The Official Kaveri Review Thread (A10-7850K, etc)

[blastingcap]

We haven't seen mobile Kaveri yet and it's probably not coming before H2 2014.

Ah I see. Thanks. I have not kept up with the news and thought Kaveri *was* mobile. My cheap Pentium-based laptop is over 3 years old now and still runs fine (it's like a modest Core 2 Duo in performance), but I'm missing USB 3.0 and wouldn't mind a graphics boost.

 

[Sweepr]

We haven't seen mobile Kaveri yet and it's probably not coming before H2 2014.

Just in time for Broadwell then.

Anyone else notice the Anandtech review features only gaming evolved titles?

What are the chances of that?

Shh! You might actually ruin an old fallacious argument posting things like this.

 

[Slomo4shO]

Lol, I forgot about that!

Except this time there really isn't any gains at all thanks to the lower core frequencies... :whiste:

Seriously, if all you're going to do is game, this seems like a great option when paired with a high end video card.

The AMD 760K at half the price says Hi. At the price point of the 7850K, it is better to go with an i5 if you are pairing with a high end GPU...

Power seems way too high for a laptop doesn't it?

You base this off what? The desktop power consumption figures?

 

[Shivansps]

The power comsuption seems way high for a 45W part, someone tested it on a case with a low profile 45W cooler?

Its seems way too efficient to consume 90W of power and dissipate just 45W of heat.

 

[monstercameron]

linux is gtg http://www.phoronix.com/scan.php?page=article&item=amd_a10kaveri_linwin&num=1

 

[frozentundra123456]

Seems like a very solid CPU choice for a gaming only PC.

Seriously, if all you're going to do is game, this seems like a great option when paired with a high end video card.

:shrug:

Actually, if you are using a high end gpu, Kaveri seems like one of the *worst* choices you could possibly make for gaming.

 

[Lonbjerg]

Seems like a very solid CPU choice for a gaming only PC.

Seriously, if all you're going to do is game, this seems like a great option when paired with a high end video card.

:shrug:

Why would I choose a CPU that woukld CPU limit me in most games?

 

[blackened23]

Yeah. It would be pretty stupid to pair an APU with a high end video card. Downright stupid in fact.

Kaveri with discrete GPU benchmarks:

The APUs don't have enough horsepower to keep a high end discrete GPU fully fed. If you're blowing 500$+ on a discrete card, going cheap on the CPU makes pretty much...zero sense. Besides which, a Haswell i5 is not expensive. Especially not if you're already spending 500$+ on a GPU.

 

[GRAFiZ]

Except this time there really isn't any gains at all thanks to the lower core frequencies... :whiste:



The AMD 760K at half the price says Hi. At the price point of the 7850K, it is better to go with an i5 if you are pairing with a high end GPU...


You base this off what? The desktop power consumption figures?

Actually, if you are using a high end gpu, Kaveri seems like one of the *worst* choices you could possibly make for gaming.

Why would I choose a CPU that woukld CPU limit me in most games?

I just wanted to be different.

I just wanted AMD to make me happy here.

 

[Ventanni]

I'm not impressed. I'm usually always the individual who thinks the glass is half full and focuses on the positives of a product, but I'm just not impressed with Kaveri. The GPU performance is a nice step up, no doubt, but in reality it's really the CPU performance that *has* to improve, or else the product is a wash. After all, if your GPU performance is insufficient, you can always go and add a dGPU. Yes, it costs additional money, but the option is still there.

I was really expecting 15-20% IPC increase with Steamroller given the amount of pre-release content we saw. With a +30% increase in this and a +20% increase in that, I was honestly unimpressed when the final numbers pointed more towards ~7.5% overall IPC increase. Here's to hoping Mr. Keller (is that his name?) is leading the team to take the Cat cores to a new level of awesome.

AMD - Never missing an opportunity to miss and opportunity.

 

[Slomo4shO]

I just wanted to be different.

I just wanted AMD to make me happy here.

Rewarding bad behavior leads only to more bad behavior...

 

[TeknoBug]

With an i3 in it that does everything they want it it to, is faster for work/office/home, and costs less than Kaveri.

This is Derpdozer all over again, complete with all the broken performance promises - albeit without the monster power consumption. A three year old Athlon is just as fast of a CPU. Maybe a five year old Athlon is too.

History continues to repeat itself, even the moar cores mantra carried over - Kaveri has 12 cores!

