AMD A10-7850K vs. Intel/AMD CPU/APU Comparison

[chuckula]

On the server side, AMD is updating, if your dream is another high end AMD workstation see what ends up coming out for single and dual socket Opteron systems.

Have you actually ready AMD's own server roadmaps? The only "update" is that in 2015 AMD will relabel the exact same Piledriver based Opterons that it was selling last year. That's it. Oh, and Kaveri is being re-badged as a "server" part FWIW.

If all this stuff about not needing a high-end CPU for games is what you really care about (I care about a whole lot more than just games although I play a few casually), then the i3-4330 that's been available since last year and costs $60-70 less than Kaveri parts that are on the market and is vastly more power efficient has all the CPU power you need too. Sure, its IGP is weaker than Kaveri but... all you need is a cheapo (and upgradeable) video card to easily beat Kaveri... in fact an AMD HD-7750 is one such card.

In another year or two, you can get another sub-$100 GPU as an upgrade that's cheaper than AMD's next-generation $150-$200 APU that will still easily beat whatever AMD can produce in the IGP, and you can keep the (more power efficient) CPU since CPU doesn't matter.

That's AMD's problem, to really promote Kaveri you have to basically ignore other AMD products, so AMD is competing against itself.

In mobile Kaveri could be more successful because the IGP is more important in mobile setups and a lot of mobile parts are still dual core (e.g. 2 Kaveri "modules") anyway. In those setups you can't upgrade the GPU. I'm curious to see what the mobile flavors of Kaveri do, but when there are real quad-core Intel CPUs on the desktop out for the same price as Kaveri, it's silly to think that in the long run you'll be better off with a guaranteed low-end CPU when GPU upgrades are easy and even sub-$100 GPUs are beating Kaveri by a comfortable margin.

[Laser]

Did anybody expect Kaveri to beat Intel performance-wise? It should be clear by now that with their current CPU architecture, they won't beat Intel neither performance- nor efficiency-wise. Intel is at 22nm, while Kaveri is produced in 28nm - not the only, but also a reason. What is really good about Kaveri is the tight integration of GPU and CPU with hUMA. Will software support this? We don't know yet, but AMD's showcase - a spreadsheet for a modified LibreOffice that makes use of OpenCL with hUMA - looks very promising for the right workloads. AMD also added a lot of other stuff into Kaveri, like an updated UVD and VCE or the GCN graphics supporting Mantle.

I'll certainly buy a Kaveri PC in the next weeks, and I know it is kind of a bet: We don't know how these features will look in practice or if software support for them will ever be available on a wide scope. But I think AMD has delivered a piece of very good engineering work, it is just unfortunate that the CPU component is relatively far behind the competition.

Anandtech has a very detailed review about Kaveri's internals at http://anandtech.com/show/7677/amd-k...7600-a10-7850k and I recommend checking it out.

[nightmarex]

Did anybody expect Kaveri to beat Intel performance-wise? It should be clear by now that with their current CPU architecture, they won't beat Intel neither performance- nor efficiency-wise. Intel is at 22nm, while Kaveri is produced in 28nm - not the only, but also a reason. What is really good about Kaveri is the tight integration of GPU and CPU with hUMA. Will software support this? We don't know yet, but AMD's showcase - a spreadsheet for a modified LibreOffice that makes use of OpenCL with hUMA - looks very promising for the right workloads. AMD also added a lot of other stuff into Kaveri, like an updated UVD and VCE or the GCN graphics supporting Mantle.

I'll certainly buy a Kaveri PC in the next weeks, and I know it is kind of a bet: We don't know how these features will look in practice or if software support for them will ever be available on a wide scope. But I think AMD has delivered a piece of very good engineering work, it is just unfortunate that the CPU component is relatively far behind the competition.

Anandtech has a very detailed review about Kaveri's internals at http://anandtech.com/show/7677/amd-k...7600-a10-7850k and I recommend checking it out.

Eh... I think AMD APU's do beat intel's performance wise... just not CPU wise... usually.

The benchmarks i have seen with opencl or HSA are impressive for what they are. I think of AMD apu's as trucks where intel cpu's are cars, yeah cars or usually faster, but lack the utility. Of course unless you got something to throw in that truck it's just wasted gas mileage.

[TemplarGR]

What i would like to see, is how the unified ram architecture of Kaveri simplifies open source driver performance optimization. I suspect it will become easier for radeon developers to optimize apus' performance from now on.

