• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

DB 4170!

You mean BD FX-4170.

And I don't know about wanting a 4-core version of what is already not exactly a stonking processor, especially given that it's TDP is the same as the top-end 8-core.
 
Depends on pricing, if it competed with Phenom II 965BE or similar, it may be worth a shot. The four core processors are not quite so power hungry and would not be running flat out most of the time.
 
A 4 core BD at 4.3ghz will probably perform worse than a Phenom X4 980BE and probably cost about the same.


You may be right, I have not investigated it, thats why I thought it should compete in the sub £80-90 to have a reasonable fit.
 
problem is, at a lower price the eight core versions are an interesting prospect if used effectively, but the problem is the less cores the less flexible the processor becomes, so the lower end Bulldozers are a bit pointless. though at a lower price the higher end ones could be interesting, especially to have a mess about with. :)
 
If these chips are actually 4core, and not 2core*2modules they could be interesting good performing chips.

However due to the cache listed on the op's chart It looks like it wont be. A shame.
 
they are identical to the eight core versions, just with two modules disabled, likely due to manufacturing defects. the 4MB L2 is just two modules 2MB L2 x 2. ideally in the coming times AMD want to stop being idiots and stop using the same die for all their processors, I mean they all have four Hyper Transport connects, which is only for the server chip, the desktop one only uses one I believe. so that is just wasted die space, same goes for the four core ones, with all that extra L2 and extra modules sitting doing nothing, just make a two module version?!

the whole design is modular so I don't see what the problem would be, lower die size, better yields, better margins and get rid of the pointless Hyper Transports from the chips except servers, that is just taking up die space as well! :confused:

Edit: though I applaud you for being one of the only person who sees these as 'two module, four thread' chips rather than 'four core', because that is the way it is, regardless of marketing!
 
I may have worded things badly. However I shall try and use pictures to explain my understanding.

Both the below diagrams could represent the 4170 Chip, based on a cut down 8 core version, (possibly due to manufacturing defects as you say)
However the top version would be significantly faster than the bottom.

4 Real independent cores over 4 modules. - result faster
fastj.gif


2 modules * 2 cores. - result slower
slowb.gif


My understanding that one of the issues causing bulldozers limited performance, is when the modules have 2 cores assigned. performance drops significantly. Pulling figures out of my memory (not accurate) 20-30% ?
 
that is the sort of thing I was on about in the main Bulldozer thread, about how 'not' sharing can be about 15 - 20% more efficient, but at the end of the day it is all based on workload, games would make great use of that sort of thing i.e. shutting down one integer core for each module since they tend to use at most four 'hefty' threads.

however in programs that are more multi-core orientated, it could prove inefficient to work that way, assume a 'lighter' thread can be run just fine by half a module, so two of those occupy a module nicely, the advantage in that situation is that you wouldn't have to use the 'chip shared' and slow L3 cache for data on these specific threads, rather the much faster 'module' L2 cache and L1 cache, so in that sort of situation one would take a performance hit rather than see any gains.

to be honest, regardless of many peoples thoughts (which a lot of the time I see is short sighted) to get the most out of Bulldozer one has to dynamically schedule tasks, identify what sort of task it is, what sort of 'action' is best required to execute the task and then how best to use the rest of the chips resources, all in very short space of time and on the fly, an example:

an older game for example consists of two 'hefty' threads and another program is very multi-threaded, as an example for some reason we want to use them at the same time. the best scenario would be two modules reduce resources by 33% (i.e. shut down one integer core and operate with the other and the full Floating Point Unit) and the other two modules take on the multi-threaded task, one would get the 15 - 20% benefit of not sharing with the two modules (-33% or one integer core, lower power consumption due to one whole integer cluster being shut down) and the latent multi-threaded efficiency of the modular, shared architecture for the program that is making use of it.

gaming scenarios are being ruined by that sort of inefficient tasking, the processor needs to be instructed to run the game threads in the manner described above, to see maximum performance, it is still power efficient to and extent because 33% of each module will be dormant and not consuming any power. Bulldozer isn't a bad chip, its just being used in a manner that doesn't take advantage of strengths a lot of the time, use it dynamically and it'll be no worse (better than most of the time) than Phenom II, use it like it currently is and its worse most of the time than Phenom II. ;)

Edit: person above, are you implying that the 4.3GHZ speed is an act of desperation to improve performance? the one major flaw with that is the fact Bulldozer is intended to run at those sorts of speeds.
 
You mean BD FX-4170.