What's the expected pricetag for the Kaveri? I noticed the prices of i3 has gone up, at least the Haswell ones.

 

[JDG1980]

I was really expecting 15-20% IPC increase with Steamroller given the amount of pre-release content we saw. With a +30% increase in this and a +20% increase in that, I was honestly unimpressed when the final numbers pointed more towards ~7.5% overall IPC increase. Here's to hoping Mr. Keller (is that his name?) is leading the team to take the Cat cores to a new level of awesome.

Agreed. When Bulldozer came out, it could be dismissed as growing pains - a new architecture with lots of room for improvement yet to come. Piledriver consisted of relatively small tweaks, so getting single-digit IPC improvements wasn't that unexpected. But Steamroller was supposed to be BD/PD done right - they finally got rid of the shared decoder, and had time to make plenty of other tweaks and incremental improvements as needed. And the results are still underwhelming: probably no better than Core 2 Duo in terms of floating-point IPC, and almost certainly inferior to Nehalem in integer IPC. Given that Sandy Bridge is coming up on its 3rd birthday, this is just not acceptable any more. AMD really needs to get IPC up, and it's increasingly looking like there is no way to do that with the construction equipment cores, at least within their current constraints (limited budget and fab technology that isn't up to speed with Intel's). Maybe the cat cores will be the way forward - how much could those be souped-up if the execution units were widened, more cache added, and the branch prediction lessons learned from BD/PD/SR incorporated?

AMD doesn't have to roll over and die. Intel did Sandy Bridge on 32nm, and improvements since then have been fairly marginal; if AMD could create a chip that good, there would still be plenty of market for it. But the construction equipment line is looking problematic on all kinds of levels: it can't compete with Intel in terms of IPC, it can't compete with Intel in terms of power efficiency, and it uses a lot of die space. Intel had the same problems with Netburst, which had terrible IPC, heat problems, and was too inefficient for mobile applications. The solution was to beef up the mobile Pentium-M and make this their mainstream CPU, which they did with Core 2 Duo. AMD is facing the same problem, and it seems to me that they may want to look to the same solution.

 

[ShintaiDK]

What's the expected pricetag for the Kaveri? I noticed the prices of i3 has gone up, at least the Haswell ones.

Haswell i3s cost the same as always. The prices are listed in the Anandtech review.

 

[frozentundra123456]

So does anyone know if the rumors of BF4 bundled with Kaveri are true? I havent heard anymore about it. If it were, and you were interested in that game (which I am not, but a lot are of course) that would make the effective price decent, assuming also that you dont have the game.

Otherwise, I would expect the price to fall rapidly on the highest end model.

 

[Lonbjerg]

Agreed. When Bulldozer came out, it could be dismissed as growing pains - a new architecture with lots of room for improvement yet to come. Piledriver consisted of relatively small tweaks, so getting single-digit IPC improvements wasn't that unexpected. But Steamroller was supposed to be BD/PD done right - they finally got rid of the shared decoder, and had time to make plenty of other tweaks and incremental improvements as needed. And the results are still underwhelming: probably no better than Core 2 Duo in terms of floating-point IPC, and almost certainly inferior to Nehalem in integer IPC. Given that Sandy Bridge is coming up on its 3rd birthday, this is just not acceptable any more. AMD really needs to get IPC up, and it's increasingly looking like there is no way to do that with the construction equipment cores, at least within their current constraints (limited budget and fab technology that isn't up to speed with Intel's). Maybe the cat cores will be the way forward - how much could those be souped-up if the execution units were widened, more cache added, and the branch prediction lessons learned from BD/PD/SR incorporated?

AMD doesn't have to roll over and die. Intel did Sandy Bridge on 32nm, and improvements since then have been fairly marginal; if AMD could create a chip that good, there would still be plenty of market for it. But the construction equipment line is looking problematic on all kinds of levels: it can't compete with Intel in terms of IPC, it can't compete with Intel in terms of power efficiency, and it uses a lot of die space. Intel had the same problems with Netburst, which had terrible IPC, heat problems, and was too inefficient for mobile applications. The solution was to beef up the mobile Pentium-M and make this their mainstream CPU, which they did with Core 2 Duo. AMD is facing the same problem, and it seems to me that they may want to look to the same solution.

Don't leave out Intel's FAB's...and their process advantage...that is not something you overcome overnight

 

[el etro]

The 45 W TDP Kaveri chip seems to be the one that performs best (within its TDP category).

