[gongo]

http://www.lostcircuits.com/mambo//i...&limitstart=16

IIRC HD4000 = 2 ROPS, 25.6gbps memory bandwidth, 1.15ghz 16EU gpu cores ( = 256gflops sp?)

at 720p...it runs good Farcry2 and RE5....the later being almost too good! 60fps! Hawell is rumored..and Intel rumors are almost true....to have 40EU gpu cores!!

granted it took the best microchip company 5~6 years to come up with a ....SoC with the power of large HD twins...nothing extreme....though we would need to think..why has Sony and MS not shrunk the PS360 to smaller casing yet?

Are we no longer able to have...PSOneTwo type of HD consoles going on?

[eastmen]

Hmm

The cpu is $313 by itself , couple it with a $200 mobo , $50 ram and a $150 ssd.


Seems like we still aren't quite there yet , although it seems like the much cheaper trinitys are there at a price that rivals the cost of the ps360

[Rangers]

I'm going to guess, yes. It competes with Trinity which is 384 SP's surely a more powerful GPU than PS360.

Two rops is awfully low for what it's worth, the consoles have 8. Do these rops run at 1.15ghz? If so it's more like 4+.

I'm guessing they're more capable ROPS as well. The CPU will be much more powerful than PS360 too, regardless it's obviously running benchmarks at 720P without issue.

[hoho]

Originally Posted by eastmen

couple it with a $200 mobo , $50 ram and a $150 ssd

That expensive stuff is required about as much as gold-plated case for the console.

Though I agree on the general part that the thing is more expensive than consoles but that wasn't the original question.

[Rangers]

It's price is higher, not necessarily cost. Ivy Bridge is a very small chip. 160 mm^2.

[french toast]

Haswell will put the hammer down next year, 40 eu's, some likely dedicated L4 cache and other improvements for massive bandwidth, combined with another new architecture not to mention the REAL 22nm after maturing like a fine cheese over the course of 12 months - AMD will be in serious trouble.

Not that i want AMD to go down the plug hole, quite the contrary i own AMD products, i just want AMD to be able to match or exceed Haswell next year.

integrated graphics have gone into overdrive the last few years, thanks to competition, if AMD goes then we will be looking at a slow stale technilogical progression, for over inflated prices.

Back to topic, yea i would say HD4000 is on par with a xbox360, if you combine the CPU in that and swap the two then i think games would improve in the 360.

[Rangers]

AMD is competing on 32nm though, if they go to 22nm and lets say double the 384 SP's they currently have, yeah...

Out of curiosity was the die size of the newest 360 iteration ever measured? Anand never measured it cause he didn't want to tear his heat spreader off, but I seem to remember we got a measurement from somebody...

[HMBR]

Originally Posted by eastmen

Hmm

The cpu is $313 by itself , couple it with a $200 mobo , $50 ram and a $150 ssd.


Seems like we still aren't quite there yet , although it seems like the much cheaper trinitys are there at a price that rivals the cost of the ps360

what?
ivy bridge can be used in a 50 USD motherboard,
and soon they are going to release the lower end ivy bridges, cheap mobile Core i3s, and the special models of desktop core i3s with the HD4000 (most will use the HD2500), so I expect that soon you will be able for some 150 usd to buy a i3+HD4000+h61mb+cheap ram,

but in reality wasn't llano already faster? you can buy some 400sps llano for a really low price right now...

[sebbbi]

Originally Posted by HMBR

but in reality wasn't llano already faster? you can buy some 400sps llano for a really low price right now...

We had this discussion already last year. AMD E350 netbook APU (Brazos 1.6 GHz) runs many console ports at 30 fps if you set details to low (matching console settings) and lower resolution to 1280x720 (matching console resolution). So you would argue that Brazos already matches the 7 year old consoles in many games. Ivy Bridge CPU is around 8x faster than E350 on average (http://www.anandtech.com/bench/Product/328?vs=551), but the GPU speed difference isn't that big. Games are most of the time GPU bound on all integrated GPUs, so the extra CPU performance wouldn't likely improve the gaming performance at all.

Originally Posted by gongo

at 720p...it runs good Farcry2 and RE5....the later being almost too good! 60fps!

That's twice the frame rate of the console version, so you could argue it's twice as powerful as the current consoles (in running games designed for current consoles). However it's not that simple. The console frame rate is often more stable than PC frame rate, because the developer can specially optimize the code for console bottlenecks. There are too many different PC CPU/GPU configurations to properly optimize away all the bottlenecks (improve minimum frame rate). Of couse if the game would be properly optimized just for Ivy Bridge (or E350) it would likely run faster (and if 3d API could be skipped completely on PC there would be another extra speed boost). But back to reality: Basically nobody optimizes their games specially for low end integrated GPUs (high end GPUs are often a much higher priority). Good thing is that we finally have integrated (laptop) GPUs that are capable of running console ports with a playable frame rate (and with a slightly higher detail settings to boot).

