AMD Athlon 5350 And AM1 Platform Review: Kabini In A Socket

[__Miguel_]

idle power is dependent more on the platform, bios and o.s. even if the cpus/socs have power saving tech built in. most motherboards disable some of the power saving options for wider range of compatibility. reviewers often test platforms at default settings (leaving optimizations to vendors) which doesn't often show how low/high a customized setups can go.

Good point.

Still, I find it hard to believe that, unless vendors disable stuff like low power states, the new (platform-independent) CPU power gating techniques that have appeared since 2008, combined with smaller lithography, should be able to push idle power draw down, right?

I wouldn't expect much better results than the massively undervolted and optimized platforms from that time (I remember seeing reports of sub-20W setups), but I mean, it has been 6 years of supposed progress on power efficiency...

 

[Jaroslav Jandek]

Still, I find it hard to believe that, unless vendors disable stuff like low power states, the new (platform-independent) CPU power gating techniques that have appeared since 2008, combined with smaller lithography, should be able to push idle power draw down, right?

They do disable low power states by default, because some PSUs freak out and reset when the system enters a lower power state (probably because UVP/UCP).

I wouldn't expect much better results than the massively undervolted and optimized platforms from that time (I remember seeing reports of sub-20W setups), but I mean, it has been 6 years of supposed progress on power efficiency...

The PCs (i3, i5, i7) at work consume about 8-19W (wall socket) when idle (Away Mode in Windows 7/8). The CPU itself goes to a ~2W TDP (yes, even the i7-4770K: all cores shutdown - except 1 that just underclocks to several hundred MHz). The rest of the power draw is caused by other subsystems (like the HDD, network, MB, input devices, ...) and PSU inefficiency at low power.

 

[Isaiah4110]

Still, I find it hard to believe that, unless vendors disable stuff like low power states, the new (platform-independent) CPU power gating techniques that have appeared since 2008, combined with smaller lithography, should be able to push idle power draw down, right?

They do disable low power states by default, because some PSUs freak out and reset when the system enters a lower power state (probably because UVP/UCP).

I wouldn't expect much better results than the massively undervolted and optimized platforms from that time (I remember seeing reports of sub-20W setups), but I mean, it has been 6 years of supposed progress on power efficiency...

The PCs (i3, i5, i7) at work consume about 8-19W (wall socket) when idle (Away Mode in Windows 7/8). The CPU itself goes to a ~2W TDP (yes, even the i7-4770K: all cores shutdown - except 1 that just underclocks to several hundred MHz). The rest of the power draw is caused by other subsystems (like the HDD, network, MB, input devices, ...) and PSU inefficiency at low power.

Just imagine how low that power consumption could become with just a SSD and no HDD in the system.

 

[Avro Arrow]

There seems to be a small error in your bar graph for Adobe Photoshop CC. The graph measures time in minutes and seconds but it says "higher is better" which seems wrong somehow. lol

 

[mikat]

Well... streaming boxes for example don't have hard disks. I don't see the memory differentiating them because they both need it. If the box has less, so will the Kabini equivalent. Actually the new Amazon FireTV has a good amount of memory. Operating system? All streaming boxes have something so why does this rule out Kabini? So if a company wants a way more capable set top or streaming box, Kabini could be a way to go. It will cost a little more for more performance and flexibility (more PC-like).

I don't agree.

Hard disks... if you're honing in on streaming media box exclusively, we're talking about a very specific use case here. You can run the OS off of a USB, sure, but you're limiting the system in pretty much every other scenario. And don't forget, you could run a streaming OS off of an FM2+ platform, too.

But that's getting off topic a bit. The real point is that the motherboard and CPU are not the entire expense in a system. Price out a cheap machine and you will find that the other components are a significant percentage of cost.

So we're not talking about $60 motherboard & CPU vs $110 motherboard and CPU, which sounds like a huge savings... we're talking $450 system vs $490 system. Know what I mean?

I see what you mean, but I don't think you don't understand what I mean. I still don't see the differentiation. Why can't Kabini be used in new purpose built devices on the higher end? You say to price out a cheap machine - well that isn't the point. I can't price it out as if I were Amazon or Microsoft or Sony. "we're talking $450 system vs $490 system. Know what I mean?" --- well what does the Xbox One cost? PS4? What CPU is in there? Pretty similar? It's a custom design too. And those systems have way more bells and whistles than something else! Take a big co, have them set a spec like they do, take out everything you don't want depending on the device you're going for compared to XBox One/PS4 (bluray/dvd playback, fancy OS, less 3d gaming video power, connectivity options, input controllers, local storage?, game dev-game server-cloud subsidies), and add in big volume pricing and you'll have something that's pretty affordable and just a little pricier than the other boxes out there with a commensurate improvement in performance. So why not $100 system (Amazon FireTV?) vs $200-$250 system (that can play some low end x86 games too if you keep a bit of the storage or use USB flash drives)? I'm guessing on the price but you get the idea. It will be much less than Xbox One/PS4! (If MS subsidizes it, then someone else can subsidize device X so that is also a red herring.)