Does this mean we can expect coming mobile 25/15 W TDP Kaveri based chips to perform well too? :hmm:

Yes, we can. But the real doubt is if Kaveri is a step foward in terms of power efficiency or the its better power efficiency relies only on GLO-FO 28nm process.

OEM-1 to AMD = We want low TDP APUs for SFF Desktop
OEM-2 to AMD = We want low TDP APUs for SFF Desktop
OEM-3 to AMD = We want low TDP APUs for SFF Desktop
.
.
.
.
.
.
.
.
OEM-n to AMD = We want low TDP APUs for SFF Desktop

Now look at Newegg all those Intel SFF Desktops. There are 21x Core i3 and 59x Core i5 Slim/SFF Desktops in that page. If you didnt know already, 90%+ of Desktop Users don't debate/fight about Kaveri and Core i3/5 in Hardware forums like AT, they simple get a SFF desktop with integrated GPU for work/office/home. :whiste:

The 45W A8-7600 is the best AMD Product of the last 3-5 years, it may not be the fastest and is not competing in the HighEnd but it has the potential to be the most successful AMD Desktop product so far.

This is the answer to the ones who think that PC is dying...
...But PC industry still needs better and faster answers to the SteamBoxes..

 

[ViRGE]

What are the chances of HSA taking off? Do you guys think Intel will support something similar?

Kaveri looks like garbage to me, but HSA seems interesting. Hopefully it has a future.

I'm not sure if HSA in particular will take off, but I absolutely think that something like HSA will eventually take hold in the market. The ideas behind it are fundamentally sound, both with respect to hardware features and software abstraction. Of all of the ideas presented up until now for making a hetero environment low enough in overhead to be viable and easy(er) to code for, HSA is the first idea that's good enough to work.

 

[know of fence]

According to Hardware.fr review 3.7GHz Phenom II 980 (2009 Deneb core) is still faster than 3.7-4.0GHz A10 7850K (2014 Steamroller core) - both applications and games.

Just barely and in a totally obscure cumulative average. Considering that the Phenom II was a 130W part without any graphics, at that level having an APU less that half of which is dedicated to CPU cores, is actually quite impressive.

AMD is obviously moving sideways, they finally caught up with themselves in regards to GCN, they committed to a lot of innovative stuff, they rule the consoles, you don't need a vice or a razor to cool them! For anyone without a graphics card, Kaveri is a safe recommendation.

With all the discussion, about who APUs are for? Now, maybe there are people out there who drop more than a grand on a gaming PC, to play it 10 hours a week and have the best possible experience. But I suspect, from a time when I used to own graphics cards, there are plenty of blokes who'd want to get a lot of mileage out of the purchase of a pricey discrete GPU card, which in retrospect usually is a bad decision.

Casual gaming can imply staying say below 100W TDP, sticking with shorter, smarter, independent games, instead of chasing the biggest high, be it total Immersion in a Sim, the most spectacular Battlefield stunt kill or the shiniest armor in an RPG. For the first time it's viable, you even can test and run any game and then decide to commit to an allegedly better experience. Computing is stuck in a holding pattern anyway, waiting for mobile to catch up.

People who say a frequency gimped, i3, with pitiful graphics, disabled energy saving states and 20 degees worth of thermal insulation is a better choice than a fully featured, top of the line APU, those people are full of something, but it ain't wisdom. OK, Maybe in an ultra-book an i3 makes sense, or in a office nettop.

The funny thing is that for the longest time it seemed that notebooks were going to replace desktops, but actually AMD is better than ever positioned in the growing HTPC, console segment than say the blasted ultrabooks, or thousand dollar graphics cards in their respective markets.

 

[blastingcap]

I'm not sure if HSA in particular will take off, but I absolutely think that something like HSA will eventually take hold in the market. The ideas behind it are fundamentally sound, both with respect to hardware features and software abstraction. Of all of the ideas presented up until now for making a hetero environment low enough in overhead to be viable and easy(er) to code for, HSA is the first idea that's good enough to work.

I agree. Memory speed is gonna be an issue for HSA in particular, going forward. Not sure if there will be another HSA-like attempt later by intel and/or amd with big on die memcache or something.

 

[el etro]

http://translate.google.com/transla...eview-test,testberichte-241474-9.html&act=url

50% increase in FPS (26--> 39) with 50% increase in memory bandwidth (1600-->2400) for this game at the tested resolution.

Same 50% bigger score goes to 1333-1600Mhz. IMO 1866-2133Mhz is a real deal for the 7850k...