[tunafish]

Originally Posted by eastmen

Hmm

The cpu is $313 by itself

At present, the cheapest CPU with HD Graphics 4000 is $235, but as soon as Intel gets to launching their dual cores, that's going to go down to $150 or so.

couple it with a $200 mobo

Why on earth would you do that? Intel has spent the last 5 hardware generations moving the performance-sensitive stuff from the motherboard to the CPU die. There really isn't much left. Unless you plan to go for really heavy overclocking or need some less common ports, there is never a reason to pay more than ~$120 for a MB. And if you don't plan to overclock at all, the $70 Gigabyte MB is the best choice.

After Intel releases Haswell, which moves the VRM on-package, there won't be *any* sensible reason to buy an expensive MB left. None at all. I wonder how the MB manufacturers are going to sell their expensive edition motherboards then.

$50 ram

You can get 4GB of 1600MHz ram for ~$25. No real need for any more for a budget box.

and a $150 ssd.

Seriosly? For a low budget box? How about a $50 HDD instead.

[Fafalada]

Originally Posted by sebbbi

That's twice the frame rate of the console version, so you could argue it's twice as powerful as the current consoles

That kind of neatly coincides with SB being close to consoles IME, and IVB roughly doubling the GPU.

so the extra CPU performance wouldn't likely improve the gaming performance at all.

The neat thing is that it's now actually possible to do things on PC like use left over CPU to run free MLAA or something. Granted that's assuming you actually spend time doing optimization for a specific type of hw, but the recent integrated chipsets arguably make for a viable platform to optimize for unlike most others.

[TheD]

Originally Posted by sebbbi

We had this discussion already last year. AMD E350 netbook APU (Brazos 1.6 GHz) runs many console ports at 30 fps if you set details to low (matching console settings) and lower resolution to 1280x720 (matching console resolution). So you would argue that Brazos already matches the 7 year old consoles in many games. Ivy Bridge CPU is around 8x faster than E350 on average (http://www.anandtech.com/bench/Product/328?vs=551), but the GPU speed difference isn't that big. Games are most of the time GPU bound on all integrated GPUs, so the extra CPU performance wouldn't likely improve the gaming performance at all.

Are you really, really sure you are thinking about the E350?

A quick search on youtube shows a lot of videos with the E350 running games at framerates that are kind of really, really crappy (BF3 SP at 640x480 at 19FPS on lowest, Bops at 800x600 and lowest settings @ 9FPS).

[gongo]

Interesting...very interesting...when i created this thread...my impressions were that HD4000 and PS360 wins on certain areas...HD4000 does not have the fillrate and mem bandwidth but comes with faster....shaders...? I did not expect to hear developers....hinting that IB is twice as fast as PS360! Granted...RE5/Capcom 'pc ports' were built for pc too!

so i went to youtube looking for HD4000 gameplay videos..this guy made many recent games at 720p ...running on HD4000! Nice!
Some games were very smooth..some jerky..could be due to fraps..and/or his i7 3770k HT is on....what do you guys think?
http://www.youtube.com/watch?v=bF0X1...1F6B94AC0738F9

Originally Posted by tunafish

After Intel releases Haswell, which moves the VRM on-package, there won't be *any* sensible reason to buy an expensive MB left. None at all. I wonder how the MB manufacturers are going to sell their expensive edition motherboards then.

Think that got misreported....only the dual cores ulv haswell for ultrabooks get the vrm on die...or was it on package...? either one...it is interesting how Intel can accomplished that..i took a look at my matx htpc and....those vrms..mosfets...caps are big and many!

[gongo]

One more thing about the HD4000...that was not widely reported...is that it has its own 8mb LLC...that is one big scratchpad edram like thingy ...!?! Do the resident devs feel there can be optimization to have for 8mb of fast...cache??

So it means IB is a cpu with a total 16mb of L3 cache...for a pretty tiny microchip!

[ToTTenTranz]

Originally Posted by gongo

One more thing about the HD4000...that was not widely reported...is that it has its own 8mb LLC...that is one big scratchpad edram like thingy ...!?! Do the resident devs feel there can be optimization to have for 8mb of fast...cache??

So it means IB is a cpu with a total 16mb of L3 cache...for a pretty tiny microchip!

I believe the GPU's access to the L3 cache in sandy/ivy bridge was designed to be completely transparent (thus not "optimizable") to game developers.

No that any developer would ever bother to make specific optimizations for Intel iGPUs.

[sebbbi]

Originally Posted by TheD

Are you really, really sure you are thinking about the E350?

