AMD Cutting GPU Prices Again

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
[citation][nom]luc vr[/nom]They should have done this with the Bulldozer and Opterons also.[/citation]

I don't know about the Opterons, but AMD did cut FX prices. They are priced excellently right now, especially for enthusiasts who know how to configure them properly and are willing to do so. People who value highly threaded performance and don't want to spend $600 or so on an i7-3930K are also better off getting at or between the FX-6100 and the FX-8150 because the 61xx/62xx model can fight with the i5s in highly threaded performance and the 81xx models are just a little behind the i7s in highly threaded performance.
 


Have you actually looked into those CPUs? Take an FX-4100, bring the CPU/NB frequency to 3GHz from the 2GHz or 2.2GHz stock, use PSCheck to lower the P state frequencies of the second core of each module, and and heighten them for the first core of each module will let it beat the i3s significantly even without overclocking the CPU frequency. You can do the same with the other FX CPUs and you get some huge improvements. The FX-8120, with this, can give even the LGA 1155 K edition i5s and i7s some excellent competition in OC versus OC performance and at a much lower price.
 

Yes I did. I test tons of different hardware on a daily basis, and garbage like Faildozer ain't an exception at all.


1.
2.

...now scram.
 
[citation][nom]Novulux[/nom]Don't know about other places, but $360-380 HD 7970s (reference, non-GHz) have been available for months at Microcenter. Even before I bought my HD 7870 two months ago for $280~. I really didn't want a reference card, but I know many people won't care.[/citation]
FYI, tax in my state is around $30, so that's well over $400. Neweg, at least in my state, doesn't charge tax 😀
 


I see no reason to "scram". You haven't proven anything other than that you judge a book by the cover and try to offend anyone whom disagrees with you. Testing something and looking into are two quite different things. Unless you tested it with what I said, then any results that you draw are irrelevant except for comparisons at stock.
 
[citation][nom]blazorthon[/nom]Did you ever hear of the CPU/NB frequency and PSCheck? The CPU/NB frequency controls (among other things) the L3 cache frequency and overclocking it significantly has a substantial effect on performance per core per Hz of the CPU frequency. Using PSCheck to cut-down on the second core fo each module lets the fist core mroe or less get the front end all to itself, another significant performance per core per Hz boost at a given CPU clock frequency.FX can almsot match SB and can easily make up for the now minor loss in performance per Hz per core with significantly higher frequencies. The FX-81xx CPUs can easily hang with even the K edition i5s and i7s in OC versus OC, greatly because of the PSCheck configuration being able to make the second core of each module either be disabled (using roughly zero power) or run at very low clock frequencies/voltage and thus very low power when not running all eight cores at load.AMD's CPUs can compete quite well, you just havr to know how to use them. AMD obviously doesn't know how to set up their CPUs properly, so it's no surprise that most people (even within the tech communities) don't know of this either.[/citation]

I dont really know why this is downrated. I read it completely and it is reasonable. And actually, pretty much the truth.

Its amazing the amount of fanboyism going on these days, every 'gamer' buys extreme elite machines for ridiculous prices, when you can get all that you need for much less. The stupidity on people always brings out the asshole in me, and downrating a witty comment like this is an insult to people that knows computer hardware.
 
[citation][nom]ninjarubberduck[/nom]PC Perspective reported this same thing but updated it saying that AMD denied any additional near future price cuts.[/citation]
The source of the article is indicated as PC Perspective. Also, I and probably tons of people hope you're kidding.

[citation][nom]Schnitter[/nom]Thanks AMD for making it easier to boycott nVidia. So long as PhysX is still around forcing physics calculations on video cards while my CPU stays at 50% usage or less, I will not buy any nVidia product, or any game that supports PhysX (Borderlands 2 for example, no matter how good it is, won't buy it)[/citation]
I'm not sure what you meant as to why you wouldn't buy PhysX-related stuff. Mind explaining?
I'm just guessing that you meant that PhysX doesn't seem to help your CPU. If that's the case, I think it's because (maybe just in some cases) PhysX allows games to add more physics effects that run on your GPU rather than offload calculations from your CPU to your GPU.

