can Intel's Iris PRO inside Broadwell be a low-end discrete video card killer?

[Dabombinable]

I cant believe how many people are buying into them numbers.

 

 

 

So the i5-5675C will hold 122 FPS on average in GTA V regardless of the scenario? Because if you disagree (like Anandtech does) then you're just agreeing with me this whole time. I would not urge anyone to go out and buy one because they are under the impression that it performs that well in games. Numbers can be skewed and misleading like Tom's Hardware is known for. I can't believe you guys would egg people on by approving such results when it only performs 45% total of what Tom's Hardware advertised.

That power consumption says it all.

[Dabombinable]

-snip-

[patrickjp93]

That power consumption says it all.

The TDP says it all. AMD is still far behind when it comes to efficiency.

I'd hold off on that until all the Carrizo numbers come in.

[Dabombinable]

I'd hold off on that until all the Carrizo numbers come in.

Ignore the last post-forum glitched then I had to throw my keyboard to get backspace and delete working again.​ (Either that or Project Spartan has issues)

[NAP51DMustang]

"will put the hurt on AMD gpus". yeah maybe versus 6 year old AMD gpus

 

also nice benches done at 720p...... . .
 

[patrickjp93]

"will put the hurt on AMD gpus". yeah maybe versus 6 year old AMD gpus

 

also nice benches done at 720p...... . .

 

There's also some of 1080p where Intel beats Kaveri as well by a decent margin, and no, it's officially more powerful than a GTX 740.

[NAP51DMustang]

There's also some of 1080p where Intel beats Kaveri as well by a decent margin, and no, it's officially more powerful than a GTX 740.

I'd like to see it do MGS:GZ at 1080p maintain 55-60 fps with a mix of medium and high settings (an HD 6850 can do that). Then I'd consider it a good contender for low end (but still, we won't be seeing these take over the medium range and higher cards anytime soon really).

[patrickjp93]

I'd like to see it do MGS:GZ at 1080p maintain 55-60 fps with a mix of medium and high settings (an HD 6850 can do that). Then I'd consider it a good contender for low end (but still, we won't be seeing these take over the medium range and higher cards anytime soon really).

That'll come around 2020 by industry experts as iGPU moves to take up 98% of "CPU" dies, and you'll still want dGPUs to supplement iGPU performance. I was starting to wonder where Nvidia's lower end Maxwell cards were, but now I have my answer. It's not worth trying to put them out when AMD and Intel have that performance range basically covered.

[Megahurt]

Low end discrete cards IMO are more about giving some life back to an old system than a viable option for a new system, and I don't think this iGPU will change this.

[ForsakenLive]

If they can put this into an i3 or Pentium, then we have a deal.

[SteveGrabowski0]

Umm, a core i3 comes in at some 140€. any serious GPU comes in at 150 or more. So you get a higher price. this way, you take an i5 with 4 real cores and a great IGPU and later upgrade (if you feel the need) to an x60 or higher class

 

Wow, I never expected their iGPU to be this good on Broadwell. But yeah, this is really a great option. How many times do you see people coming on here with builds with throwaway temp CPUs and temp GPUs because they don't have the money for that i5 + GTX 970 system right now? Now they can pick the i5, still play new games on it at 720p, and a couple of months later they're not throwing away a Pentium or an R7 260x when it comes time to put a powerful CPU or GPU in to balance the system. Now there is no need to tell people they're best off waiting until they can afford both the i5 and the expensive GPU.

[niofalpha]

RIP Dedicated APU's. Hopefully this will make quick sync look better/ faster.

[patrickjp93]

Low end discrete cards IMO are more about giving some life back to an old system than a viable option for a new system, and I don't think this iGPU will change this.

The Skylake flagship iGPU (72 EUs) will change that if it comes to desktop, and Cannonlake even more so since it will be Gen 9 and not 8.5 at (expected) 96  EUs.

[sgloux3470]

They don't need to if the game has its own built in benchmark. Benchmarks are written purposely to match what performance impacts that you will experience in game at that graphic quality. Tom's Hardware could of been driving out in the middle of no where for all five of their tests with two of them staring off into space (skybox frames). They earned their bad reputation among the professional tech communities and they are continuing to drive the same garbage that earned them that reputation (LTT members seem to be the only ones oblivious to what's in front of them).

 

The in game benchmark is derived of scenes that the player will actually experience in the game (in game camera movement). So you can keep throwing a blind eye trying to boast the company that you root for although anyone with common sense can easily tell Tom's results are beyond skewed. As no one cares how many frames you get when you sit there and look at the sky. Wait until more benchmarks leak and I guarantee they will fall into place with Anandtech numbers. 122 FPS... shit I'd be lucky to get that with my HD 5870. Once I get my other hard drive installed I will benchmark it to prove it.

 

A lot of people here assume that because someone made a graph in excel it means their numbers are valid or representative of actual performance.

[patrickjp93]

A lot of people here assume that because someone made a graph in excel it means their numbers are valid or representative of actual performance.

Is it safe to assume Opcode assumed the same thing only because it's Anandtech, an operation that actually has a bigger vested interest in Intel since it has a pipeline to them?

[sgloux3470]

Is it safe to assume Opcode assumed the same thing only because it's Anandtech, an operation that actually has a bigger vested interest in Intel since it has a pipeline to them?

 

If there's only two choices in a vacuum without a frame of reference you can't really determine which one is accurate if the results disagree by such a large margin.  

 

The fact that two different places are running the same game on the same settings and have drastically different results simply highlights that people really, really shouldn't put so much stock in benchmark numbers.

 

The whole point of ingame canned benchmarks is to try and minimize these sorts of discrepancies by providing a standardized test.  Instead, we have every review site running their own "benchmark courses" which is just fucking stupid if they want results to be reproducible and comparable.

[patrickjp93]

Sure, if there's only two choices in a vacuum without a frame of reference you can't really determine which one is accurate if the results disagree by such a large margin.  

 

The fact that two different places are running the same game on the same settings and have drastically different results simply highlights that people really, really shouldn't put so much stock in benchmark numbers.

 

The whole point of ingame canned benchmarks is to try and minimize these sorts of discrepancies by providing a standardized test.  Instead, we have every review site running their own "benchmark courses" which is just fucking stupid if they want results to be reproducible. 

But the other issue (as I pointed out) is a number of canned benchmarks actually run several fps slower than the same scenes in-game, and as Opcode claims, some run faster. There has to be that unshaken reliability factor.

[Megahurt]

The Skylake flagship iGPU (72 EUs) will change that if it comes to desktop, and Cannonlake even more so since it will be Gen 9 and not 8.5 at (expected) 96  EUs.

I don't know, by the time skylake is considered an old system, low end gpus might still be viable, apus are guetting better very fast, but so are video cards, Pascal from nvidia seems very promising. CPUs are pretty much stagnating so it will stay relevant for quite a while, meanwhile gpu performance is pretty much skyrocketting.

[Blackie Sheen]

No, the $275 price tag on the Core i5-5675C ruins the value proposition vs. a cheap Core i3 + dedicated GPU.

Well if gaming is only on the side then it makes perfect sense which is what is expected for a consumer buying this kind of chip

[patrickjp93]

I don't know, by the time skylake is considered an old system, low end gpus might still be viable, apus are guetting better very fast, but so are video cards, Pascal from nvidia seems very promising. CPUs are pretty much stagnating so it will stay relevant for quite a while, meanwhile gpu performance is pretty much skyrocketting.

It isn't CPUs that are stagnating. It's consumer software design. Intel's flops numbers have multiplied by a factor of 6 in as many years. We have wide vector instructions. Code just has to be allowed to optimize for them. Most consumer software is still compiled as to support all the way back to the minimum of windows XP: the Pentium 3. -O0 vs -Ofast on any C++ compiler is a huge difference. Go take a look at my HPC for Dummies blog here on LTT. I demonstrate very clearly what the differences are.

[patrickjp93]

I think the more important takeaway here is the 5775C is actually keeping pace with the 4790K in a lot of CPU-bound games and benchmarks, and I think Anandtech correctly attributes that to the 128MB L4 cache which is used to house things kicked out of L3 and move more things closer to the CPU. That's actually a rather large IPC gain just from the extra cache in the right scenarios.

 

If the 6700K doesn't come with an Iris Pro graphics SKU, then a potential flagship refresh chip with it could actually see a 10+% jump on its own, and I think seeing the overclocked Broadwell results will be interesting for everyone.

[sgloux3470]

But the other issue (as I pointed out) is a number of canned benchmarks actually run several fps slower than the same scenes in-game, and as Opcode claims, some run faster. There has to be that unshaken reliability factor.

 

The point of benchmarks isn't to show people exactly what the performance they can expect is.  It's to show how different pieces of hardware compare to each other in a test with as little variance as possible so that you know the only difference in results is from the hardware.  

 

If I run the Shadow of Mordor ingame benchmark 3 times, I get identical average framerates, 3.325% RSD on Max and 1.322% RSD on min FPS.   This means that if I swap out my GPU and run this benchmark again three times, I can be reasonably confident that my results can be compared.  Maybe the framerate is slightly higher or lower than I would get if I just played the game, but that doesn't matter.  What matters is relative performance percent, because I'm comparing hardware not software.

 

If I'm manually running a course, I'm introducing a whole lot more error, no matter how consistent I think I'm being.  Moreover, my course probably doesn't cover all the representative elements of the game.  If my game has fancy weather effects, then if I'm not benchmarking those effects I'm not doing a representative benchmark of the game.  

 

A scripted course is always the better option.

[bob345]

wow, and just like that amd's apu's are dead. That seriously looks to be a viable option for low cost gaming now.

[patrickjp93]

The point of benchmarks isn't to show people exactly what the performance they can expect is.  It's to show how different pieces of hardware compare to each other in a test with as little variance as possible so that you know the only difference in results is from the hardware.  

 

If I run the Shadow of Mordor ingame benchmark 3 times, I get identical average framerates, 3.325% RSD on Max and 1.322% RSD on min FPS.   This means that if I swap out my GPU and run this benchmark again three times, I can be reasonably confident that my results can be compared.  Maybe the framerate is slightly higher or lower than I would get if I just played the game, but that doesn't matter.  What matters is relative performance percent, because I'm comparing hardware not software.

 

If I'm manually running a course, I'm introducing a whole lot more error, no matter how consistent I think I'm being.  Moreover, my course probably doesn't cover all the representative elements of the game.  If my game has fancy weather effects, then if I'm not benchmarking those effects I'm not doing a representative benchmark of the game.  

 

A scripted course is always the better option.

Though a scripted course never covers it all either, and I'd argue it can even cover less in a lot of cases.

[patrickjp93]

wow, and just like that amd's apu's are dead. That seriously looks to be a viable option for low cost gaming now.

Mmmm, not yet. If AMD brings out higher clocked Carrizo SKUs before Zen, it could close that gap pretty easily, especially since Carrizo gets GCN 1.2 with Delta Color Compression, and the Excavator core IPC is higher than Steamroller by a 5-15% margin depending on the workload.