erek
[H]F Junkie
- Joined
- Dec 19, 2005
- Messages
- 10,875
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Windows recognises the fx 8120 as officially a 4 core now. Interesting
- Thread starter erek
- Start date
mvmiller12
[H]ard|Gawd
- Joined
- Aug 7, 2011
- Messages
- 1,519
My FX-8350 was ALWAYS listed that way...
Fact is this is false, because it's an 8 core Integer with shared FP 2 by 2. Not sure how Windows treats that because of Windows limitations but Linux may say otherwise. Means your OS will work as an 8 core because data treatment and integers make the core on CPUs. But justice has decided it's 4 core.
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
Theory can be argued either way; AMD's 'modules' look quite a bit like an SMT2 core from Intel or AMD's own Ryzen. The bigger challenge for Bulldozer was that AMD declared their modules to be comprised of two 'cores', yet each of these modules were slower than the single-threaded cores on the CPU architecture they were meant to replace.
One can understand being a bit miffed about AMD selling a shiny new 'eight-core' CPU that is slower than the quad-core CPU it replaced.
Master_shake_
Fully [H]
- Joined
- Apr 9, 2012
- Messages
- 17,794
it's not a 4 core.
again there is still no legal definition because amd settled the case.
again there is still no legal definition because amd settled the case.
If I remember there were some highly customized workloads or maybe it was custom Linux Kernels that really allowed the FX line to blaze! The chips showed massive potential when coded for properly. But no way was Microsoft going to change their entire empire for one silicon iteration.
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
Bulldozer was always promising, on paper. I still like the idea personally, but like many ideas that are great, quality implementations must follow and in the case of Bulldozer, they just didn't. The architecture still excelled in some use cases, but in general that just meant being almost competitive rather than being a decent challenge further up the stack.
Bulldozer was a wait to cheat people out of performane
saying 8 core but only getting 4 cores of performance in FPU ( and thne the FPU was pretty bad as well)
dont care what marketing gimmicks say when a task that can scale linearly with threads doe no longer scale linearly with threads its because you don't have the core to pack it
in FPU heavy process the "8 core*" cpu show same scaling ability as a cpu with 4 cores and SMT but at least Intel only sold those as 4 cores and not market as 8
Integer different story. that's why it get convoluted. on integer heavy process the cpu scaled like an 8 cores
so yeah its kinda both 8 and a 4 core depending on the instructions/part of the cpu yoi are using
the idea was good. but they marketed as it best situation which leaves ppl with a bad taste in their mouth
P.S.
this is just as easy to test as SMT is with affinity. take 5 minuttes to show it
however if recent windows 10 1909 ryzen patch is also applied to FX we should see speed improvemtns
if not there is still project mercury to overcome the bulldozer lthread conflicts and get more game performance
in short:
Calling it 8 core is over prommising
calling it 4 core is not giving it enough praise
is both at the same time 8 core integer. 4 core FP
Thatwhat the CMT design was all about
saying 8 core but only getting 4 cores of performance in FPU ( and thne the FPU was pretty bad as well)
dont care what marketing gimmicks say when a task that can scale linearly with threads doe no longer scale linearly with threads its because you don't have the core to pack it
in FPU heavy process the "8 core*" cpu show same scaling ability as a cpu with 4 cores and SMT but at least Intel only sold those as 4 cores and not market as 8
Integer different story. that's why it get convoluted. on integer heavy process the cpu scaled like an 8 cores
so yeah its kinda both 8 and a 4 core depending on the instructions/part of the cpu yoi are using
the idea was good. but they marketed as it best situation which leaves ppl with a bad taste in their mouth
P.S.
this is just as easy to test as SMT is with affinity. take 5 minuttes to show it
however if recent windows 10 1909 ryzen patch is also applied to FX we should see speed improvemtns
if not there is still project mercury to overcome the bulldozer lthread conflicts and get more game performance
in short:
Calling it 8 core is over prommising
calling it 4 core is not giving it enough praise
is both at the same time 8 core integer. 4 core FP
Thatwhat the CMT design was all about
Last edited:
PliotronX
2[H]4U
- Joined
- Aug 8, 2000
- Messages
- 2,070
Its just funny how history repeats itself because FPU deficit was what held K6 back but users who never gamed got a good deal. Then K7 came along like Ryzen and the rest will be history 
I think without Bulldozer we wouldnt have chiplets at least not yet. I think that failure led to the incredible ryzen.
Last amd I had before Ryzen 1st gen was phenom ii 965 be and way before that was Athlon 64 and before that was probably something I forgot about but definitely first amd was 486 dx4-100mhz.
I've had intel between those gaps.