A quick search on youtube shows a lot of videos with the E350 running games at framerates that are kind of really, really crappy (BF3 SP at 640x480 at 19FPS on lowest, Bops at 800x600 and lowest settings @ 9FPS).

There was a forum where users collected their E350 gaming benchmark scores. Over 50 games were listed, and roughly half of them ran at around 30 fps or higher. I can't find it anymore, but here are some scores:

Battlefield 2 (2005) - 1024x768, all max, 25-35 fps
Crysis 2 (2011) - 1280х720, 20 fps
Dirt 3 (2011) - 1024x600, very low, 25-28 fps
Mass Effect 2 (2011) - 1024х600, 20 fps
Call of Duty: Modern Warfare 2 (2009) - 1280х720, 25 fps

Here are some more from Anandtech:
Batman, low: Arkham Asylum 29 fps
Battlefield 2, low: 50.8 fps
Company of Heroes, low: 44 fps
Need for Speed: World, low: 33.5 fps
Quake 4, low: 57.3 fps

Metro and BF3 PC versions are unfortunately too much for E350. There are some other games such as Civilization V that are completely unplayable (lots of draw calls). Draw calls are actually a bottleneck for Intel drivers as well. I hope that DX 11.1 (and DX 12) further reduces the overhead of draw calls, because spending lots of extra cycles doing extra work doesn't sound right on (battery operated) tablets and mobile phones (Windows 8 & Windows Phone 8). Luckily for us Windows 8 already runs games a few percent faster, so OS/driver/service side optimization seems to be a big focus for them right now.

[Grall]

Originally Posted by Fafalada

The neat thing is that it's now actually possible to do things on PC like use left over CPU to run free MLAA or something. Granted that's assuming you actually spend time doing optimization for a specific type of hw, but the recent integrated chipsets arguably make for a viable platform to optimize for unlike most others.

Would it actually be possible to render tiles - I assume the Intel HD graphics works on tiles like previous GMA solutions - to L3, then read that back in with the CPU for MLAA processing, without having to take a roundtrip via bandwidth-limited (starved, really) main memory? I'd be extremely surprised if that was the case, but it'd be really cool of course.

Although with IB being OpenCL-certified I guess it would be extremely beneficial if you could access GPU results directly on-chip without having to hand everything off via main memory. Much less latency of course, and less power spent (always a concern these days), and any bandwidth not wasted on needless transactions hither and yon the better.

[Svensk Viking]

Originally Posted by sebbbi

There are some other games such as Civilization V that are completely unplayable (lots of draw calls). Draw calls are actually a bottleneck for Intel drivers as well. I hope that DX 11.1 (and DX 12) further reduces the overhead of draw calls, because spending lots of extra cycles doing extra work doesn't sound right on (battery operated) tablets and mobile phones (Windows 8 & Windows Phone 8). Luckily for us Windows 8 already runs games a few percent faster, so OS/driver/service side optimization seems to be a big focus for them right now.

Isn't draw calls what DX11 multithreaded rendering is supposed to help with?

Civilization V is confirmed to use that, and Nvidia GPUs showed clear performance improvements when their drivers enabled support for it

[sebbbi]

Originally Posted by Fafalada

The neat thing is that it's now actually possible to do things on PC like use left over CPU to run free MLAA or something. Granted that's assuming you actually spend time doing optimization for a specific type of hw, but the recent integrated chipsets arguably make for a viable platform to optimize for unlike most others.

Ivy Bridge has full speed AVX (256 bit wide vectors) and it has new float16<->float32 conversion vector instructions as well. It would be pretty good for image post processing (esp 16 bit float HDR data). Performance analysis (clock rate) data shows that Ivy Bridge CPU clocks are usually very low when it's running games, so there's certainly lot of untapped performance left. Most games just utilize 2 or 3 threads, and the rest could crunch 256 bit AVX data at full speed. But so far I don't know if any game supports AVX yet.

Originally Posted by Svensk Viking

Isn't draw calls what DX11 multithreaded rendering is supposed to help with?

Civilization V is confirmed to use that, and Nvidia GPUs showed clear performance improvements when their drivers enabled support for it

On early AMD drivers recorded command buffers didn't improve performance at all compared to doing similar draw calls again and again every time (I tested on Radeon 5850). I remember reading that both AMD and Nvidia optimized their command buffer implementations after Civ V release. They must have first implemented that feature simply by fully performing recorded draw calls at command buffer execution time instead of transforming those commands to a hardware display list (at capture time), and just putting an execute token+pointer to the ring buffer (at command buffer execution time). Executing a command buffer shouldn't cost much more than issuing a single draw call (but I am not sure what kind of extra safety checks PC drivers must perform, so it might be a bit more expensive on PC).

