Monday, July 11th 2011
AMD FX-8130P Processor Benchmarks Surface
Here is a tasty scoop of benchmark results purported to be those of the AMD FX-8130P, the next high-end processor from the green team. The FX-8130P was paired with Gigabyte 990FXA-UD5 motherboard and 4 GB of dual-channel Kingston HyperX DDR3-2000 MHz memory running at DDR3-1866 MHz. A GeForce GTX 580 handled the graphics department. The chip was clocked at 3.20 GHz (16 x 200 MHz). Testing began with benchmarks that aren't very multi-core intensive, such as Super Pi 1M, where the chip clocked in at 19.5 seconds; AIDA64 Cache and Memory benchmark, where L1 cache seems to be extremely fast, while L2, L3, and memory performance is a slight improvement over the last generation of Phenom II processors.
Moving on to multi-threaded tests, Fritz Chess yielded a speed-up of over 29.5X over the set standard, with 14,197 kilonodes per second. x264 benchmark encoded first pass at roughly 136 fps, with roughly 45 fps in the second pass. The system scored 3045 points in PCMark7, and P6265 in 3DMark11 (performance preset). The results show that this chip will be highly competitive with Intel's LGA1155 Sandy Bridge quad-core chips, but as usual, we ask you to take the data with a pinch of salt.
Source:
DonanimHaber
317 Comments on AMD FX-8130P Processor Benchmarks Surface
Keep in mind boys and girls, I'm not only the motherboard reviewer here @ TPU, I also game on an Eyefinity rig, so the bleeding edge of performance is where I've been for years now.
If you don't understand the how SuperPi relates, then I'm sorry, but you do not understand CPUs, and thier performance, very well. Just go to one of my reviews and check out F1 2010 results, if you want to see how SuperPi relates to gaming, as gaming is pretty important to alot of members here.
A graph for edification:
AMD's current lack of memory bandwidth affects F1 2010 in such a way that Intel 1155 rigs are a good 50% faster than Phenom II x6 rigs. Why? MEMORY PERFORMANCE.
Unlike many people posting here, I have benchmarks to back up why these things are important. Anyone else got some to show me wrong? Don't worry, I have more benchmarks, too. I haven't been harping on AMD's memory perforamce for years for no reason.
GPUs are already at 5 Teraflops
CPUs lol below 200 GFlops stillIt is a roller coaster ride because by the time LGA 2011 comes out we will be worrying about how Komodo performs compared to Zambezi
Zambezi (2011)
Komodo (2012)
Next-Gen Bulldozer(No codename yet) (2013)AIDA64 uses only 1 core for that benchmark what do you expect happens?
Stock Clock -> TC1 -> TC2
Turbo Core Mode 2 = 2 Modules are gated/turned offI noticed it right away :|Wait and see?
There will be an increase how big I don't know
10% to 30%
BUT, do not BE surprised when Zambezi Retail CPUs go up to 4.8GHz on Turbo Core Mode 2Moving/Downloading doesn't use much processing power these days run it with IntelBurntestZambezi isn't Phenom II
That 50% increase in performance well....
30% increase in NB(IMC) performance+a Predictor or Prefetcher(what are those things that keep data called in Computer Engineer Jargon) and what is the percentage of going from 1333MHz official support to 1866MHz official support?
With both of those memory performance skyrockets from Phenom II
Edit: you guys post like crazy
Unsurprisingly, as you've been prone to do today, seronx, you didn't actually read what you were responding to. I'm starting to think it's a reading comprehension deficiency.
I cannot drop 1155 memory performance to equal PhenomII, nor can I increase PII performance to SB levels, so the request you make is not possible.
I can, however, tell you to look at all the result in my reviews, because all the results together, they DO give the perspective you are asking for. It's not a theory, BTW, it's a FACT that AMD's memory performance suffers compared to 1155, and a FACT that this has it's impact in gaming.