And I don't know about wanting a 4-core version of what is already not exactly a stonking processor, especially given that it's TDP is the same as the top-end 8-core.

People really need to get to grips with WHAT tdp is. Its the max power the chip CAN use, not DOES use.

I overclock a 2500k to 4.4Ghz, I leave the power settings in bios as stock, thats 95W normal 125W short term(this is for a few seconds). I run a benchmark and the chip changes frequency constantly and I get X in a benchmark. I go back and change the bios setting to 125W normal and 150W short term power, and still at 4.4Ghz, the clock speed stays at its highest far more often and the score would end up being X+20-25%.

TDP, power, these days it is not a case of, buy a chip with 125W tdp, it will always use 125W.

A octo core 125W chip and a quad core 125W chip will run VERY differently.

A 8150 will only hit its top speed occasionally, like the 2500k, and constantly clock down as it hits the limit and throttles itself, a quad core has the same LIMIT, not default power usage, it means when clocked up it will stay at the top clock speed far longer. In this case, a 8150 maxed out on all cores will use very close to 125W at default clock speed, turbo will barely kick in while a quad core will stay at 4.3Ghz the whole time.

This was one of the issues I had with reviews, they very rarely went in to what, say the average clock speed was throughout any benchmark. They listed would say turbo 4.2Ghz, default clock 3.6Ghz, but what if its running 3.6Ghz for 99% of that benchmark? The process is doing ok but not great, in 6 months the SAME chip being made, might run the same benchmark, have the same TDP, the same default clock and turbo, but it could run 4.2Ghz for the entire benchmark run and be 20-25% faster.

A 4 core BD at 4.3ghz will probably perform worse than a Phenom X4 980BE and probably cost about the same.

It WILL be faster.
 
that is the sort of thing I was on about in the main Bulldozer thread, about how 'not' sharing can be about 15 - 20% more efficient, but at the end of the day it is all based on workload, games would make great use of that sort of thing i.e. shutting down one integer core for each module since they tend to use at most four 'hefty' threads.

however in programs that are more multi-core orientated, it could prove inefficient to work that way, assume a 'lighter' thread can be run just fine by half a module, so two of those occupy a module nicely, the advantage in that situation is that you wouldn't have to use the 'chip shared' and slow L3 cache for data on these specific threads, rather the much faster 'module' L2 cache and L1 cache, so in that sort of situation one would take a performance hit rather than see any gains.

to be honest, regardless of many peoples thoughts (which a lot of the time I see is short sighted) to get the most out of Bulldozer one has to dynamically schedule tasks, identify what sort of task it is, what sort of 'action' is best required to execute the task and then how best to use the rest of the chips resources, all in very short space of time and on the fly, an example:

You seem to be assuming that a new scheduler would ONLY be capable of running it one way or the other, not both.

Think about it this way, what's faster on a i7 2600k, running two threads on the same core, or two separate threads on two different cores? It's quite ridiculous people think this is somehow "new", for years schedulers have had to handle whats best in multiple different scenario's, just that its quite easy on a "normal" chip and a new chip needs a new scheduler. This is the problem, a new scheduler will be able to both identify when two threads should be run on separate modules AND when two threads should be run on the SAME module, and when a system is set for power saving it should put them in the same module anyway and when set to highest performance WHEN IT WILL HELP, put the threads on separate cores. The scheduler doesn't do this now, because its not aware, Windows 8 scheduler will be.

They are still gonna sell nicely to people that don't know better and think the MOAR COARS and higher GHHHHHHHzzzz means better and faster CPU :D

It's obvious who their targeted customers are, and it's curtainly not us :p


I really wouldn't mind having an explanation for you in how a 4.3Ghz quad core won't be faster than a quad core thats, errm, got a lower clock speed, and how that is bad, and how only certain types of people will buy it because they won't understand that more Ghz isn't faster?
 
the idea I was trying to get across was for a new scheduler to do both, at the moment it doesn't and at the moment that gives the chip quite a sizeable penalty in some situations, especially those that aren't heavily multi-threaded. the biggest issue is we need software to support the hardware, software that runs Bulldozer in an efficient and dynamic way. :)
 
You mean BD FX-4170.

And I don't know about wanting a 4-core version of what is already not exactly a stonking processor, especially given that it's TDP is the same as the top-end 8-core.

Uhm its got a TDP of 95W? What wrong with that?

To me that seems it will be a decent chip, I wouldn't be surprised if they hit 5.5ghz+ as they will be built on mature silicon and the design facilitates those kinds of clocks.
 
Back
Top Bottom