[citation][nom]blazorthon[/nom]Have you actually looked into those CPUs? Take an FX-4100, bring the CPU/NB frequency to 3GHz from the 2GHz or 2.2GHz stock, use PSCheck to lower the P state frequencies of the second core of each module, and and heighten them for the first core of each module will let it beat the i3s significantly even without overclocking the CPU frequency. You can do the same with the other FX CPUs and you get some huge improvements. The FX-8120, with this, can give even the LGA 1155 K edition i5s and i7s some excellent competition in OC versus OC performance and at a much lower price.[/citation]
Hey blaz. :) I've noticed you've been posting around about that PSCheck method along with the L3 cache one. The latter is understandable, but can you elaborate why tweaking the P-states of the 2 cores of a module in way that you're somewhat "transferring" some of the clocks from the 2nd to the 1st core. Would that really help with the scheduling/shared module resources problem or does it address another issue? I'm imagining that since the Windows hotfix activates the 1st core of every module first before the 2nd, this would just provide better overclocked performance for the 1st cores which are probably being used more.
Thanks. :)
 
[citation][nom]army_ant7[/nom]Hey blaz. I've noticed you've been posting around about that PSCheck method along with the L3 cache one. The latter is understandable, but can you elaborate why tweaking the P-states of the 2 cores of a module in way that you're somewhat "transferring" some of the clocks from the 2nd to the 1st core. Would that really help with the scheduling/shared module resources problem or does it address another issue? I'm imagining that since the Windows hotfix activates the 1st core of every module first before the 2nd, this would just provide better overclocked performance for the 1st cores which are probably being used more.Thanks.[/citation]

Good catch, I forgot to add the CPU prioritizing settings need to be edited too.

Cutting down the P states would mean that utilization on the second cores of each module would skyrocket when Windows runs pretty much anything on them, so they might actually be avoided for with heavy threads because Windows shuffles things around based on utilization, but I'm not sure of how good it would do and I doubt that it would be as good as manually setting prioritization.

The Windows 7 hot fix would probably be detrimental to performance with this done because it tries to load modules one at a time rather than semi-randomly throwing things around the cores.
 

Wow! A pretty clever way to work with how Windows...works. Kudos to the person who though of that and to you for sharing and getting the word out.

I'm thinking that hot fix would still help (not in a direct way with that method though) since it would help with keeping the the 1st cores of the modules active thus having more cases of only them to throw around threads to.
 
Can you post some benchs of this? I got a 2500k last year august b4 I knew, but since all the pcs I've built/bought have been AMD. (phenom ii but yea). I'm wondering how good it gets and just for me to show on the net to shut up intel fanboys.
 


Honestly, palladin9497 was the member who first told me about PSCheck. Before that, my recommendation was disabling the second core completely and although that would let the FXX-81xx CPUs trade blows with the i5s quite well, they simply didn't have enough in the way of highly threaded performance in this usage to compete with the i7s' Hyper-Threading Technology.

I think that the hot fix would hurt performance in this usage because it tries to fill the second core of a module before filling the first core of the next module. The hot-fix might not hurt in single threaded tasks or in tasks that rely mostly on a single thread for performance, but even most games can effectively utilize at least two threads nowadays and the same is true of a lot of other software. At worse, the hot-fix would hurt performance significantly and at best, it wouldn't help much at all, if at all, so it is unlikely to be a good idea to install it if you alter the P states as I've described above IMO.
 
nice, but i stuck with nvdia vision 3d but like it, amd doesn't produce the same level of graphic in game than nvdia despite might have higher frame rate.
 

Sorry if I'm not getting something, but I know that you're open to different thoughts so here goes. I'm trying to create in my head how the hot fix and this PSCheck method would work when both applied and they don't seem to conflict.

If the hot fix prioritizes the 1st core of each module anyway, then if 4 threads are active at one time for an FX-8000 series CPU, then the desired effect is reached (stressing 1 core each module to monopolize the shared module resources).
If there are let's say 6 threads active at one time, then at least 2 modules would only have 1 core running each, and then the 2 other modules would have both of their cores active, but maybe having the heavier thread being used on their higher clocked 1st core because of the PSCheck method. Both the hot fix and this method working at the same time seems to be even more beneficial in some situations (like those that don't have 8 or more threads active at one time) rather than just utilizing only one of them.