SeronX asked for wPrime numbers...oh look, those are in my reviews too. :laugh:
Just take alook here:
www.techpowerup.com/reviews/Biostar/TZ68A_PLUS/9.html
Also, you're acting like I'm not reading your reviews. I told you once I bought my board BECAUSE of your review, I'm not going to tell you again. :p
The Biostar review just provides more data points for the same CPU. Also, the wPrime numbers indicate there's bandwidth to spare if you have extra cores to use it.
I want to see the Flex FPU at work ugh :cry: in wPrime
The only things I have been seeing so far is what I already know!
show me the wPrime 1024M scores Donanimhaber
Second, I will wait until I get my own hands on one of these before I make any judgments or assumptions.
Finally, while I don't expect miracles, I expect a nice improvement over PII in test that are properly setup, as Bulldozer supports 5 new instruction sets including SSE 4.1, SSE 4.2, and AVX. I don't expect miracles, but I expect consistent, across the board improvements over PII's.
And while I am posting here, I also expect a lot of other testing platforms to be outed for basing their tests instruction path on the name of the chip instead of the chip's actual abilities before this year is out.
While the 8130P is the current flagship, I would like to note the A8-3850 in their naming scheme. We all know the Llano is a 32nm Athlon II with better memory support so their is not much more they can do with it, thus the rather high number for the last 3 digits. The only reason AMD would leave such a huge gap in the FX line's numbering scheme is if they expect to release more chips later on with higher clocks and/or improvements from later stepping. By the end of next year we could have a AMD FX-8550P clocked higher with more OC headroom. No way to know for sure, but wait.
The fact of the matter is that each and every benchmark you see in my reviews was vetted extensively to validate it as a performance compare.I know, I know, but you know, as staff here, I do feel it's my job, not only for you, but for everyone, to make sure that things stay factual, and away from the FUD.Actually, the bottom two numbers are an 1100T. The P7P55 D-E Pro has a i5 750 in it. I do also have an I7 870 too. Interesting thought about wprime, but to me, that shows that although memroy performance is lacking in AMD, thier core performance is very good, so my perspective is not the same as yours, I guess.Yeah, and that's AMD saving grace with Phenom II...overclocking. maybe 3% of users here run stock...and time has shown that overclocking to the maximums presents a very different picture than stock numbers. But again, not everyone is going to have a CPU as good as yours..the 1100T i have is a much worse sample than yours, so while it worked out for you, if you had MY AMD CPU, I do not think you'd be as happy. :laugh:You should hop on our TS server. We do talk about these things there too(from time to time), and through these discussions, it's kinda become clear that L3 cache performance is what's also lacking in AMD CPUs(well, my benchmarking has brought it up too).
I mean really, get rid of AMD's "Unganged" mode, and thier memory performance is even worse, but at least then both Intel and AMD would be using 128-bit memory controllers. The real truth of the matter is that Phenom II CPUs have a memory controller that can split into 2x 64-bit, and even then, they are still very lacking.
Bulldozer is similar...it's a quad-channel controller, in it's "native" socket(I do not consider AM3+ a real BD socket), so I do beleive that part of the problem with desktop Bulldozer parts is that 1/2 of thier memory performance will never be realized on the desktop platform...these chips WERE designed with quad-channel in mind.
BD FX 8130 is not an 8 core as SB 2600k is not one either even if the task manager shows 8 cores, but they are not classic 4 cores either.
Just because AMD marketing calls it that way doesn't mean anything, Its just for the average joe who is buying a cpu based on frequency or number of cores and what will he buy if he can choose between an AMD 8 core 4.2Ghz(turbo) or Intel 4 core 3.8Ghz(turbo) , of course he will choose AMD because he will think It will give him more than twice the performance because of double amount of cores and a bit higher frequency, nothing more. 8 cores are really just for marketing purposes, but in my opinion it was a really stupid move from AMD, they could have claimed BD is a 4 core which has almost the power of CMP designed 8 cores with much lower die space and not how they call it as an optimized 8 core with a bit lower performance but with a much lower die space.
They are both CMP designed 4 cores but to gain even more power without substantially increasing the core area they use additional design in combination with CMP(chip multiprocessing), AMD is using CMT (cluster-based multi-threading) which needs more core area but can give you more performance and Intel is using HT what is a form of SMT(simultaneous multi-threading) which on the other hand requires less additional space but will give less performance.
