News Headlines
- Mon, Aug 08
- Indie adventure The Girl and the Robot escapes on August 18
- Bethesda is talking to Nintendo, considering game development for NX
- Rumor: Nintendo NX lineup to deliver new Mario, Legend of Zelda, and Pokémon titles in the first six months
- No Man's Sky day one patch adds multiple endings, requires new save file to experience
- PlayStation Neo reportedly planned to be revealed on September 7 in New York City
New Articles
Related Articles
AMD A8-3850 "Llano" & Gigabyte A75M-UD2H Review - PAGE 1
Carl Poirier - Thursday, June 30th, 2011 Like (1) ShareFinally, the first desktop processors from AMD using the 32nm manufacturing process are out. The mobile chips which launched a little earlier were reviewed by a few websites, and generally they were very well received.
By this time it should be known that Llano APUs comprise of an integrated Radeon GPU as well as its four x86 cores. Nearly every tech site has told the story of the merger of ATI and AMD, so you should be well seasoned over how AMD came to leverage ATI's Radeon technology.
At the Fusion Developper Summit '11, AMD made a great presentation about its upcoming Fusion System Architecture, truly designed to address heterogeneous computing. In fact, Llano and Bobcat are the first step to these very ambitious plans, which I will explain.
To drastically improve performance when handling large heterogeneous workloads, AMD first and foremost wanted to unify the address space for the CPU and the GPU. Anyone who has taken a dive in parallel computing languages such as OpenCL knows that the steps needed to efficiently process a given data set in parallel on a GPU aren't that simple; there are many of them to take and plenty of details to look at. Most importantly, the data needs to be transferred from the main memory to the GPU memory, and vice-versa after it's been processed. This could require lots of bandwidth in many cases, so unifying the address space would instead allow the GPU or CPU to access the data directly without transfers. But this requires a huge amount of work for both the silicon and the software teams, and it is by no means going to be done in one silicon iteration. The roadmap shows this as a four-stage process:
The Llano is the materialization of the first step; the CPU and GPU were grouped together in the same silicon, obviously using the same process technology. Then, the memory controller running the DIMMs was designed to serve both components. This was accomplished by keeping the same CPU and GPU architectures of the older generation K10.5 and Evergreen platforms, more commonly known as Athlon II and Radeon HD 5000 Series. Of course they were tweaked a bit, but nothing major on the performance side.
For the next step, it is question of supporting C++ constructs, virtual functions, DLLs and much more, on the GPU itself. AMD is currently working on a brand new GPU architecture which will do exactly that. This will be the compute unit, which now contains vector and scalar units, and has little to do with the Radeon cores found on current AMD cards. In fact, it was said at the FDS that the Southern Islands GPUs (AKA the Radeon HD 7000 series) should be utilizing this architecture, and they are rumored to come out this year, so it will be interesting to see how this turns out. We'll also witness bi-directional power management, which will most probably allow for a Turbo mode on the graphics engine, much like Intel already does in its Sandy Bridge lineup. Gaming performance reportedly will not be sacrificed along the way, obviously.
The third step is the architectural integration. As mentioned earlier, this is one of the most important difference in the architecture reviewed today. Whereas the buffer read and write operations would be computed on the GPU, at least in current versions of OpenCL, with Llano the data would be utilized in place directly. Also, the unified address space architecture would support direct access to the data by either processor using the same pointer, outside of OpenCL. A key component needed to do this is the IOMMU. There is one present in the 890FX and 990FX chipset, but only in the upcoming versions will the chipsets be able to cater to those needs. Finally, at the end, the memory unification will be extended to discrete GPUs, with some other optimizations such as context switching of the GPU.
Article Index |
|
So, with just this processor, that costs $135, you can get the same performance of a lower mid ranged dedicated GPU in games, and also a quad core processor that performs well in just one chip?
If I'm correct in that, would you recommend, if my friend wants to buy a gaming PC but has a restricted budget, that this would be something to buy until he can afford a better dedicated GPU?