So I never experienced the bulldozer line simply because it came out when I was on a good Intel platform. I think my 3930k at the time.
Last amd I had before Ryzen 1st gen was phenom ii 965 be and way before that was Athlon 64 and before that was probably something I forgot about but definitely first amd was 486 dx4-100mhz.
I've had intel between those gaps.
So I never experienced the bulldozer line simply because it came out when I was on a good Intel platform. I think my 3930k at the time.
FlawleZ
[H]ard|Gawd
- Joined
- Oct 20, 2010
- Messages
- 1,691
I ran a 980 BE at 4.4Ghz during this time and Bulldozer was not an upgrade for gaming at the time. The only potential upgrade would have been an X6 Phenom II but Thuban topped out around the same as most C3 Deneb chips and most games still couldn't maximize anything over 2-3 threads. With my 980 @ 4.4Ghz and my OCZ Reapers at 1500Mhz 6-6-6 it was about as good as as the AM3 platform could do for gaming.
The speed has never been the spec used to count the number of cores.
It's not SMT. There are 2x real cores. As everybody knows, the kernel of an OS relies on computing integers. Floating point is another problem, and should be made so. Not sure how Windows handles that now, since Windows NT4 (which is basically an old Windows 10 kernel) and up to Windows 98 needn't any FP unit to run, for sure. And this was the case of sth around Linux 1.2 that I ran over a 386 without a 387 FP unit too.
They settled because they didn't want to create bad publicity when they launched new products that had nothing to do with their older one. However, AMD was right. There were 8 cores, not 4.
This is just like Windows 10 up to 1903 unable to use the 32 cores of the Threadripper 2990Wx. This is not because the AMD product was bad but because Windows product was bad and bugged.
Showing in Windows the FX 8xxx had 4 cores was buggy. This also has to do with that line of AMD CPU being much slower on Windows than Linux, again. The fact that it didn't sell well, made Microsoft to not even bother about the problem in their OS. Now that people are buying Threadrippers, and those people claiming they will run Linux because Windows' slow, just scares Microsoft who simply made the effort to correct the bugs in the Windows Kernel to support AMD products.
No, it wasn't. It never cheated anyone. Anyone who read reviews at the time knew what they were getting.
This is barely English, but it appears to be wrong from what I can tell. The FPU has nothing to do with whether or not the processor is an 8 core or a 4 core. FPU's have often been entirely separate or not part of a CPU design at all. There is no legal or technical definition where a CPU core has to have an integrated or paired FPU to be considered a full CPU core. It was a bad design, but calling it an 8-core CPU isn't over promising. Giving a CPU any amount of cores does not by itself promise anything.
That's speculative at best. I don't know that I agree with it either. There is no lineage between Ryzen and Bulldozer beyond Ryzen coming after the latter chronologically. I'd say that at the very least, this does not have to be the case. That is, there is nothing specifically in Bulldozer that necessarily lead to AMD coming up with Ryzen. That's like saying the Pentium 4 lead to the Core 2 micro-architecture. There is literally nothing in the design carried over. In fact, the Core 2 was based on the Pentium Pro's P6 microarchitecture. The only thing that influenced the Core 2's design to be better was Netburst sucking so bad. Multi-die chip designs date back to the 1970's in IBM bubble memory. The Pentium Pro had a separate cache die and so on. The modern "chiplet" isn't something that belongs to AMD alone. The point is I don't think Bulldozer has a damn thing to do with it beyond sucking bad enough to force AMD to regroup and do things differently.
At the current time this DPU releases no CPU in the consumer market haved had cores with missing FPU parts for decades. That we once had no FPU with a core back in the early 32bits days of a 386sx does not really apply to todays expectation to a CPU sold today
Rver since the 486 the FPU has been includrf in the "core so ever since multiprocessor existed the FPU unit was part of the core
CMT design isolated that and gave a single FPU unit behind 2 integer units. So 2 threads had to share thesame FPU unit just as if it was only 1 core per each 2 core said on the box.
That the design that how AMD built it. but in te box it did not say "hey we changed the common percepetion or core by removing half of he FPU units.
If they had been honest with the marketing it would have had to be a problem.
You can argue that in fact there has never been a "core" with no FPU units as when the term came around with multicore. They all had FPU units behind each INT unit.
AMD is the only one in the consumer market to my knowledge that did this kind of under delivering.
They put Themselve in a hard position with this hybrid design of 8 INT units wit only 4 FPU units.
but the fact remain these CPU could not scale to more than 4 cores when it came to FPU
Also:
AMD does not promise with review are showing. i cant go to AMD and say hey this site tested you cpu to give me ex amount of performance.
