Kaveri review with meaningful data

Level1Techs INBOX.deb
1

http://forums.overclockers.co.uk/showthread.php?t=18572557

 

Picked up a 7850k this morning and an FM2+ board and some ram etc etc. So far, mixed results. I looked on Youtube and couldn't find any reviews, save general chat, and then there is a lot of stuff on forums which is just either AMD hate or an overview. Even the Anandtech review is pretty poor. The Bit Tech review is (as usual) poor too. Some guy on the OCUK (a forum I would urge anyone to avoid like the plague usually) has done an interesting little test which most of the results tally up with my experiences comparing my 7850 to the 6800k I have.

 

So, hopefully someone will find these results useful..

2

Hardware Canucks has a review up.

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/65031-amd-kaveri-a10-7850k-a8-7600-review.html

It's a disappointing read and seems like the 7850K isn't much better than the 6800K in most situations, even the power consumption hasn't improved despite the newer 28nm manufacturing process.

3

Glad i went with the i5 now lol

4

When the 5800K came out it was noted that the 3870K did some stuff better. I think i've found that the 6800k does some things better then this 7850k too. If Mantle brings the 20% gains it could be a good upgrade. However seeing as it is still pushing under 30mb/s bandwidth it is likely to be bottleknecked. Getting a 20% gain on a discreet 7850/7870 that has 175mb/s or similar bandwidth is a whole different ball game.

5

your build is way better like 2x the performance of the new 7850k performance wise the gpu is like a under clocked 7750. 

6

routinely outpacing the A10 6800K by a wide margin

I did not see a single test where the 7850k beat the 6800k by more than ~2%, yet this is part of their conclusion.

(for the hardware canucks review)

7

gona have to wait for Fedora 21

8

Very true, I can't see these "wide margins" anywhere. The biggest improvement I found was Battlefield 4 had a 5 frame improvement over the 6800K (whoopie  -_-). In almost all of the productivity tests and the synthetic benchmarks, the 6800K beat or narrowly lost to the 7850K.

This is another incremental update and Steamroller, appears to be, nearly no better than Richland.

9

I can't get my setup to play BF4 at 1080 well at all. I've got 2400 speed ram...this link is fairly interesting too :

 http://semiaccurate.com/2014/01/15/exploring-effect-memory-bandwidth-amds-kaveri/

 Settings wise, everything is LOW, no SSAO or any of that stuff, that's all off. Weapon DOF is OFF.

No AA pre or post or whatever. Everything is as low as a snakes belly, res is of course 1920x1080.

I'm running Win 8.1 with this driver : http://support.amd.com/en-us/kb-articles/Pages/AMDCatalyst13-11APUBetaDriver.aspx

 

China rising is unplayable even if you one of two guys on a 16 man server, it never goes over 32 and hangs about at 22-25 most of the time. 

The main game, most maps are also unplayable, too many FPS drops, particularly explosions. Golmun Railyway I was getting 30-35 most of the time as long as I stayed in a 'Choppa. As soon as I ran about on the ground it was dropping frames like anything.

It does play at 1280x720 fine runs at about 44-46fps average, looks like crap on a 1080 monitor. Memory usage on LOW at 1080 was going between 550mb and 650mb according to GPU-Z as Afterburner/Riva won't work in BF4, in fact the Beta 3 just crashes the game.

My PSU is a Corsair 500w and the disk is an EVO 128GB Samsung. RAm is COrsair 2400 Dominator and Mobo is Gigabyte A88X Sniper.

The reason i'm griping is they were making its ability to play BF4 at 1080 a feature. I have a pc with a six core xeon and an R270 this was merely an experiment for a new PC for my son, who will be playing Minecraft and Spore and who has no interest in BF4.

The CPU does play Skyrim quite well, but no real upgrade over a 6800K.

 

 

10

Hoped for a bigger improvement in the CPU part. Slower clock speed but same performance aint what I was wanting.

11

So you are running the 7850k?

A few questions: What is the latency of that 2400mhz ram? (CL_) If the latency is high, performance will suffer.

How much ram have you allocated to be used by the GPU? You may be running out of memory for the GPU.