SMT or HT in this case increases core size by a bit more than 5% probably 5-7% I don't know If it is with or without L2 cache
CMT increases core size by 20% without L2 cache and with L2 cache just by 12%.
www.xtremesystems.org/forums/showthread.php?265710-AMD-Zambezi-news-info-fans-!/page8
SMT, CMP, CMT
citavia.blog.de/2009/07/07/more-details-on-bulldozers-multi-threading-and-single-thread-execution-6464533/
and here is an important old presentation picture about multi-threading done right
www.blog.de/srv/media/media_popup_large.php?item_ID=3663732
What is a CMP? 2 cores, 3 cores, 4 core and so on
And what is a CMT? It's a modul
info.nuje.de/Bulldozer_Core_uArch_0.4.png
you can see a modul is just a single core with 2 integer clusters and not 2 cores based on CMP design. If you wanted to call a modul as 2 cores you would need to have everything twice and not just 2 integer clusters, thats why it shouldn't be called 8 cores because its not but instead 4 moduls or 4 cores with CMT design or something similar.
8100 = 3.5GHz 8110 = 3.6GHz 8120 3.7GHz and so on
8550P would be this godly 8GHz beastI quoted you for some reason but I forgot after I saw someone saying the FX Processor wasn't an 8 core
ISA behind, Memory not so much they are just missing a component Intel CPUs have (I think it is either called the prefetcher or predictor as said a million times before)((It doesn't help that the K10.5h CPU is using a rehashed/super retweaked K7 IMC and Zambezi isn't)
Overclocking isn't a big field that is why to optimize usage AMD invented the 2nd turbo core mode so people didn't have to overclockAMD Zambezi is an 8 core CPU
It's an 8 core not because of CMT/Some awkward sound french guy or some smart person who owns his own tech forums
It is an 8 core BECAUSE EACH core has it's own resources
CMP over provides
SMT under provides
AMDs CMT under marketing terms says it provides balance
Meaning you can't compare it to either cpu design and say it's not an 8 core
Edit: I ate and I ready to talk about Zambezi again
www.techpowerup.com/reviews/AMD/Phenom_II_X6_1090T/5.html
1090T is a 3.2GHz part, so you can compare it to the 955. 248 seconds vs. 372 seconds (using rounding), which is roughly 50% faster. Linear scaling. That means that the CPU is not memory bottlenecked at 6 cores or 4 cores. The memory performance isn't lacking if scaling is linear. And in that case, the performance per-core is just terrible.
I'm more interested in the thoroughly awful scaling of SuperPi scores, because that *can* imply that there is a memory wall in that benchmark, but only if the benchmark can make full use of all 6 cores. Does it? I don't know that answer; I only have 4 cores available to me in any one machine.
But it's not necessarily any better on Intel, considering the i5 661 with its 2 cores scores second-best at 1M. There's more L2 cache on the i5 and i7 and more processing power available (in terms of it being a function of clock speed x # cores). Does that mean LGA 1156 had awful memory performance? I dunno, because we can't put the Phenom II's memory controller on an i5 or i7. Isolating this stuff is darn-near impossible outside of AIDA 64 or Sandra benchmarks, where Intel has a lead, but not a 30-40% lead.
We can argue til were blue in the face but it still doesn't explain anything. When we get better benches and NON ES chips then we can start to see what is really going on.
But I don't think you could disable such "modules" in bulldozer. So they are permanent cores. So in this I can say that BD has 8 real cores anyway. My question is all about scaling.
Real cores on an opteron G34 for example, scale pretty well, not sure the exact number, but I bet is better than 80% in efficiency terms.
We all know intel HT helps a lot in certain scenarios, but its far away from being magical and cannot be considered as an additional cores, well all already know that. BD is different, is so different that we are discussing if we should consider each modules as core or both modules as a core. A new term could be introduced....
AMD Bulldozer module
There are two cores in there beyond belief I proved it hahahaha mwaahahaha :laugh:
I'm a little worried about the size of the "cores". They look terribly small...