What AMD promises is what is on the box when I buy the product
review are not a free pass to play loosey gossey with terminology
Last edited:
No. This is simply false and your missing the point. The i386SX, i486SX and NexGen Systems NX586 all lacked FPU's.
Again, there is absolutely no legal definition that requires an FPU to be part of a "CPU core". There is no legal definition or constraints for what constitutes a CPU core, nor should there be.
False. This assumes the masses have any preconceived notion of what constitutes a CPU core. They do not know what a CPU is in any detail. They don't know L2 cache from an FPU, much less equate FPU units as being part of integer units. It was also well known at the time that AMD's design centralized many resources and shared them among execution units. People who were reviewing these CPU's and working with them at the time knew the performance was terrible. They were way behind Intel on FPU performance and that's frankly all anyone needed to know. The average public knows dick about semiconductor design.
Whether there ever had been another multi-core CPU with fewer FPUs than integer units is irrelevant. It is neither a technical or legal requirement to include them along side each integer unit.
It's a bad design that performed horribly. No one disputes that. It's why AMD was relegated to bargain basement builds and upgrades to existing AM3 motherboards during this era. Its why AMD couldn't hope to compete with Intel on equal footing and didn't. Again, you have provided no evidence that a CPU core must contain both an integer and FPU unit to be considered a fully functioning core. It's frankly absurd.
Actually, reviews tell you what kind of performance you can expect in the applications tested against to a reasonable degree of accuracy. AMD never hid that it under performed beyond the usual cherry picked benchmarks they released which down played their weaknesses. Guess what? AMD and Intel still pull that crap today. I've seen plenty of press kits on such matters. Right now, Intel press kits don't mention the competition and highlight AVX-512, Deep Learning Boost and gaming performance. Intel even touts its overclocking prowess. What it doesn't do is talk about how its gets owned for less money in everything else. The point, never believe what some marketing department tells you. Do your research. Buyer beware and all that.
AMD never promised an equal amount of FPUs to integer units. And really, if you can back far enough integer units alone would be all that's required for a CPU core in an x86 architecture based processor. Everything else is just fluff. Besides, non-Intel CPU's having shit FPU performance has been the standard for about 30 years. If you thought AMD had to give you 1 FPU for each integer unit despite plenty of information that told you how these were built being available at the time, then that's on you.
You show me one piece of material from the time Bulldozer was released, from AMD that states it has more than 4 FPU's, and you'll have a case. Without it, you simply don't. Your misconception that FPU and integer units have to be equal is just that, and it isn't evidence to support your position.
DeathFromBelow
Supreme [H]ardness
- Joined
- Jul 15, 2005
- Messages
- 7,315
Imagine if AMD just flat out bought reviews and journalists to cover up bad news about their products.
AMD wasn't deceitful, it's an 8 core processor with 4 shared FPUs. The weak single-threaded performance is from the small cores. It was a perfectly fine 1080p gaming processor that could do a lot at once, and that's how they marketed it. I still have a Bulldozer and Piledriver in service. They were a lot of fun to play around with back in the day and they're still solid CPUs (4.6/4.8 GHz @ 1.45v since I bought them). No huge security nightmares like Meltdown, either.
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
Who says they haven't? They certainly have a history of misleading marketing -- after their drivers, it's one of the areas that they most need to improve.
I didn't say they were.
The implementation of their idea was just weak period. It failed to surpass its predecessor and was stomped by the competition right out of the gate.
This is a wild claim -- and really, purely subjective. 1080p is not a remotely comprehensive measure of CPUs.
Their marketing is the basis for their own payout.
I never claimed they didn't work; quite the contrary, actually.
If you wanted a space-heater? I owned one too, by the way. Bought used because it was cheap, and price was the only reason to ever buy one.
Why would anyone go looking for security vulnerabilities in a CPU architecture that was never deployed to the enterprise?
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
I haven't seen any prominent ones that couldn't be attributed to poor methodology. Producing good reviews is really very hard work that not everyone is cut out for; there's a reason I never tried to get into it myself despite having an interest and an aptitude for the technology.
I think people forget how difficult it is to rigidly test something and then to follow the numbers despite where their preconceptions started them. Whole branches of philosophy have been created to figure out how to do that successfully.
Sometimes its about properly placing a piece of technology in the proper context. A given CPU might be terrible, but really great in a certain price point etc.
This is true. I ran tests on the Linux kernel. In Linux it was comparable to Sandy bridge and in many tests it would beat it if you considered cache sizes during complication. The main problem that chip had was its cache was tiny and slow. Many things such as games are very cache sensitive and there was no way to fix that. But if things stayed in cache then performance was actually ok. TDP sucked though.