That's my point. Why is it different? It fits into a different price and performance spot. I don't see why someone has to build a PC like a retail consumer and say now it costs $450, oh noes it's not as good as an Xbox One.

Remember my quotes:

"It's going after a new pseudo-desktop arena referred to as "PC-like devices". This is the battleground where Android-equipped set-top boxes and media players are taking pieces of the traditional desktop machine's pie."

Ok cool

"A PC is so much more than just a CPU and motherboard, though. The rest of the components, such as memory, hard disks, and an operating system, already make up much of a budget machine's price tag. So I'm skeptical of this platform's ability to reclaim ground for the PC."

Whaaaa? Now it can only be a PC? Strawman.

Oh, and I agree with the other posters - getting a power supply that fits the usage makes a big difference. Efficiency isn't constant per draw and the 80 PLUS spec doesn't cover a broad range. Differences will appear less than they should in your tests if you use a huge one for no reason.

 

[HomeSkillenSlice]

this could be useful for a htpc ... but with boards that max out at 4 sata-ports and often only one pci-e port its just no use. 1x blu-ray, 1x system 3x storage (raid5) ... you run out of ports ... same with pcie if you use a dvb-decodercard and a ci-module you need at least two ports, so even in the best case, where you have those ports you have no expansion to solve the sata-port problem through extension-cards ...

would have loved to build a low-power system like this but as is i will fall back to core i5 with 35 watt and 1150 socket. for those you at least get nice boards.

those systems just dont cut it ...

WOOAAHHH WOAAAH WOAHHH !!!!!!!!!!! Your talking about a completely different usage scenario tho. Who the hell buys an APU like this and expects i5 performance and all those other ports (many of which are irrelevant to most people). Raid configurations are EXPENSIVE , this platform is not. So who would run it with this system.

This is a mainstream platform with efficiency and power consumption in mind, not all out performance.

 

[fteoOpty6488]

"Socket FM2+ is a vastly superior vehicle from a performance perspective, and scales many orders of magnitude higher than AM1."

This closing statement is simply NOT true!. As a quad 5350 is almost exactly HALF the performance of a Kaveri 7850 flagship!. This means the frequency scaling is very linear compared to Steamroller+GCN with HSA. So the Jaguar cores are pretty solid for their power envelop.
Note: An "order of magnitute" means a factor of 10 by definition!. These Kabinis are NOT slow as it might be perceived. So surely a test with Kaveri, Richland and Trinity chips should put the rekative performances in perspective.

 

[Onus]

I have noticed that all the AM1 boards have x4 PCIe slots, not the full x16. This really cries out for some testing. In an ultra low power-enforced environment (e.g. a solar-based system) where every watt counts, would pairing this with a HD7750 or GTX750 actually allow reasonable gaming on at least medium settings?

 

[__Miguel_]

I have noticed that all the AM1 boards have x4 PCIe slots, not the full x16. This really cries out for some testing. In an ultra low power-enforced environment (e.g. a solar-based system) where every watt counts, would pairing this with a HD7750 or GTX750 actually allow reasonable gaming on at least medium settings?

Phoronix has done quite a few articles on AM1 boards, you might want to check up on those.

Also, either one of those articles covered add-on graphics, or I believe a link on one of its forum threads (not sure on which one, though) had a link to yet another review, which showed that an NVIDIA GTX750 was about the maximum you could get out of the platform, with a reasonable added performance over the base configuration. The GTX750 Ti was only marginally better than the GTX750.

Hope this helps.

 

[HomeSkillenSlice]

I have noticed that all the AM1 boards have x4 PCIe slots, not the full x16. This really cries out for some testing. In an ultra low power-enforced environment (e.g. a solar-based system) where every watt counts, would pairing this with a HD7750 or GTX750 actually allow reasonable gaming on at least medium settings?

Thats exactly what I'm wondering. Luckily for me and u PCPER just did a test a couple of days ago with a 750 Ti and the results look phenomenal , which is also why im having trouble believing the results . I may just get these parts because of this test

-------> http://www.pcper.com/reviews/General-Tech/AMD-AM1-Platform-and-Athlon-5350-GTX-750-Ti-1080p-under-450

Theres also the fact that this a quad core and its not modular either so this may perform quite well in future titles.

EDIT : My mistake , I wrote that it was modular, its not. Well from what I know of anyways.

 

[Onus]

Wow, the temptation is quite strong to build one of these now...