Again, sorry if I'm sounding silly because I didn't get something, and thanks. :)

I agree with rexdale_punjabi's sentiment. It would be nice if you would be so kind as to dig-up some benchmarks of this if you have some. :)
 
[citation][nom]army_ant7[/nom]Sorry if I'm not getting something, but I know that you're open to different thoughts so here goes. I'm trying to create in my head how the hot fix and this PSCheck method would work when both applied and they don't seem to conflict.If the hot fix prioritizes the 1st core of each module anyway, then if 4 threads are active at one time for an FX-8000 series CPU, then the desired effect is reached (stressing 1 core each module to monopolize the shared module resources).If there are let's say 6 threads active at one time, then at least 2 modules would only have 1 core running each, and then the 2 other modules would have both of their cores active, but maybe having the heavier thread being used on their higher clocked 1st core because of the PSCheck method. Both the hot fix and this method working at the same time seems to be even more beneficial in some situations (like those that don't have 8 or more threads active at one time) rather than just utilizing only one of them.Again, sorry if I'm sounding silly because I didn't get something, and thanks. I agree with rexdale_punjabi's sentiment. It would be nice if you would be so kind as to dig-up some benchmarks of this if you have some.[/citation]

The hot-fix doesn't prioritize for the first core of each module, it prioritizes by module, not core. For example, it tries to fill each module before moving on to the next. If there are two heavy threads, it will try to let them both run on the same module instead of letting them run on the first core of two modules. The problem with this is that the module's total throughput is a little lower than it is even without the P state mod (you'll probably have to drop the second core's frequency more than you can increase the first core's frequency) and you're basically wasting the core configuration improvement.

As for benches, this is the best that I have right now:
http://techreport.com/articles.x/21865

It shows the effect of completely ignoring the second module by using thread affinity and compares the threading to non-threaded with a higher frequency, but it proves that changing thread configuration has considerable impact on FX performance. I know, you've seen this one before, but it's what I have available. The Windows 7 hotfix basically does what the third threading test in this link does, if I remember correctly, and you can see that it did not really help even with the stock threading config.
 


I'm sorry to say this but you may have that backwards there, blaz.
Two threads running on two separate modules have access to two front ends (and two FPUs), while two cores running on a single module must share both the front-end and FPU. A smarter OS might know the most effective way of distributing the load, which AMD stated would be a feature of Windows 8. Fortunately, MS released a hotfix to address some of what was purportedly going wrong in Windows 7.
Click to expand...
http://www.tomshardware.com/reviews/windows-7-hotfix-bulldozer-performance,3119.html
I really don't mean to be annoying and I wish I knew a better way of disagreeing with someone. 🙁
 
Windows 8 has the proper threading as described in that caption and helps Bulldozer performance considerably as a result (current versions of Linux also have more proper threading). The Windows 7 hot fix does not apply the same changes that the Windows 8 improvements do. Windows 8's improvements aren't the same as hat I'm saying here, but they're better than Windows 7 because they really are smarter.
 

Yeah, I do remember learning that Windows 8's improvement is better than this Windows 7 hotfix from you and other places I think, but the Windows 7 hotfix (which just seems to be a poorer implemented fix compared to Windows 8's) could still stand to show some improvements with the PSCheck method by the way it works(now that we have what the hotfix does, cleared out), right ? :)
 


Look at it this way:

FX-8120: cores: 8
[1-1][1-1][1-1][1-1]
The hot-fix tried to force programs to fill each module before it fills any other modules, so at best, you get slightly better performance from the Turbo feature being more aggressive.
Altered core configuration would probably be something more like this:
[1.7-0.2][1.70.2][1.7-0.2][1.7-0.2]
With the second core of each module prioritized to only be used when either all four modules have their first core in-use or when a thread is related to the thread that is on the first core, so it'll benefit from sharing a front end with it, performance is now increased greatly in any task that uses four or fewer threads efficiently. Not only does it get to have its four threads run at a much greater frequency, but the cores in use also benefit from getting the front end of their modules to themselves, so the performance could roughly double in such situations quite easily and with reasonable power consumption.

The hot fix would try to load both cores of the first module and then the same for each successive module. It would be a roughly 10-40% performance drop in comparison because of that. Windows 8 would be better and avoiding this hot-fix while setting the prioritization settings in PSCheck to focus on loading the first core fo each module before the second would also be better.
 
Here's the thing blaz. I believe you've got backwards what the hotfix does. The article I shared in the link mentions how Windows 7 by default just crams threads into one module at a time instead of the 1st core of each module, not the other way around, and the hotfix attempts to fix this behavior by activating each module before using the 2nd core of one module. This uses more power but provides better performance as we both know. :)
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom