting one, I'd be very interested.
Your ideology is flawed

Two modules has a peak throughput of sixteen macro-ops much more than the eight macro-ops of one module

When two modules are used you have a higher throughput thus seemingly higher FPS but you then get blockade by the slowest cache

blibba
Until then, I think we're better off with the generally accepted explanation, which is that in treating BD CPUs as "normal" 8-cores, Windows 7 isn't (yet) understanding the Bulldozer architecture properly.
Windows 7 sees Bulldozer correctly it has 8 normal cores

Windows 8 will just fix the problem of legacy programs show when a new architecture is introduced that changes the number of decoders

A normal core is registers to ALU clusters
Bulldozer has 8 individual registers to 8 ALU Clusters thus can be called 8 physical cores
Sandy Bridge has 8 registers that have pairs shared to 4 ALU Clusters thus can be called 4 physical cores

Other than that you can go by Database licensing which concludes the amount of cores being the amount of logical cores thus Bulldozer is 8 cores and Sandy Bridge is 8 cores
Posted on Reply
Add your own comment