![screenshot-[2015-02-18] [15-29-37].jpg](https://4chanarchives.com/image/image-404.png "screenshot-[2015-02-18] [15-29-37].jpg")
Are AMD APUs even relevant anymore /g/?
AMD has given us 28nm kaveri but it still seems like a better idea to just get a separate cpu and gpu.
I know Intel has the technology to make a killer APU but they would probably charge like $1,000 for such a chip so they are out of the question.
>>46623495
Intel BARELY keeps up with AMDs APUs and those chips that do cost ridiculous prices
>>46623495
Yes, they are relevant and pretty good in low budget builds.
The newer A10s can get 60fps in 1600x900 on BF4 in medium, easily over 120fps in CSGO, etc.
>>46623495
They're good in their budget. If i would have to make simple, everyday PC for light usage and some games, a6/a10 would be great idea.
>>46623971
a8*
derp
>>46623860
>2016
>1600x900
>>46623495
It depends on where you buy them, and what you want to use them for. An A10 7850K can routinely be purchased for $110 or $120 at Microcenter, and often with further discounts on a matching FM2+ mobo. Those prices make it untouchable in terms of value. A $72 8GB 2400mhz RAM kit thrown it combined with a little CPU overclock give you a really well rounded system.
The IGP isn't the most powerful thing around, its seriously memory bottlenecked, but it can still handle a ton of games at 1280x720+ on high settings. There are even a lot of titles that'll run at 1080p on them.
>>46623495
Intel considers keeping up with AMD graphically warrants a high premium price tag, they can go fuck themselves
>>46624828
So ddr4 would actually make it competitive?
I couldn't find any information on bandwidth restriction while searching for ddr4.
can and would you elaborate?
>>46623495
It's more efficient to deliver a chip that has all the bells and whistles (i.e a SoC) than it is to separate out all the different parts into separate chips, not to mention the silicon data lanes into the chip are far faster and short than PCIe lanes so latency is reduced and bottlenecks of chip to chip communication are non existent.
Also it makes installing drivers less of a pain when you have the DSP, Southbridge, GPU all in one section and you can download one file and install all the shit you need in one swoop (inb4 AMD has no drivers).
>>46625182
Quad channel to be specific, AMDs next APUs coming in 2016 will either be using HBM, DDR4 or a mix of both, IIRC quad channel DDR4 providdes bandwidth of around 95gb/s which is roughly 4x DDR3 dual channel, ideal since the next APUs will be built on 14nm, allowing for around 3-4x as many transistors on a chip compared to the current 28nm chips.
wait for new apus and ddr4
shit's gonna be so cash
gpus are dead long live the apu
>>46625182
More bandwidth would improve performance.
Dual channel 2400mhz RAM gives you 38.4GB/s memory bandwidth.
Kaveri has 8 GCN CU, similar to the Radeon 7750. The Radeon 7750/7770 have 72GB/s memory bandwidth. The 7750 has 2CU disabled compared to the 7770 so it has a little more bandwidth than it needs here, but this shows roughly how much bandwidth 8CU require to not be bottlenecked.
Quad channel 3000mhz memory gives you 96GB/s memory bandwidth, more than enough to feed 8CU. A quad channel IMC would give enough memory bandwidth whether you were working with DDR3 or DDR4. The kicker here is that a single package of HBM1 can provide you with 128GB/s. They have everything lined up perfectly to facilitate their IGPs becoming ever more powerful in future iterations.
Pic related, DDR3 Radeon 7750 vs a normal GDDR5 7750. The DDR3 variant shows roughly how Kaveri's GPU performs. The normal reference card shows how it could perform with adequate memory bandwidth. Frame rates almost double in a couple cases.
>>46625104
The HD5200 Iris Pro doesn't match the Kaveri A10 7850k in heavy GPU workloads. Its neck and neck in some game benches but its due to the large difference in CPU performance. Intel has a problem with their GPU arch not being fast enough. AMD *had* a problem with memory not being fast enough.
AMD's decision to jump on the HBM wagon is going to result in Intel getting their asses handed to them in the APU market. If AMD can manage a chipset with HBM instead of DDR4 then Intel basically loses the APU performance market thanks to memory bandwidth.
>>46626037
They'll definitely be dominating when it comes to integrated graphics, theres no question about it.
https://asc.llnl.gov/fastforward/AMD-FF.pdf
A single package of HBM1 can be as high in density as 1GB, and provice 128GB/s bandiwdth.
HBM2 reaches 8GB, and can provide 256GB/s.
AMD can have IGPs as big and powerful as their process node, thermals, and power consumption allows. Zen based chips will probably be a fucking beast.
>>46626142
This is the first time in a LONG while that I'm optimistic about AMD delivering a great product
>>46626497
GPU performance is in the bag, and they have access to a solid 14nm FinFET process. All they have to do is deliver a CPU core arch that doesn't suck.
I sincerely hope they don't deliver another disappointment like Bulldozer, but it would be pretty hard to do that. Bulldozer was trying to make a smaller core and clock it higher to deliver more performance per mm2. They won't be going with another stupid 2ALU wide core again.
>>46626142
Not just integrated graphics, anon. There's little doubt in my mind with that kind of memory bandwidth AMD will make their cards able to utilize the memory on the motherboard too. Benchmarks will get interesting when you can suddenly give a mid range card an extra 8 gigs of memory at that kind of speed.
>>46626589
Its going to be a huge uplift for UHD gaming, and VR if it ever takes off commercially, I'd imagine. Even more impressive if you take that new color compression into account, thats an awful lot of effective bandwidth and memory for high res textures.
>>46623495
>Are AMD APUs even relevant anymore /g/?
When did they stop being?