How much ram does the system have in total?

No other programs or apps running in the background? (Chrome tabs are memory hogs)

12

Ram is this : http://www.scan.co.uk/products/8gb-(2x4gb)-corsair-ddr3-dominator-platinum-pc3-19200-(2400)-non-ecc-unbuffered-cas-10-12-12-31-dhx-   I've had it a while though, 8GB total. I've tried some other ram (see below) to make sure the ram isn't playing up and it was slower (the ram was 2133),

Ram allocation is 1024mb in bios. 

I didn't have anything running, I even alt'tabbed out to close firefox which Origin battlelog uses.

Also this is an optimised install of WIndows with un-necessary services disabled.

Without sounding like a know it all.. there was nothing to jeopardise the fps like you suggest could be possibility.

I have a Richland system to comare it to back to back, except that is running an MSI A85 board with 2133 Team ram. I also added an Asus DDR5 6670 to that setup and it still kicks the crap out of the Kalveri system in everything...

In my testing a single 6670 with 480 'old' cores (whatever they called the 6XXX architecture) kicks the living shit out of Kalveri, and I know why...because its DDR5 can offer 65mb/s. As for a 7850k offering the same performance as a 7750...I really doubt it.

Even with Mantle the Kalveri will still only have 30 ish mb/s bandwidth.

 

13

That ram timing isn't fantastic, but it's also not horrible either. Certainly not enough to cause that much of a performance hit. I'm really curious to find out why you can't replicate the performance they demonstrated running the same game (unoptimized/non-mantle version) at the same resolution and settings. 

What do your CPU temps/frequency and voltage under full load look like while gaming?

 

 

14

Well it can't replicate the memory bandwidth. When the Nvidia GT640 came out, they were making lots of noise about how it had, 384 shaders so was 'twice' as fast as a last gen mid range cards with 192 like the 550. But it had poxy DDR3 with a bandwidth of 28mb/s. Even if the Kaveri GPU portion of the chip has 1280 shaders like a PS4 chip or 7870, it would be nothing like either, as they are both using DDR5. Seems nobody is paying any attention to the 800lb Gorilla in the room, the crappy memory bandwidth. 

 

Short of having FM2+ motherboards with 1gb of DDR5 soldered to them for graphics use...I can't see any APU being that great for a long time.

 

15

"even Anandtech". Seriously?

16

The problem (in my eyes) is that the processor is based around HSA and there isn't really anything that uses it yet. Fedora 21, like anarkist said, will have HSA included and I think that will allow for a better comparison of how this processor is SUPPOSED to work rather than how it is working currently. They got the processor out there to steer the market toward making HSA compliant things because they want that to be the future of computing, right now that HSA compliance is limited and thus we don't see what the processor is said to do.

Conclusion: I wouldn't count Kaveri or Kabini out yet.

17

AMD must be aware of the memory bandwidth bottle neck though and know they can only take this so far with DDR3. Wondering if this next wave of APU's will just be a stop-gap until DDR4 arrives. 

None-the-less, for those on a tight budget or want a very slim/compact machine for some light to moderate gaming or HTPC use, these do have a place on the market. 

 

18

If Kaveri laptops use less power for similar performance to the best Richland the 5757m and Mantle brings a noticeable boost to games AND the price drops on desktop parts...it is certainly worth it. 

For those that may be interested. Kaveri whilst not great in BF4, outpaces 6800k Richland in Saints Row and Sniper V2 by 8-9 fps at 1080 on low/medium.

19

That was one of their main goals with Kaveri; to deliver the same performance at lower power usage. In that regard, they certainly succeeded.

I was quite impressed with the A8-7600 and the fact that it delivers performance not far behind the 6800k and 7850k while only drawing 65watts. Heck, even it's performance at 45watts was pretty darn good, all things considered.

Watching closely for it's release later in Q1. If the price is right, it will most likely be the brain of my budget slim steambox/HTPC build. For what it delivers, there's really nothing better on the market (currently).

20

Not sure what you mean/think of Anandtech. People seem to think they do good reviews. Seems to me they have an Apple and Intel bias.