![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.
A10-7850K Performance Review
- Thread starter omega-x
- Start date
Totally agree. It's probably 30% slower realistically, but I suspect the tdp drop is greater than 65% (seen claims that the 100W desktop part actually exceeded 100W). Man those linked imgur cpu benches don't look so hot, unless I'm mis-reading them
... I hope those results are not correct, it barely looks any different from richland, unless I am not reading the slides correctly. Maybe they need to release an optimized driver.
And what about single threaded cpu performance? Look at the cinebench numbers, why is it doing worse than richland? in both single threaded and multi threaded. I thought kaveri would increase ipc?
Something seems fishy. What were the clocks? was richland clocked much higher?
grain of salt until we see stable results
And what about single threaded cpu performance? Look at the cinebench numbers, why is it doing worse than richland? in both single threaded and multi threaded. I thought kaveri would increase ipc?
Something seems fishy. What were the clocks? was richland clocked much higher?
grain of salt until we see stable results
- Joined
- Mar 1, 2003
- Messages
- 2,610
On paper, the GPU should be outperforming Trinity/Richland's much more than seems to be so in those charts. On the CPU side, I'm not sure what to expect; there's definitely an IPC improvement, but I'm afraid any gains are going to be offset by the 10% drop in clock speeds.
Nothing that can't be overcome with a mild overclock, obviously. So I'm more worried about what GPU results are going to show.
Anybody have an idea when the NDA on reviews lifts?
Nothing that can't be overcome with a mild overclock, obviously. So I'm more worried about what GPU results are going to show.
Anybody have an idea when the NDA on reviews lifts?
- Joined
- Sep 7, 2011
- Messages
- 10,214
The problem with the GPU on Kaveri is the clockspeed. It needs to be ramped up, hence we need to see overclocking numbers.
Dear [H]: please, PLEASE overclock the balls off the GPU on this thing. Do it properly. Underclock the CPU portion if necessary if it helps get the GPU to "holy shit" overclock levels. I want to see this bitch run Crysis.
On very high.
Dear [H]: please, PLEASE overclock the balls off the GPU on this thing. Do it properly. Underclock the CPU portion if necessary if it helps get the GPU to "holy shit" overclock levels. I want to see this bitch run Crysis.
On very high.
octoberasian
2[H]4U
- Joined
- Oct 13, 2007
- Messages
- 4,081
I saw January 14th. mentioned somewhere above or on that removed article. That or January 23rd. One of these dates is for AMD's Kaveri.
Forgive me, my memory hasn't been that great lately.
Michaelius
Supreme [H]ardness
- Joined
- Sep 8, 2003
- Messages
- 4,684
IPC is improved but the clock speeds are lower.
Might be a reason why they are not wasting time releasing FX variant of it.
- Joined
- Aug 3, 2004
- Messages
- 24,323
So when do we get 8, 12, and 16 core versions for real gaming?
You wont, real gaming = Intel
- Joined
- Aug 18, 2006
- Messages
- 11,754
Dumb question, but in regards to the FX processor line (assuming they're still going to make them) - couldn't they just take an APU and use the die space saved by not incorporating the IGP and put even more cores in there?
I'll have one running on dc-2400mhz ram soon enough.
What's surprising is the modest gain the GPU gets vs the previous generation. It looks like Iris Pro 5200 may still hold the crown in most games. Too bad it's really rare or Iris Pro might have made a difference.
At least Gigabyte has brought both the i5-4570R and i7-4770R to the Brix Pro, hopefully more manufacturers will do the same so we have options for Iris Pro.
yes, they could... and release it on FM2+ (not AM3+)
Could they? Of course.
The better question is: are the engineering / financial hurdles small enough to make it worth it?
I don't suppose it's that easy to just remove/shrink the GPU part and add more cores.
Pieter3dnow
Supreme [H]ardness
- Joined
- Jul 29, 2009
- Messages
- 6,784
AMD will not do anything on the FM2+ platform that has only cpu cores unless it is a disabled APU. They won't turn back now/
not sure how they overclocked the Igpu, but if you give me the settings, i can run the same benchmark on my A10-5757m laptop. 512mb vram, dual channel 1600.
Honestly it really depends on what screen your gaming on. I found that I can run pretty much all games I've thrown at my A10-5757m laptop @ 720p with decent fps. Considering this is a ultra thin book and 35 watt tdp its impressive. Kaveri A10 mobile chip should bring about some gains, vs the richland processor, so it will only get better.
Your comparing a desktop processor to a mobile processor. Your also comparing a Richland Processor to a Trinity one. Don't get me wrong the A10-5757m is a decent cpu, but even though it lacks 25% of the clock speed of the desktop version it is actually more than 25% slower. The turbo core is very aggressively scaled in the mobile version. Only time you see that 3.5ghz clock speed on the mobile part is strictly single thread.
