I found the review itself quite disappointing. Where are power consumption measurements? Where are comparable discrete-gpu+cpu comparisons? Dual-gpu results?
I agree. They compared the GPU to Intel Iris GPU, but only on certain benchmarks. Seems like the meat of the review was thrown together and they attempt to distract from that with excruciating levels of technical detail.
If you care enough to know what a discrete gpu is, this is probably not the CPU for you. While this might be the best integrated GPU to come around yet, it's not going to offer the performance of a good discrete card.
the benchmarks include comparisions with a Radeon HD6750 which is an older entry level card that performs significantly better. I don't think it would make sense to put the APUs up against even more powerful discrete GPUs as the gap would be huge.
I got a bit confused when he compared the Kaveri to the Intel. Not one of the charts had the Intel CPU even showing up. And at the end, the graphic compares the Kaveri A10-7860 to the i5-4670K whereas the charts were using the i7-4770R.
They were confusing charts, not what you are used to seeing. The charts were normalized to the Intel chip; thus a positive bar for an AMD chip in that graph means relative improvement to the Intel chip, and a negative bar relative loss to the Intel chip.
Confusing, and of course the meanings reverse with "lower-is-better" graphs.
The way I usually see to show normalized performance is normalizing to 100 rather than 0. An example, normalizing on the A10-6700:
This was a lot to read so I tried to find a quote that boiled it down:
"The reality is quite clear by now: AMD isn't going to solve its CPU performance issues with anything from the Bulldozer family. What we need is a replacement architecture, one that I suspect we'll get after Excavator concludes the line in 2015.
"
And from the last paragraph I took that it would be a good CPU for a budget gaming box.
For as much as AMD shows promise in their graphics line, if one thing is clear it is that the Bulldozer architecture has been a humiliating failure on the CPU side: extremely high TDPs and terrible performance vs. equivalent Intel *y Bridge/Haswell cores. It's been universally panned for most workloads since its release date made worse by AMD's marketing efforts leading up to its launch.
I wonder what has happened to the Bulldozer design team at AMD at this point and whether or not they even still work there at all.
I remember reading an article a while back re: changes from hand-drawn manual layout to automated layout methods for the chips. The assertion was that manual layout is often better for compute performance, while automated layout makes more efficient use of die area.
AMD, of course, spun off GlobalFoundries years ago and now pays a 3rd party to fab their processors. They also led the charge to incorporate graphics cores into the same die, and modern GPUs are very transistor-intensive.
Basically, I would not be surprised if this is as much about process as it is about design.
It's good for a small form factor budget gaming box. Kaveri closes the gap but at this point you can still buy a cheap non-APU processor + $100 faster discrete graphics card that will perform better at about the same cost. Even if simply buy the same-ish GPU in discrete form with GDDR5 (~$80) it will perform better than an APU paired up with DDR3. Even more so if you don't pay the premium to get fast DDR3 modules. APUs take a big performance hit when paired up with less expensive / slower DDR3 modules.
One of the prominent features of Kaveri we will be
looking into is its HSA (Heterogenous System
Architecture) – the tight coupling of CPU and GPU,
extending all the way down to the programming model. Gone
are the days when CPU and GPU cores have to be treated
like independent inequals, with tons of data copies back
and forth for both types of cores to cooperate on the
same problem. With Kaveri, both CPU and GPU are treated
as equal class citizens, capable of working on the same
data in the same place in memory. It'll be a while before
we see software take advantage of Kaveri's architecture,
and it's frustrating that the first HSA APU couldn't have
come with a different CPU, but make no mistake: this is a
very big deal.
It's not a reason to buy one for a gaming box today, but the article is right: this architectural change is a very big deal.
Love Anandtech, clutter free straight to the point review with the stuff that matters.
Boils down to: Kaveri itself is a nice evolution from the previous APUs and offers healthy performance gains and the lower end model matches performance of previous top APUs at lower power consumption. CPU performance is largely unchanged and they wont be able to challenge Intel with the Bulldozer architecture which will be around until 2015.
Its a very nice APU for Desktops that need some gaming/gpu chops, but nothing revolutionary as the Intel CPUs with integrated Iris Pro are in the same league when it comes to GPU performance and even outperform the Kaveri in many benchmarks.
HSA (Heterogenous System Architecture) has a lot of potential and could be a model for the future, if that works out with the market adopting remains to be seen though.
From the looks of AMD, Moore's law hasn't just ended it's reversed: CPUs are getting worse every year.
I bought a Thuban Phenom II X4 and an hd5750 about five years ago. If I had known then that AMD would refuse ever again to sell me a CPU with IPC that good, I would have sprung for an X6. In other words AMD can't compete with it's own 5 year old products. An 1100T now sells for more on ebay than a new Vishera; apparently the market has decided that it's more valuable.
Anand's reviewers (Cutress and Garg) almost admit as much, saying "Kaveri is just another iteration of AMD’s APU line up that focuses purely on the integrated graphics side of things, while slowly improving the CPU side back to Thuban levels of performance."
It seems like Kaveri, and AMDs previous APUs in general, are perversely configured so that they don't make sense for any usage case.
--The marketing pushes desktop gaming, but it's worse than most low end GPUs (and old GPUs too).
--It's not an upgrade to my five year old desktop CPU. It's probably worse at lightly threaded performance then similarly priced intel products which are, curiously, not included in this review...
--Apparently most mobile Kaveri will have a 35 watt TDP. If it's turns out like mobile Richland, you'll be lucky to find one anyway. There are 1597 laptops on newegg right now, 9 have high end A10s, and they are all huge.
