AMD’s FX-8150 vs. Core i7 & Phenom II – Bulldozer Arrives!
What is Bulldozer?
Phenom II has reached its end of its useful life. There have been few changes to AMD’s basic architecture since about 2003. Bulldozer is the next iteration of AMD CPUs that are particularly useful for servers and have also been adopted for the desktop as the basis for AMD’s next generation CPUs that will probably be in use for years to come.
Here is AMD’s now-public roadmap:
Now that we know that Bulldozer is expected to scale beyond 2014, lets look at the architecture.
Bulldozer’s design allows two cores to each share a larger, higher-performance function unit – as a floating point unit (FPU), for example – with less total die area than having separate, smaller function units for each core.
The floating point unit in “Bulldozer” has been re-designed from Phenom II. It has been improved to support new instructions and to allow resource sharing between its cores. There are two 128-bit FMACs shared per module which allows for two 128-bit instructions per core or one 256-bit instruction per dual-core module.
“Bulldozer” Front End
Bulldozer’s front-end unit drives the processing pipeline and is designed to make sure that all of the cores are kept busy with information. The front end works with each dual core unit to allocate threads to individual cores. AMD made significant changes that include decoupled predict and fetch pipelines, as well as prediction-directed instruction prefetchers. It also means that Bulldozer’s pipeline is longer.
A Prediction Queue manages direct and indirect branches that are fed with a L1 and L2 Branch Target Buffer which stores destination addresses. Bulldozer’s modules can each decode up to 4 instructions per cycle versus 3 on AMD Phenom II processors.
The prediction pipeline produces a sequence of fetch addresses. The Fetch pipeline does a look up in the instruction cache, and pulls 32 bytes per cycle into the fetch queue which feeds the decoders.
Bulldozer uses a physical register file (PRF) which is a single location that holds the register results of executed instructions. This reduces power by eliminating unnecessary data movement and data replication by keeping one copy instead of broadcasting the data.
Caches
Each Core is equipped with a 16 KB Level 1 Data cache, a 32-entry fully associative DATA TLB, and a fully out of order load/store – capable of two 128-bit loads per cycle or one 128-bit store per cycle. Each dual-core module includes a 2 MB 16-way unified L2 cache with an L2 TLB capable of 124 entry, 8 way that services both instruction and data requests. Bulldozer supports up to 23 outstanding L2 cache misses for memory system concurrency.
Finally AMD has designed a shared 8 MB L3 cache with 64 way associativity for both cores in a each module. And Bulldozer is planned for use in both the desktop and server areas.
AMD also redesigned the Northbridge.
Turbo Core
One impressive new feature that appears to work well in practice is AMD’s Turbo Core technology. It is far more sophisticated than similar technology used in their 6-core Phenom IIs. It allows for the FX-8150 to increase from 3.6GHz to 3.9GHz across all eight cores, and a further double-boost on 4 cores to 4.2GHz if TDP headroom allows it. Generally the boost effect can range from instantaneous to over a longer period, depending on the thermal situation.
Of course, this could give some significant performance improvements as AMD’s internal testing shows.
And of course we need to look at the AM3+ chipset.
It all comes together looking like this at launch:
According to AMD, FX processors offer:
- the World’s First 8-Core Desktop Processor designed for highly threaded applications,
- new instruction support for FMA4, XOP, AES, AVX, and SSE 4.2 for next generation PC applications
- unlocked to allow customer customization and higher system performance
Specifications:
FX Processors:
We just had a very brief overview of Bulldozer’s architecture. As with all CPU designs, there are compromises that make Bulldozer particularly attractive for AMD’s large server market but not so perfect for desktops, perhaps. We we need to look at is how the Bulldozer Flagship FX-8150 performs in applications and especially in games. Please read on.
Your review is a lot different than most of them out there. And that is a good thing. Its a different view comparing BD with the phenom 980 and the i7 920. Most other sites used the phenom x6 and intel’s Sandy Bridge. With the latter configurations its harder to see BD in a good light. Especially considering AMDs own 6 core phenoms which in, my opinion, currently a much better value. As a slightly revised model, I believe many ppl have overlooked the 1100t. it pulls away from then phenom x4 by a decent amount. Its a bar that bulldozer is struggling to surpass.
But your review focus is interesting because it does show BD isnt that bad compared to the phenom2s (4 core versions). Its not so bad at all in this comparison. AMD can only improve on this design from here. It will get better. Its just a terrible way to start it off. only if that 1100t (and 1090) wasnt in the picture, bulldozer would be a much better looking AMD option!
Thanks for your angle in reviewing the BD, it is something unique and useful. Its a review that i enjoyed mostly because it helps see a more complete picture.
The benchmark charts was a clusterfuck that I couldn’t understand. The gaming chart was much better. Need to create readable charts and organize the scores accordingly instead of copy and paste results at tiny pics which are hard to see, this review is a fail.
I don’t understand why this review was posted in the shape it’s in.
First off, apoppin, this is way below your standard just in terms of presentation. The pasted screenshots make comparisons awkward.
SATA scores with some 500GB seagate drive.. wtf? Use some high-speed SSD, maybe some fast USB flash to evaluate the south bridge performance.
Sometimes you have Intel scores, sometimes you don’t. How much work is it really to run SuperPi on the Intel setup?
And lastly, why the 920? The AMD chip just came out, I think it makes more sense to compare it to current Intel chips. The Sandy Bridge comparison is fair in terms of price and availability. Who buys a 920 today?
Overall this isn’t up to ABT’s standards, I feel.
First of all, I have to agree. And I am not going to make excuses for it. There is not even a real conclusion in my article.
Let me start with the last issue first – the i7-920. Don’t forget that it is benched at 3.80GHz with turbo on to 4.0GHz. That is faster that the current stock i7-960, Intel’s fastest LGA 1366 quad-core on their lead platform – the X58 motherboard (until it is replaced this month). As noted in the article, it would be about as fast as a stock i5-2500K. So it is a very valid comparison. That said, we are also again evaluating new CPUs from Intel and I just received a Core i3-2105 for comparison so we can see how the dual cores perform, including the Phenom II X2 in gaming.
The main issue with the FX-8150 evaluation was one of lack of time. Reviewers mostly had 7 days from the receipt of the FX-8150 until publication which is time pressure enough. However, the ABT review kit’s ASUS AM3+ motherboard was DoA; a hardware issue and it would not even post.
It took over two days to get a replacement AM3+ MB and that left only about 3 days for the entire evaluation (considering that it takes a day to set up Windows and 20 games and all of the benchmarks, updates and patches – and I did it twice).
Worst of all, CrossFire didn’t scale well with the stock-clocked FX-8150 on the replacement motherboard and it also had issues which meant a lot of retesting. While overclocking the FX-8150, the MB’s LAN became defective and a PCIe riser pulled loose from the MB with the video card, and I am now waiting for a third motherboard as replacement so that I can finish Part two of this FX-8150 Evaluation.
In the meantime, I will clean up my charts and put up easier to see images as needed. And I will add in the few comparable Intel benches which I had no time to run. An overall summary chart would also be nice. But I am saving that for Part two which will also explore CrossFire scaling with the overclocked FX-8150 (at 4.4GHz on air and perhaps further under watercooling).
All of ABT’s evaluations are always done from a gamer’s point of view. That is why you see 20 games benched as a minimum with far less emphasis placed on other CPU functions. Also, normally I run Total War: Shogun II and Deus Ex: Human Revolution, but they BSoD the PC when I tried to launch them with the FX-8150. AMD acknowledged it and believes that it may be due to an issue with Steam.
Over the next couple of days, I will clean up this article. But watch for Part II – that is the important part with a real conclusion about FX-8150 as a gaming CPU. Having two weeks with it (now) makes a lot of difference than taking a superficial 3-day look at it. I am not under time pressure now.
If I may make a suggestion to everyone reading this, follow us on ABT forum. The members find out exactly what is going on – first, before it is polished up for an article and you may even have input into the way the testing is conducted before we do it. How and what we test is driven by our members.
Fair enough — waiting for Part II. Thanks for the explanations!