 

[de5_Roy]

if you want to build a gaming pc like that, wait for a die shrink and new core for ipc and clockrate to go up within the power limits, as well as more mantle-enabled games and dx12 games. too bad amd is limited by tsmc. puma shoulda been 20nm or smaller this year.

 

[Wisecracker]

My temptation is two fold, the first is what i can shove it in too, my parents have been asking for a custom integrated htpc system for there motor home and second is to see how high this baby will scale with a good cooler.

FlanK3r at XS ... on stock cooler

Raised the system clock 32% stable -- overall wattage up 5w

 

[logainofhades]

This kind of makes me wonder if the overall architecture of these chips is superior to that of Faildozer and its successors. Would love to see one clocked at a similar clock speed of FX.

 

[Onus]

With single-channel memory, I wonder how it scales with RAM speed, especially since it doesn't have L3. Damric has posted some very interesting commentary about the L3 in AMD chips over in the Best Gaming CPUs thread.

 

[ewok93]

It would be fun to overclock one of these as high as possible with a ridiculously powerful cooler, like a phase change one, maybe, see what happens. I'm curious about their overclocking headroom and what kind of performance you'd get. Also, an $800 cooler on a $50 APU would be funny.

 

[thedoginthewok]

Page 6 in the last image there is a slight mistake.
It says "Encoding time bla bla higher is better".

It should say "lower is better".

 

[godsucker]

What I really, really, really do not understand is how on earth a experienced reviewer like Don actually pairs such low power system with a 850W (!!!) PSU?!?!? I mean, what...? I would appreciate a lot if in future reviews such systems would be tested with an appropriate PSU or even with a picoPSU to get the power consumption correctly! The guys at computerbase.de did compare a 75W (LC-Power LC75ITX) with a 400W ATX PSU: with the 75W PSU the power draw was <21W under max. load, while with the ATX PSU it reached 35W (5350) and 25W (J1900) under max. load, i.e. 10W less than what is found on here...

 

[logainofhades]

Because it is the same exact PSU they use for all the reviews. Why go buy something else, if you have what you need on hand?

 

[godsucker]

Because it is the same exact PSU they use for all the reviews. Why go buy something else, if you have what you need on hand?

How about to correctly reflect the properties of a given system? Even if Don says in the comments that the power draw is the same independently on the PSU, the test on computerbase.de does say otherwise.

Check these numbers, all measured under full load for the Athlon 5350:

• 
  ■ 850W: 48W
  ■ 400W ATX: 35W
  ■ 75W PSU: 21W

and when the system is idle:

• 
  ■ 850W: 31W
  ■ 400W ATX: 23W
  ■ 75W PSU: 8W

A difference of over 200% under full load and 400% when idle!!! Even if differences in the measurement method or the mainboard etc were considered it would not make up for such a large difference.

If to somebody power consumption is very important he would not consider these systems if he were reading only the Tom's Hardware review. Therefore, I'd strongly suggest and support to use a different PSU for such low-power systems, because as the name says they are built to get a PC that consumes as less power as possible. Correct me if I'm wrong.

 

[Onus]

I agree with godsucker. Where low power is the goal of the entire system, the peripheral components must be selected with that goal in mind in order to get meaningful results that translate to actual use.
If the "power budget" is provided by solar panels, for example, those power differences become huge.

 

[__Miguel_]

Check these numbers, all measured under full load for the Athlon 5350:

• 
  ■ 850W: 48W
  ■ 400W ATX: 35W
  ■ 75W PSU: 21W

and when the system is idle:

• 
  ■ 850W: 31W
  ■ 400W ATX: 23W
  ■ 75W PSU: 8W

Oh, wow, that IS a massive difference, indeed. Granted, the 850W PSU setup is probably different from the other two (as well as the way load is induced - CPU/CPU+GPU, and which software, since there can be variants), not to mention measuring points, but a 13W difference is quite large, and even discarding that difference, 14W with the same setup, only differing the PSU, is still huge...

In any case, I believe Don has stated they will look into that.

Thanks for the info, btw.

 

[godsucker]

In any case, I believe Don has stated they will look into that.

Yes, I saw his post. I really do hope that he reconsiders the setup for low-power systems.

 

[kviksand81]

Try viewing the power usage difference in percentage instead of absolute numbers.

Global wattage, CPU+GPU load: ~40% (!) difference

Active idle: ~12% difference

Under any other circumstances I would guess that you would care a lot about 40% higher energy consumption. The attitude here is more like: "Well... I don't care if my car does 10 km/l or 14 km/l" which I would assume that you would care about... But if you consider the performance advantage of the AMD system, it might be worth it... But you don't mention that.

 

[logainofhades]

If I had the TV to take advantage of it, I would probably go with Kabini for an HTPC. I still have a regular TV, so I would have to use my HD 3870 for video, since it is the only card I have on hand with S-video.