I can get mine to average 3.1ghz across 4 cores of load. But yeah. It's closer to half the cpu as desktop richland under normal circumstance.
As to how I overclocked the iGPU, the AMD overdrive utility can do it. The GPU is rated for boost speed up to 825mhz, but never goes past the stock 720~mhz normally.
wonderfield
Supreme [H]ardness
- Joined
- Dec 11, 2011
- Messages
- 7,396
I'm somewhat disappointed. For developer uptick of HSA (and, more importantly, tooling), there needs to be a tremendously compelling argument for it, which means iGPU performance needs to be fairly significant. It needs to be fast enough to make not exploiting it completely unimaginable.
That's not what I'm seeing here. I'm seeing a reasonably well-performing chip with good compute capability, but I'm not seeing a catalyst of anything (no pun intended). I really think it was important for AMD to have released a much higher-performing part than these mid-range parts at a higher price point right out of the gate.
That's not what I'm seeing here. I'm seeing a reasonably well-performing chip with good compute capability, but I'm not seeing a catalyst of anything (no pun intended). I really think it was important for AMD to have released a much higher-performing part than these mid-range parts at a higher price point right out of the gate.
HSA isn't something you can easily do with a dGPU, and they've crammed as much iGPU as they can into this short of console-level die sizes and prices. They may have something up their sleeve given there were rumors about an APU with more shaders and GDDR5.
I will play around with that. What model is yours? I have the Acer V5-552-x418 its a pretty solid laptop. I am trying to figure out a way to change the amount of video ram. 512mb is kinda low, and I know if I bumped it up to 1gb I would see gains in most things graphic related. The bios is locked though and needs to be hacked in order to change this option. One day I will spend the time needed to unlock the bios.
I think the benchmarks are OK, but I think expectations were set fairly high by false (non-AMD sources) pre-release rumors. Some of those rumors suggested Kaveri was competitive with Haswell, or even the nearly 3 year old Sandy Bridge in CPU performance.
The graphs came directly from the PugetSystems.com review. Bing finally refreshed its cache and the article is gone.
I have it saved, but I don't think I should post the review text here. It's a fairly basic review and the system configs and results graphs I posted earlier are pretty much the whole Puget Systems review. More thorough reviews should be coming next week, but system availability was supposed to happen a day or two ago. i don't even
The cpu side of Kaveri I had no real expectations other than a 5-10% increase in IPC. It's the gpu side that really has me doing a double take. Quite honeslty I was expecting at least a 30% performance increase across the board compared to Richland. Heck a 20% increase would have been good. This is very alarming but will anxiously wait for official benchmarks.
- Joined
- Mar 1, 2003
- Messages
- 2,610
If AMD is smart, they'll jump the NDA date to squash the negative 'leaks'.
I honestly think it's unfathomable that there's only a ~10% GFX gain.
I honestly think it's unfathomable that there's only a ~10% GFX gain.
- Joined
- Sep 7, 2011
- Messages
- 10,214
Last time someone posted 'leaked' AMD benchmarks for a new CPU, they were pretty negative and all the fans said things in line with 'don't worry, when the NDA lifts, we'll see the REAL numbers!'
Turns out that when this happens, the benchmarks are usually spot-on.
Turns out that when this happens, the benchmarks are usually spot-on.
Puget Systems is one of AMD's select system builders deemed good enough to receive Kaveri early. It was a basic review with dozens of benchmarks, not a negative leak. afaict, Puget gave Kaveri all the benefits it could, including AMD branded DDR3-2133 memory ("2x AMD RG Gamer DDR3-2133 4GB 10-11-11-30"). This doesn't appear to be a negative leak, but representative of what customers will be getting as the systems ship... if they order with more expensive DDR3-2133. The graphics performance is even worse with DDR3-1600 memory.
Puget showed in synthetic benchmarks that AMD did deliver the 20% IPC improvements AMD claimed, but with a 10% drop in clock speed from the A10-6800K, overall CPU performance only improved ~10%. The testing seems to be on the up and up.
The GPU is a kicker though. While it may be a GPGPU powerhouse in certain workloads, the limited memory bandwidth from dual channel DDR3 could be what's holding it back in games. The scaling from DDR3-1600 to DDR3-2133 is pretty significant. At some point though it's a big waste of money since you can often buy a regular set of DDR3 DIMMs and a faster discrete video card for the same price as higher end specialty memory.
I'm not sure what miracles people were expecting.
stealthy123
2[H]4U
- Joined
- Feb 4, 2005
- Messages
- 3,149
It's interesting that since the Stars cores bandwidth has been the issue with the GPUs but AMD seems committed to a dual channel solution.
Maybe I don't understand memory systems as much as I should, but if it's been clear they needed more memory speed then that they wouldn't go with triple or quad channel.
Unless memory channels don't impact bandwidth the way I think they do.