AMD are doing the right thing by throwing in to APUs. Their performance may win over many OEMs who see an opportunity to churn out cheap "Gaming" laptops or desktops via an APU solution.
Don't misunderstand when I say
>"Gaming"
The use of the quotes was not meant to imply that APUs are incapable of such things. Infact, many are good at it. Namely A8 and A10 chips. Which feature vastly similar iGPUs and offer notably better results in games than the iGPU in A6s and lower.
The current generations of APUs are heavily bandwidth starved, though this is being worked on, specifically with HBM, and I suppose you can throw DDR4 in there too.
With HBM, AMD may chose to get manufacturers onboard to produce motherboards with a new port. This could be used as input for HBM modules, or this could be done through PCIe.
I myself think this idea is very far fetched, but it could happen. And it would be a solution to the bandwidth issue we see with APUs.
DDR4 will be supported by the next generation of APUs according to AMD's roadmap. So there will finally be some demand for that on the market, which may prompt more companies to start adopting it.
The first products out with a new technology can get the idiots onboard the bandwagon and buying fast. Imagine the ads.
But seriously, APUs hold their own. And should more OEMs adopt them in anything more than A4 powered laptops I think people will finally start to associate AMD with a good product again (Outside of Tech-associated sites)
the future's lookin' bright
>>46626578
>I sincerely hope they don't deliver another disappointment like Bulldozer, but it would be pretty hard to do that
It would be even more difficult considering that Keller built K12 and Zen from the ground up.
>>46626750
Based Keller.
Lead CPU architect.
>>46624828
Holy fucking shit, it's even cheaper than the price on amazon. FUCK
>>46626768
>Cutting off the other half
How could you.
>Inb4 filename
I had to pull it off some curry nigger site.
>>46626788
Shit I was suppose to capture a larger area, sorry anons. I'm insanely impressed with micro center.
>>46626799
>>46626578
CPU performance can't be any worse than what it has been, With Jim Keller leading the pack I imagine a big jump in CPU performance
>Looking to buy a new motherboard/CPU
>AMD chip that's something like 4GHz for the same price as an Intel chip that's 2.6GHz.
>Look at benchmarks and Intel wins.
Why the fuck does AMD even exist?
I've got a question for you all: If I water cool the APU, can I disable the iGPU, slap on a R9 290X, and overclock the CPU cores to 4.8GHz? Because if I could then I will develop a huge respect for APUs.
>>46624828
Microcenter is cheating. I picked up a 8320 for $100 and a 4770k for $200 there. Those are better values imo.
>>46626887
AMD sells APUs that have the GPU hard disabled as the most recent iterations of the Athlon. I'd have to go find the appropriate threads to see how they clock though, but people are purchasing them and using them explicitly for the CPU part.
>>46626887
If you have a discrete GPU in the system then the IGP won't be used unless its in dual graphics mode that only a couple cards support. Some Kaveri chips can reach around 4.7ghz, most seem to top out around 4.5ghz. Personally I think reviewers just need to spend more time tweaking mobo settings. You don't need liquid cooling to reach those clocks though.
>>46626944
This, a similarly priced air-cooler would perform as well as a water-cooler.
>>46626984
I bought my h100 for $30. I love Frys sometimes.
>>46626880
they exist because otherwise intel would get fucked over by anti-monopoly laws.
if you mean why do they suck, it's because intel conducted massive amounts of illegal activities back when AMD was superior.
intel forced AMD to sell off it's fabrication facilities, costing them something like $15 billion. AMD is years behind as a result, and still continually hurting from having to subcontract production.
intel was caught red-handed for all of this. in the end all they were forced to do was recompense AMD 1 billion. that's like a grade-school psychopath being forced to apologize to the kid he threw off the roof and turned into a vegetable.
>>46626142
That is... small
holy crap
is it usable to replace ddr*?
>>46626683
>Its going to be a huge uplift for UHD gaming, and VR if it ever takes off commercially, I'd imagine. Even more impressive if you take that new color compression into account, thats an awful lot of effective bandwidth and memory for high res textures.
>not posting the original image
>shilling this hard
please anon, you're not even being paid
HOLY SHIT AMD, I WANT ZEN ALREADY.
FUCK YOU
WHY IS THIS SHIT TAKING SO LONG
I CANT WAIT ANYMORE
>>46627254
Not as it is currently. The signals wouldn't carry over long distances from a separate slot/socket to the CPU socket, and CPU sockets would need drastically more pins on the bottom to utilize HBM in that way.
>>46627312
Shitpost elsewhere, child.
>>46626793
old guy with short white hair + young hip guy with long hair
I trust this partnership to do great things.
APUs are about relevant as AMD is.
Which is to say, not at all.
So, where is AMD going to stick this magical HBM, motherboards, directly on chip?
Do they even have room for that?
>>46627349
I'm not holding my breath. When I do a full upgrade next year I'm already counting on it being Intel. My 8350 had a fairly long life, but AMD is sure to disappoint.
>>46627501
Dobberpuhl is more than an old guy with white hair.
>>46627545
High end GPUs first, as they are hugely bandwidth hungry. On top of that HBM will lead to a fairly significant reduction in GPU power draw, as GDDR5 is quite power hungry.
>>46627545
See: >>46626142
Its placed on the same package as the die its feeding.
>>46627545
From looking at (>>46626142)
It seems HBM is entirely on-chip.
>>46627581
As for there being physical space:
pic related is Kaveri. The die size is 245mm2@28nm. If the package were redesigned there would be room for a couple HBM modules under the heat spreader. A smaller 14nm die would leave even more room.
>>46623495
APUs are god tier
>>46623495
The newest A10 is a pretty bad buy however, since the on die graphics are now almost entirely memory bottlenecked.