If they die shrunk Thuban, I would buy it. If they put more cores on Kaveri, I would buy that. If someone would put an A10 in a laptop that wasn't a huge, cheap, POS, I might even buy that. But so far they won't.
So who is this for?
Mobile might be appealing if we care more about gaming than work or battery life, but at this point it's not clear how available it will be. As a programmer, I like the idea of unified memory but that's not a selling point for customers. A friend of mine has a laptop with a recent APU, and she just figured out that it's too slow for minecraft. Telling her about hUMA did not make the sad face go away, true story.
The only place APUs looks good is for desktop gamers who don't have a desktop already, and can't afford better, even by a few measly bucks. I'm not even sure that they wouldn't be better off with a Haswell Pentium and and $50 GPU for the same price, but Anandtech won't tell us...
Did no one else notice that Anandtech's review is posted in the site's "AMD Center", and is "presented by AMD"? I call shenanigans. And people in this thread are calling Anandtech one of the best hardware sites on the the net... Now I'm depressed.
i hear you..i still own a phenom X4 3.4Ghz and will just purchase a $150 GPU to replace the old HD5000 series and it will be much much faster than any of those APUs for the same price.
21 comments
[ 2.6 ms ] story [ 57.9 ms ] thread"AMD launches Kaveri processors aimed at starting a computing revolution (venturebeat.com)"
https://news.ycombinator.com/item?id=7057733
Confusing, and of course the meanings reverse with "lower-is-better" graphs.
The way I usually see to show normalized performance is normalizing to 100 rather than 0. An example, normalizing on the A10-6700:
http://media.bestofmicro.com/M/1/387001/original/Average-per...
"The reality is quite clear by now: AMD isn't going to solve its CPU performance issues with anything from the Bulldozer family. What we need is a replacement architecture, one that I suspect we'll get after Excavator concludes the line in 2015. "
And from the last paragraph I took that it would be a good CPU for a budget gaming box.
I wonder what has happened to the Bulldozer design team at AMD at this point and whether or not they even still work there at all.
I remember reading an article a while back re: changes from hand-drawn manual layout to automated layout methods for the chips. The assertion was that manual layout is often better for compute performance, while automated layout makes more efficient use of die area.
AMD, of course, spun off GlobalFoundries years ago and now pays a 3rd party to fab their processors. They also led the charge to incorporate graphics cores into the same die, and modern GPUs are very transistor-intensive.
Basically, I would not be surprised if this is as much about process as it is about design.
Boils down to: Kaveri itself is a nice evolution from the previous APUs and offers healthy performance gains and the lower end model matches performance of previous top APUs at lower power consumption. CPU performance is largely unchanged and they wont be able to challenge Intel with the Bulldozer architecture which will be around until 2015. Its a very nice APU for Desktops that need some gaming/gpu chops, but nothing revolutionary as the Intel CPUs with integrated Iris Pro are in the same league when it comes to GPU performance and even outperform the Kaveri in many benchmarks.
HSA (Heterogenous System Architecture) has a lot of potential and could be a model for the future, if that works out with the market adopting remains to be seen though.
However, it is sort of silly to compare the extremely expensive $400+ (they are expensive to make too) Iris Pro CPUs with Kaveri.
It is much more reasonable to compare CPU and GPU performance to mainstream chips like i5-4570 which have 4600 graphics.
Just wanted to give the perspective that, while the performance is pretty good, its not the fastest on-die gpu.
I bought a Thuban Phenom II X4 and an hd5750 about five years ago. If I had known then that AMD would refuse ever again to sell me a CPU with IPC that good, I would have sprung for an X6. In other words AMD can't compete with it's own 5 year old products. An 1100T now sells for more on ebay than a new Vishera; apparently the market has decided that it's more valuable.
Anand's reviewers (Cutress and Garg) almost admit as much, saying "Kaveri is just another iteration of AMD’s APU line up that focuses purely on the integrated graphics side of things, while slowly improving the CPU side back to Thuban levels of performance."
It seems like Kaveri, and AMDs previous APUs in general, are perversely configured so that they don't make sense for any usage case.
--The marketing pushes desktop gaming, but it's worse than most low end GPUs (and old GPUs too).
--It's not an upgrade to my five year old desktop CPU. It's probably worse at lightly threaded performance then similarly priced intel products which are, curiously, not included in this review...
--Apparently most mobile Kaveri will have a 35 watt TDP. If it's turns out like mobile Richland, you'll be lucky to find one anyway. There are 1597 laptops on newegg right now, 9 have high end A10s, and they are all huge.
If they die shrunk Thuban, I would buy it. If they put more cores on Kaveri, I would buy that. If someone would put an A10 in a laptop that wasn't a huge, cheap, POS, I might even buy that. But so far they won't.
So who is this for?
Mobile might be appealing if we care more about gaming than work or battery life, but at this point it's not clear how available it will be. As a programmer, I like the idea of unified memory but that's not a selling point for customers. A friend of mine has a laptop with a recent APU, and she just figured out that it's too slow for minecraft. Telling her about hUMA did not make the sad face go away, true story.
The only place APUs looks good is for desktop gamers who don't have a desktop already, and can't afford better, even by a few measly bucks. I'm not even sure that they wouldn't be better off with a Haswell Pentium and and $50 GPU for the same price, but Anandtech won't tell us...
Did no one else notice that Anandtech's review is posted in the site's "AMD Center", and is "presented by AMD"? I call shenanigans. And people in this thread are calling Anandtech one of the best hardware sites on the the net... Now I'm depressed.