 

[tential]

Is it just me or does kaveri seem good for a cheap gaming laptop? Esp. 1366x788 models?

I think AMD shot themselves in the foot when they mentioned 1080p gaming and kaveri. They should have said 720p gaming and kaveri or left it ambiguous and said HD gaming. I like the 3rd option best leaving it ambiguous if I'm thinking from a marketing perspective.

If I needed to make a gaming rig that was a PC that had to be CHEAP and SFF Kaveri sounds great. AMD though is really disappointing me though. You need to improve SOMETHING SIGNIFICANTLY with a new processor. People cried about Haswell but it WAS a significant improvement in mobile. It did SOMETHING and it did it very well. I'm not seeing a huge improvement at all with Kaveri.
The problem isn't AMD's ideas it's the delivery.

 

[JumBie]

Yeah. It would be pretty stupid to pair an APU with a high end video card. Downright stupid in fact.

Kaveri with discrete GPU benchmarks:

..



The APUs don't have enough horsepower to keep a high end discrete GPU fully fed. If you're blowing 500$+ on a discrete card, going cheap on the CPU makes pretty much...zero sense. Besides which, a Haswell i5 is not expensive. Especially not if you're already spending 500$+ on a GPU.

Yeh, well Kaveri's market isnt the dGPU market, its the HTPC, Grandma, School, casual market. Regardless, AMD isnt viable for anything other than that market anymore, even the FX line is pathetic. No point in looking at AMD to provide us with anything other than great dGPU's because as far as CPU's go Intel has(except the x2 and x4) and will always be the only option when it comes to us gamers.

 

[blastingcap]

Yeh, well Kaveri's market isnt the dGPU market, its the HTPC, Grandma, School, casual market. Regardless, AMD isnt viable for anything other than that market anymore, even the FX line is pathetic. No point in looking at AMD to provide us with anything other than great dGPU's because as far as CPU's go Intel has(except the x2 and x4) and will always be the only option when it comes to us gamers.

I think AMD knows their IPC is lagging, and it's why they are pushing Mantle which effectively does away with CPU bottlenecks supposedly.

 

[Homeles]

The one thing I take for this is that desktop CPUs are now "completed". When looking at what's on the horizon they will simply not provide much more performance going forward.

Both Intel and AMD are bringing us such minuscule performance increases from one generation to the next that it is pointless to upgrade your desktop PC. We're heading towards 10 year upgrade cycles. And no, I'm not bitter...

I'm not so sure about that. Even in a "mobile first" world, there are a lot of technologies and improvements that carry over from the mobile world that can provide benefit to the desktop world.

For example, L4 caches are in consumer products, thanks to a quest for performance per watt, where they previously had not been. Right now, there's not a whole lot of use for a 128MB L4 cache, but I don't think it'd be difficult to take advantage of it.

The classic example of "mobile first" bringing benefits to the desktop was Conroe.

At any rate, even though benefits are being focused at the "bottom" of the performance stack, look how much better the bottom has gotten. We're almost to the point where an Atom-level core can power a computer for cheap. I say almost, because I don't feel like Silvermont is quite enough, and because it's rather overpriced right now (although there is no technical reason it couldn't be priced much lower). Airmont might bring us there, perhaps Goldmont, but it's not hard to see that it'll catch up quick.

The industry is on the verge of making some major breakthroughs. New channel materials (SiGe, Ge, III-V -- next 2-4 years), new transistor designs (FinFET, TFET, NWFET -- already here in the case of FinFETs, other designs coming within a decade), stacked memory (this year, and in increasingly higher usage as costs come down), and newer memory technologies (PCRAM, ST-MRAM... the latter which has actually made it into the real world).

There's also been a growing trend of fixed-function hardware that was (I believe) pioneered by Intel's quick sync. AMD now has VCE, TrueAudio... it's not going to stop there either.

Look at mobile displays. There's strong potential for mobile focused display technologies to make their way into desktop panels.

Even with this lull of CPU performance gains, there has been something pretty incredible that is already making a huge difference with computing performance, and that thing is NAND. Solid State Drives are a simply huge improvement.

There's a lot to be excited for. Kaveri may not be one of those things, and you may not consider Haswell or Broadwell to be enough of an improvement to win over your hard earned dollars, but it's coming. Call me crazy, but the computer industry has only recently become "big." It still pales in comparison to energy, retail, finance, defense... as the industry grows, so will the improvement.