Ask AMD about Bulldozer

Status
Not open for further replies.
Why all the tests in network for performance processor FX 8150

they use memories 1600 MHZ ?
if the processor need at least 1866 MHZ, is it not hurt the test ? Performance ?

Are memories of 2133 mhz can improve the performance ? with OC ?
 
Since many applications simply aren't suitable for multi-threading, and others can only be threaded so much, can we assume that AMD will in the near future start to focus on getting a competitive IPC again?
 
Is the floating point unit truely shared between cores for 128-bit code? In other words, does one core get full access to both FMACs for 2 X 128-bit operations if the other core is not using the FP unit?
 
Does using partially-depleted SOI over bulk still bring any meaningful performance or power benefits?
 
Does Piledriver fix any of the supposed cache latency and write-bandwidth shortcomings?
 
Are you working with Microsoft and/or the BIOS developers on any performance improvements at this time / if so, what are the projected improvement results? Please provide separate answers for each.
 
There are opinions that the strategy of labeling Zambezi with FX label is bad marketing, since FX has always been known for top performance parts, but some real world tests has shown that it doesn't actually have significant improvement in terms of performance compared to previous offerings.

My question is, will you revisit the performance of current Head of Marketing or at least consider using different marketing strategy?
 
I feel as though something went wrong with bulldozer, like AMD expected more out of it.

All of the market hype, the FX moniker, and the early rumors of extremely high total performance, all seem to point to AMD having much higher initial performance results.

Question: Was some sort of critical flaw found that required AMD to hamper performance in order to ensure such flaw would not end up in the final silicon?

This seems likely to me for a couple reasons:
1. Delays, it is obvious that you were trying to fix something. Yields have not been reported to be low, so I don't think that is the issue, Global Foundries' 32nm node seems to be relatively efficient.
2. High Clocks, over high IPC. Everyone learned long ago that more instructions per clock are a better solution than more clocks. the high clocks of Zambezi seem to indicate a last ditch effort to improve performance.
3. Current issues. BSOD in certain games. Some retail CPUs showing strange issues even at stock settings (random BSOD, black screen, etc.), maybe remnants of such flaw?
4. relatively high transistor count. It seems to me that you should have been able to reduce # of transistors. Possibly indicating that some of the transistors are left overs from the original "flawed" design.
5. Not a huge advance over Thuban. With the many years AMD sank into BD, it seems that a greater improvement could have been achieved.
6. Wonky BIOSes, and some older boards that originally had support for Zambezi, not working at all. To me this shows that something changed between last year when BD rumors, and such started swirling.

Thank you for your time, and I look forward to a response.
 
Can future revisions of Bulldozer or later CPU's allow mobo manufacturers to enable a more convenient placement of ram and 3rd party CPU cooling?
I hate eliminating certain cases/coolers from my list of options because they will most likely conflict with how close the cooler will be to my ram, or that the cooler will point toward the top of my case, and therefore need an exit.
 
one good idea is to rename them as 4 cores not 8 cores and by this you can artificially increase their prestation. being just 4 cores their level of performance will be fairly (kinda) decent vs a 4 core from Intel.
 
Is Bulldozer/Zambezi the last series to make use of the AM3+ mainboards ?
 
Why,after a 2 year development period, did you release a product that is on par with pentium4?(barely)
 
Compared to the older Phenom II design.

What is the primary reason or reasons, why the new Bulldozer CPU's 2 Core/1 Module design leads to slower single core instructions executed, per clock cycle?

Thank you, Justin. :)
 
Last edited:
Dear AMD,


I sincerely want to support your company as much as I can afford to, but I simply cannot justify or rationalize paying more for a chip that performs worse, uses more electricity and runs hotter than previous generation chips. I plan on building a gaming machine with the A8-3870K when it is released as I view it as a superior architecture for my needs than the bulldozer:

So here is my question:

If you cannot improve IPC on Bulldozer to Phenom II levels, and cannot reduce power consumption levels to 2600K levels, with future Bulldozer revisions, Will you please consider please continuing the stars core architecture used in Phenom II and llano on a 32nm process?
 
Last edited:
1.) When will the FX-8120 95W (FD8120WMGUSBX) be available in the U.S.?
2.) Why wasn't it made available when the other 8 core processors came to market as scheduled?
3.) When will the giveaway happen?
 
Last edited:
I am curious about how Bulldozer will perform once AMD releases the HD 7000 Series graphics cards, which communicate with the CPU differently than prior generations and alter GPU computation.
 
Any secret sauces between the Bulldozer FX CPUs and 7xxx GPUs whereby their total performance will exceed that of the comparable Intel I7's and I5's and same GPUs?
 
I think the memory bandwidth is something to blame for 8 cores performing like this debut CPU. Is there a quad channel chipset in the works?
 
I have question about cache:

Why L1 cache uses write-through instead of write-back approaches, when write-through cache means worse results ? (not to mention, that L1 Data Cache shrunk to only 16kB)

What impact to performance have shared write coalescing cache between L1 and L2 cache in terms of latency ?

Why all cache subsystem of BD have terribly increased latencies compare to K10 ?

Latency L1D, L2
'K10 ': 3 ; 12
'BD' : 4 ; 18-20

Do you have any plans for making Bulldozer without L3 cache ?

Bulldozer Front-end : previous K10 architecture have 32 byte fetch window. Bulldozer have 32 byte fetch window too, but for two cores ! It leaves only 16 byte fetch per core, which is in level of K8 or Bobcat. Why such a solution ?

is FPU working at the same frequency as integrer units ?
 
Last edited:
We have seen in some benchmarks that this cpu can be a multithreaded beast. I understand that Battlefield 3 and future games will use multithreaded DirectX11 drivers. Do you believe that Zambezi will have any performance advantages over the competition in this area given the extreme multithreaded nature of the chip? And are you working with Dice and or other Developers on similar optimizations for future game and multimedia titles?

Thank you for your time. I have bought your CPUs for many many years and I appreciate the fact that you stepped forward to answer questions.
 
Thanks for the input and keeping it above the belt. Submissions are now closed.
 
Status
Not open for further replies.
Back
Top