Maybe I don't understand memory systems as much as I should, but if it's been clear they needed more memory speed then that they wouldn't go with triple or quad channel.
Unless memory channels don't impact bandwidth the way I think they do.
- Joined
- Apr 10, 2003
- Messages
- 23,532
Probably all comes down to maintaining a sub $200 price range for all CPU / APU products.
There are a lot of things AMD can potentially do, but there are costs involved. Whether it's development time, money, keeping pin count down (not just an issue for the CPU, but may also necessitate more PCB layers on motherboards) or other things, they all have consequences. Chalk it up to compromises all products face.
It's certainly a down-side to having an iGPU architecture which is so memory bandwidth dependent, without having mitigating solutions. The 2 steps Intel took (high bandwidth, shared L3 and 64GB/s Crystallwell "L4" device in Iris Pro equipped products) have a very positive effect on its deferred rendering architecture. For whatever reasons AMD still has a relatively pokey L2, not that it's suited for the same purpose that Intel uses the L3 cache. I was surprised that AMD didn't design in a large L3 into Kaveri to help with the iGPU bandwidth demands. I guess real "Fusion" is still a while away.
Maybe when DDR4 arrives for mainstream desktop products we'll see > 2 channel memory solutions.
That's not the context of the issue I quoted and responded to. That was not saying AMD needs it to compete, but AMD could have other ways of getting more gaming performance out of the GPU included in Kaveri. The small ~10% increase using DDR3-2133 seems to be disappointing.
The problem is that Kaveri's iGPU is very bandwidth constrained. Performance is lower than expected with faster DDR3, and scaling is pretty significant from 1600->1866->2133. If AMD had a large, high bandwidth L3 and used tricks similar to what Intel does (using L3 for both CPU and graphics), it could ease the dependence on external memory bandwidth (i.e. extract more performance).
A modest 4MB L3 would probably increase the cost of each chip by 10-15% due to the extra die area. That's the easy part. Integrating the GPU to use it is apparently the hard part.
Ozymandias
[H]ard|Gawd
- Joined
- Jun 21, 2004
- Messages
- 1,610
Should've gone with the GDDR5, just like our console overlords.
Pieter3dnow
Supreme [H]ardness
- Joined
- Jul 29, 2009
- Messages
- 6,784
Should've gone with the GDDR5, just like our console overlords.
Does not work like that, PS4 is just a special piece of hardware. If GDDR5 would benefit AMD in windows then it would have been there already.
In a system where you have no different hardware it easier to manage certain changes. Nor does it have to be compatible with 1001 different configurations.
wonderfield
Supreme [H]ardness
- Joined
- Dec 11, 2011
- Messages
- 7,396
Memory is more complicated than "more bandwidth = better", unfortunately, particularly with respect to CPUs.Should've gone with the GDDR5, just like our console overlords.
It may happen at some point, but don't be surprised if it isn't all that you might expect.
If slower and older iGPU's were bandwidth limited putting a faster and newer iGPU on die will give you little to no performance improvement. "Cranking" the iGPU clocks will also do little to nothing for performance in games while also causing power usage to shoot through the roof.
AMD really reeeeally needs dual channel DDR4 or quad channel DDR3 or some largish on die/package memory a la Crystalwell to make their iGPU performance go up significantly at this point.
If they don't have something in the works along those lines right now, to be ready for production in late 2014 or 2015, even their APU's will be over run or met by Intel for iGPU performance.
I'm kind've surprised they haven't already done this by now. They're not stupid and they probably realize they're mostly spinning their wheels by throwing more die space at the iGPU. Probably a lack of resources is their problem and has been since forever really.
wonderfield
Supreme [H]ardness
- Joined
- Dec 11, 2011
- Messages
- 7,396
To be fair to the 7850, it does come in considerably cheaper than any desktop Crystalwell part, the 4570R being the cheapest at $288. The R-series SKUs are available exclusively in BGA packages, too, whereas the 7850 is at least in a builder-friendly socketed package.
That said, it doesn't start looking like a particularly good value when all facets are considered anyway.
That said, it doesn't start looking like a particularly good value when all facets are considered anyway.
Wow, these benchmarks look really bad for the "steamroller" cores.
Am I missing something or is it actually SLOWER than Richland in many of the pure CPU benches?
Edit: so basically it looks ok (vs Richland, CPU only) in geekbench multi-core, best in PCmark, and actually worse in geekbench single core/cinebench.
It looks like most of the IPC gains were eaten by clock decreases. Overall, yuck.
Am I missing something or is it actually SLOWER than Richland in many of the pure CPU benches?
Edit: so basically it looks ok (vs Richland, CPU only) in geekbench multi-core, best in PCmark, and actually worse in geekbench single core/cinebench.
It looks like most of the IPC gains were eaten by clock decreases. Overall, yuck.
Last edited: