APU13- AMD Kaveri details- 856Gflops, 3.7Ghz CPU,720Mhz GPU

[monstercameron]

Certainly at 1080p, the Iris Pro can't even beat a 630 GT.

http://www.computerbase.de/artikel/grafikkarten/2013/intel-iris-pro-5200-grafik-im-test/11/

stick with popular benches so they can't find fault with the source, as to limit trolling by...

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/6

AT has the i7-4950hq iris pro and crystalwell on par with the gt640 at lowered resolutuions and the gt64m puls away at higher resolutions[bandwidth starved?]


http://ht4u.net/reviews/2012/msi_radeon_hd_7750_2gb_ddr3_test/index24.php
and these guys have the 7750 ddr3 slightly under gt640 performance, this doesnt bode well...

 

[LogOver]

You're not sure if 2.5x faster than the GT 630, which is Iris Pro level, is faster than Iris Pro?

I'm not even sure that kavery is faster than gt 630... Because you know... It's AMD presentation.

 

[Shivansps]

stick with popular benches so they can't find fault with the source, as to limit trolling by...

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/6

AT has the i7-4950hq iris pro and crystalwell on par with the gt640 at lowered resolutuions and the gt64m puls away at higher resolutions[bandwidth starved?]


http://ht4u.net/reviews/2012/msi_radeon_hd_7750_2gb_ddr3_test/index24.php
and these guys have the 7750 ddr3 slightly under gt640 performance, this doesnt bode well...

Anand tested the higher resolutions with higher details, larger textures means less things on the 128MB L4, they should have tested high resolution at the same details, high resolution will also hit the texture fill rate, so we cant be sure what will really happen.

 

[monstercameron]

I'm not even sure that kavery is faster than gt 630... Because you know... It's AMD presentation.

please keep your audacious comments to yourself and stay ot. What evidence do you have to doubt this other than anecdotes?

 

[schmuckley]

Idk but..This thing here says 130 GFLOPS on CPU side..
That's comparable to Haswell.
http://community.amd.com/community/amd-blogs/amd-business/blog/2013/10/21/what-is-your-processor-iq

 

[frozentundra123456]

stick with popular benches so they can't find fault with the source, as to limit trolling by...

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/6

AT has the i7-4950hq iris pro and crystalwell on par with the gt640 at lowered resolutuions and the gt64m puls away at higher resolutions[bandwidth starved?]


http://ht4u.net/reviews/2012/msi_radeon_hd_7750_2gb_ddr3_test/index24.php
and these guys have the 7750 ddr3 slightly under gt640 performance, this doesnt bode well...

Isnt iris pro mobile only? If you increased the TDP to 100 watts like Kaveri will probably be, the performance would be a lot higher. Problem with iris pro is that it is too expensive and very limited availability.

We need to see Kaveri in real benchmarks and if crossfire finally works well, although bandwidth and vram may still be limiting if you pair it with a 1gb card. Also I am interested in cpu performance, gpu isnt the only thing.

Edit: I just looked up the raw gflops of the 6800k, and amd states it as 780 gflops. So if the 860 gflops for kaveri is accurate, based on raw gflops alone, it is what, about 10 percent faster?

 

[Stickmansam]

ugh to stick with 4670k pruchase or wait for Kaveri/Steamroller...

AHH decisions ><

 

[LogOver]

please keep your audacious comments to yourself and stay ot. What evidence do you have to doubt this other than anecdotes?

Well, most probably kavery is faster than gt 630 but my point was that it is not possible to make any conclusions based on one benchmark done inside AMD controlled environment.
BTW, pretty much everyone tested mobile version of iris pro.

 

[LogOver]

Idk but..This thing here says 130 GFLOPS on CPU side..
That's comparable to Haswell.
http://community.amd.com/community/amd-blogs/amd-business/blog/2013/10/21/what-is-your-processor-iq

These are SP GFLOPS. Comparable to dual core haswell - yes.

 

[monstercameron]

Well, most probably kavery is faster than gt 630 but my point was that it is not possible to make any conclusions based on one benchmark done inside AMD controlled environment.
BTW, pretty much everyone tested mobile version of iris pro.

fair enough. yeah, the mba's all had them.

 

[SiliconWars]

Well, most probably kavery is faster than gt 630 but my point was that it is not possible to make any conclusions based on one benchmark done inside AMD controlled environment.
BTW, pretty much everyone tested mobile version of iris pro.

The 65W 4570R got soundly beaten by the mobile Iris Pro - http://www.anandtech.com/show/7399/...review-iris-pro-driving-an-accurate-display/3

In most cases it was far below the GT 640/750M, getting a severe beating at even 1680x1050 let alone 1080p.
http://www.anandtech.com/show/7399/...review-iris-pro-driving-an-accurate-display/3

 

[Vesku]

Idk but..This thing here says 130 GFLOPS on CPU side..
That's comparable to Haswell.
http://community.amd.com/community/amd-blogs/amd-business/blog/2013/10/21/what-is-your-processor-iq

"Sasa Marinkovic is Sr. Manager, Technology Marketing at AMD. His postings are his own opinions and may not represent AMD&#8217;s positions, strategies or opinions."

Hmm, given AMD's track record (He who shall not be namedAMD) I'd say that statement makes this guys postings not much better than stuff from VR-Zone or Semiaccurate.

 

[Homeles]

130 GFLOPs is awfully anemic. It's fitting though, given the fact that Kaveri only has two FPUs.

At 3.5GHz, Haswell has a peak FLOP rate of 32 * 3.5 GHz = 112 GFLOPs/core.

Kaveri's peak FLOP rate should be 16 FLOPs per cycle, and given a clock speed of 3.7GHz, we land 59.2 GFLOPs/core(~120 GFLOPs total). I assume the 130 GFLOPs is measured with turbo, which Kaveri would need to be clocked at ~4.1GHz to achieve.

So, interestingly enough, one Haswell core has floating point performance not far off from two Kaveri modules. That's the power of AVX2 vs AVX, really.

 

[monstercameron]

130 GFLOPs is awfully anemic. It's fitting though, given the fact that Kaveri only has two FPUs.

what are some comparison points to judge this by?

 

[schmuckley]

These are SP GFLOPS. Comparable to dual core haswell - yes.

wat? Haswell @ 5+ GHz gets about 130 GFLOPS throughput..IF you have the RAM @ 2000+ (quad-core that is)

the Hyperthreading adds 10-15

 

[Homeles]

what are some comparison points to judge this by?

Check my edit.

wat? Haswell @ 5+ GHz gets about 130 GFLOPS throughput..IF you have the RAM @ 2000+ (quad-core that is)



the Hyperthreading adds 10-15

Theoretical vs. real world, et cetera, et cetera.

 

[Centauri]

ugh to stick with 4670k pruchase or wait for Kaveri/Steamroller...

AHH decisions ><

Between the two companies AMD is developing the more interesting products, and they are in far greater need of your money.

Wait for the AMD.

 

[Homeles]

Between the two companies AMD is developing the more interesting products

That's a tough argument to make in the face of the incredible battery life Haswell offers.

 

[monstercameron]

That's a tough argument to make in the face of the incredible battery life Haswell offers.

I am sure the 4670 would kill a 45WHr battery in less that 20mins...

haswell mobile uses similar amounts of power to amd skus, it is idle where haswell seems very optimized...cpu we all know goes to intel then again amd has better gpu performance.

 

[itsmydamnation]

So, interestingly enough, one Haswell core has floating point performance not far off from two Kaveri modules. That's the power of AVX2 vs AVX, really.

no it isn't. thats the difference between 128bit vs 256bit FPU's..... AVX is 256 bit, AVX 2 has int SIMD and a really slow gather. AMD has a 128bit int SIMD instruction set (XOP). AMD does 256bit AVX via two internal ops.

 

[Stickmansam]

Between the two companies AMD is developing the more interesting products, and they are in far greater need of your money.

Wait for the AMD.

I don't think I could wait an year for Steamroller (3M/6T) if Kaveri can't beat current 3M/6T
I hear there will be nothing but APU's for 2014

Hopefully some benches will pop up soon

 

[cbn]

GPU side on the other hand is kinda of letdown in my opinion as 720Mhz is pretty low. Maybe AMD knew they will be BW limited in the end so they have found a balance where the GPU clocks stop to matter much without additional BW.

Has there been any talk of Mantle helping extract more performance from these memory bandwidth limited iGPUs?

 

[grimpr]

Heh...Broadwell's highest configuration is going to mop the floor with this one. Ouch.

Sure, at what cost? 500$ for the chip alone? Thats good for intel stock holders, bad for everyone else who wont give a cent compared to the 120$ dollar desktop apu.

 

[NaroonGTX]

Yeah, let's compare AMD's $100~$150 APU's to Intel's Iris Pro chips which are in completely different price brackets and not even available in anything outside of a BGA package. Makes sense. Let's also ignore how much bigger the Iris Pro's die size is compared to said APU's.

Just lol.

Hopefully some benches will pop up soon

Seeing as how Kaveri is supposedly hitting retail in mid-January, I guess we'll be seeing benches either later this month or sometime in December.

 

[Centauri]

^And yet SiliconWars is the one who is getting the infractions in this thread...