I am currently using an intel [email protected], 4gigs of 1000mhz ram and an ATI 3870. Since i am no hard core gamer anymore, i would love to switch this with an APU, for power efficiency and less noice/space.

Too bad it seems Kaveri won't be it. I don't have much to gain by upgrading to it. HSA isn't here yet, especially for Linux. CPU is only slightly better, gpu is memory bandwidth starved and it shows(could use a quad channel version), and the RadeonSI driver needs work(i don't intent to use catalyst). So, i will be waiting for Carizzo it seems... Or the next, as it seems Carizzo will just be a 65WTDP version of Kaveri with still dual ddr3 memory...

[Ansla]

Thanks but seeing that the 95 W TDP AMD APU has issues beating a 54 W Core i3, I'm pretty sure the 65 W AMD APUs are just going to do worse. I'm just waiting on actual retail prices for the APUs before making a final decision.

Don't be so sure 45W TDP means half the performance of 95W TDP. The only article doing this comparison even on Windows was one from SemiAccurate that shows A8 7600 @45W has 80% of the perfromance of A10 7850K @95W in most tests. That would make it quite competitive against an i3, especialy since most tasks you'll do on a HTPC should be acceleratable by the GPU.

[wargames]

You need to compile programs for the new architecture to leverage the full potential. There's a LibreOffice Calc benchmark somewhere in which the A10-7850K beats the fastest i7 by a large margin when using hUMA (CPU+GPU). Which leads me to the following question: how does one create a program for hUMA on Linux?

[juanrga]

Good article! But this part shocked me:

In terms of the results shown in this article, the processor performance was rather disappointing. In looking at the results the AMD A10-7850K is supposed to be in line with the Intel Core i5 4670K according to AMD's expectations.

What? Everyone was expecting CPU parity with a Sandy Bridge i5 at integer level and failing behind at the floating point level. Sandy has double the FPU resources (16FLOPS/core) and Haswell has four times more (32FLOPS/core) than Steamroller (8FLOPS/core).

The processor performance is as expected (especially when early predictions for 4GHz are corrected to the lower clock of 3.7GHz)

http://juanrga.com/en/AMD-kaveri-benchmark.html

[coder111]

Long road ahead for AMD APUs on Linux

I'm a long time AMD user, but I'm disappointed.

1. 90W TDP Kaveri APUs seem useless. They are not much faster than Richland. On the other hand, 65 or 45W TDP Kaver APUs seem a sweet spot- they achieve similar performance for 2x lower power usage.

2. This APU seems quite pointless on desktops. In laptops, where in 90% of cases there is no discrete GPU, this APU would be good for low-end gaming, and low power usage. On desktop, dedicated CPU + dedicated GPU beats it on price and performance. Well, maybe APU and a Radeon GPU joined in a crossfire configuration would be nice, but crossfire doesn't work on Linux...

3. Situation might change if HSA and Mantle and OpenCL are thrown into the mix. However, they are nowhere near ready. And how do you even use HSA or Mantle on Linux? I doubt they are supported in current Catalyst drivers. Open-source drivers are nowhere near ready to support any of that.

So it's a mid-level CPU with a mediocre integrated GPU, and all of its main advantages and selling points are useless in Linux, especially if you are running open-source drivers... Sad... Maybe it's useful for mid range laptops when mobile versions come out and open-source drivers improve a bit...

[Adarion]

@Phoronix-Michael

Um, I am not really familiar with the intel CPU governors but:

Processor Details- Intel Core i3 2120: Scaling Governor: intel_pstate performance- Intel Core i5 2500K: Scaling Governor: intel_pstate performance- Intel Core i5 3470: Scaling Governor: intel_pstate performance- Intel Core i7 3770K: Scaling Governor: intel_pstate performance- Intel Core i3 4130: Scaling Governor: intel_pstate performance- Intel Core i5 4670: Scaling Governor: intel_pstate performance- Intel Core i7 4770K: Scaling Governor: intel_pstate performance- AMD A10-6800K: Scaling Governor: acpi-cpufreq ondemand- AMD A10-7850K: Scaling Governor: acpi-cpufreq ondemand- AMD A10-7850K - 4.4GHz: Scaling Governor: acpi-cpufreq ondemand

Is that a fair comparison between performance and ondemand driven CPUs/APUs?

PS: Wahoo! My 686th post. Did I win a CPU now?

[Michael]

Why the frack is the Intel CPU running the Radeon R 290 and not the AMD FX-8350 paired with it?

That's only for the CPU tests. If you look at the system table on the graphics result file you see the HD Graphics were used.