Like, if his budget is $500, I could buy this chip for $135, mobo $50, RAM $35, Case $50, HDD $50, PSU $60, and a copy of Windows 7 for a total of $480, and he'd be able to play most games at medium - high settings on a 1080p HDTV, with playable framerates (in the 30's)?
That way he could save the $200 he'd need for a 6870 or something?
Also, if they're doing this, I don't see the point in buying dedicated graphics cards for the lower mid-range (usually around $80 - $120) anymore? So wouldn't this chip effectively make all those cards useless, due to the fact that you get a quad core CPU and the power of one of those $80-$120 dedicated cards for $135, instead of $250+?
The PC you have described here is exactly the market segment Llano is aiming at: casual gaming with a decent level of graphics and incredible capabilities for HTPCs. As such, it is a good build.
The lower mid-range cards are still going to be present for systems without integrated graphics, or for those with too weak ones.
Also, one thing I haven't talked about yet, as there is so much to say, is the dual graphics capability; one can add a dedicated graphics card and run both together, much like Hybrid Crossfire did with previous chipsets. The compatible cards are the lower-end segment, up to the Radeon HD 6670. I am going to add a word about it in the review, and ultimately I would like to test that out in a separate article.
Yes, on top of that you could see it easily tangling with the Phenom II 980 without much trouble at a lower clock speed even.
I would say yes, the money spent on a lower end card unless it's a previous generation would probably be wasted, there are exceptions. I know for a while they were blowing out 5850's and such for around $119.
Seeing the same benchmarks but the AMD meeting at E3 leaves me to believe yes they're aiming for exactly that, out of the box playable gaming without needing to strap in a mid-high range card to the cost as well.
The whole pricing structure is due for a shift at this point and they're stacking cards behind the APU as the new standard of basic computing. If we looked at the volume sales of the 5450 and such I have doubts that the highest selling market would be the mass consumers, chances are volume sales to OEM's would be on the list. With that said they're just reducing the build / material cost to the companies.
I doubt we'll see them kill the value market entirely but they do want to showcase how high processing power can be combined with playable video power in a single package.
I've honestly heard nothing on this or the actual release of it, the only stuff I've heard was speculation on the APU and how it would work (which got me very excited).
If this is all the case, I don't see why it wouldn't become very mainstream, and a way for big name companies to market this heavily saying that for such a price point, it can play games (like Sims 3, WoW, etc. for the mainstream casual market).
I mean, I have a few people interested in me building them some new computers, because theirs are getting old, with budgets of around $400 - $500... One is a console gamer who I've been trying to get into PC gaming, and the other is a casual gamer who plays WoW, Sims 3, and things of that nature. I would most definitely buy this for these people, and even for the other 3 who don't play games, just in case they want to in the future, or want to try some games.
This just seems very revolutionary for casual gamers (which make up majority of gamers), and the only time I've even heard of its release or really anything on it has been on Neoseeker, today.
Actually a lot of the thunder got taken for Sandy Bridge a few months back :/ as a revolution to how a single chip could do graphics and processing to reduce the need for a dedicated GPU. The lack of inclusion of course is a downer in this case because the market pricing is radically different. I know the SB chips are floating high and the A8-3850 is killing the 980 for less and probably able to dance with some of the lower end SB chips in CPU power and probably dominating on GPU power.
This is pretty much aimed right at those users, it takes the steps out of building and the need to pick based on what the rep at Bestbuy or Frys might be pushing on you. The FX line coming soon will address the hardcore market in a similar manner. The guy I spoke with noted they're looking with 32nm to push an 8 core out that is able to use AOD/ACC to deactivate 4 cores and push the remaining 4 up to 4.5Ghz or so.
Unfortunately our hardware news is pretty flat at this point, not sure if Sean (Chautemoc) might be up for helping get someone back on top of that area. After Kevin we had some trouble getting a steady hardware news guy in there and eventually it came to a halt.
So I have to change the reference clock and adjust the multipliers of others.
Right now CPU is stable 3600Mhz using Overdrive.
Ok I will try those multipliers thank you have a nice day
I changed my Ram frequency in BIOS from 1333 to 1600 and now CPU Z is showing 800Mhz before it was around 666 Mhz I don't think its real value so I am download 3D mark now.