It seems that Intel hasn't yet optimized their command buffer support. Ttheir driver team must be much smaller compared to companies that have GPUs as their main area of business. AMD and Nvidia release new drivers monthly, while Intel does that a few times a year.

[TheWretched]

AMD stopped their monthly driver releases effectively today... but AMDs drivers aren't the bees knees either (and don't even start with Linux).

Intel has a VERY bad reputation concerning their GPU drivers... Not sure about Pulsbou, but those were PowerVR derived... so it wasn't even their code to begin with. They "work"... but just barely. Never had to suffer them for long, luckily.

[gongo]

Haswell seems ready to shock the console world...just 12 more months and you may see guys like Asus, Valve, Acer, Apple...putting out ITX gaming boxes?? AVX2, FMA4, L4 cache, GT3 40 EU(like a dream!), transactional memory...all seems like it is built for gaming....could Haswell even go toe to toe with Wii U??? DX10 vs DX11....3 cores cpu vs 4 cores cpu...feels close man!

...granted the fillrate limitation do not come into effect first....AT fillrate tests seem low..
http://www.anandtech.com/show/5871/i...phics-tested/4

Wii U 4850 class gpu fillrate is 10gp/s

Ivy bridge mem bandwidth is 25.6gb/s...PS3 levels...so Haswell will improve on the numbers and efficiency, logically a Haswell ITX gaming box can run DX11 games at 720p...comfortably? Who knows what will Lucid2 brings...put a GT740M with HD5000...turn on Lucid...boom!...or hype?

[TheChefO]

Originally Posted by Rangers

Out of curiosity was the die size of the newest 360 iteration ever measured? Anand never measured it cause he didn't want to tear his heat spreader off, but I seem to remember we got a measurement from somebody...

I don't remember where I got it, but xb360 total die size at current node was 213mm2.

Projected die size at 28nm of <120mm2.

So yeah, for the size, IvyBridge should be cleaning the xb360's clock!

[pjbliverpool]

Originally Posted by gongo

Haswell seems ready to shock the console world...just 12 more months and you may see guys like Asus, Valve, Acer, Apple...putting out ITX gaming boxes?? AVX2, FMA4, L4 cache, GT3 40 EU(like a dream!), transactional memory...all seems like it is built for gaming....could Haswell even go toe to toe with Wii U??? DX10 vs DX11....3 cores cpu vs 4 cores cpu...feels close man!

In CPU terms Haswell should walk all over WiiU.. It'll be packing very near 500 GFLOPS in the AVX2 units alone.

Regarding the GPU it's not so clear. I'd peg the HD4000 at about 50% faster than the HD3000 overall even though on paper it should be much faster than that - possibly due to memory bandwidth limitations. I'd also say the 4000 is a little faster than current gen consoles, maybe by as much as 25% which Llano being about 25% faster than that.

Haswell is pegged to be between 50% and 200% faster than Ivybridge in GPU terms which would thus put it a little less than twice as fast as the current consoles at the low end or as much as 3.75x faster. Once you add in the potential of AVX2 on the CPU itself for graphics (around another 500 GLOPS) then your talking about a system which has easily 2-4x the graphical power of PS360 with maybe 4-5x the general CPU performance.

So in comparison to WiiU, I'd say odds are that Hasell will outperform it in graphics capability and certain that it will significantly outperform it in CPU capability.

Wii U 4850 class gpu fillrate is 10gp/s

I doubt it will be that quick. If it were then only the most optimistic estimations of Haswell would be competitive.

[AlStrong]

Originally Posted by Rangers

Out of curiosity was the die size of the newest 360 iteration ever measured? Anand never measured it cause he didn't want to tear his heat spreader off, but I seem to remember we got a measurement from somebody...

CGPU was 168mm^2 @45nm, and the eDRAM was 45mm^2 (likely 65nm given the scaling amount).

http://forum.beyond3d.com/showpost.p...&postcount=239

[pjbliverpool]

Ivybridge and Haswell are certainly interesting in that they offer greater than current generation console performance on a single tiny, low power chip. But I think the more interesting application for the GPU's in them is as a GPGPU/Physics co-processor working in tandem with the CPU cores and using a discrete GPU to handle the rendering.

Remember how excited many were about the PhysX PPU and the thought of a dedicated physics processor in a PC? Well we're there now. Everyone with a Sandybridge or higher CPU already has a powerful dedicated SIMD processor inside their PC sporting between 100 and 300GFLOPS that's going totally unused. Imagine what could be done if that were turned to physics and other GPGPU tasks while the discrete GPU handles the heavy graphics lifting.

Imagine what Haswell could do if fully utilised in that way with around 1.25 TFLOPS (CPU+GPU) to dedicate to none rendering workloads. That's like dedicating 4 entire Xbox 360's to power your game code, physics and AI while a separate GPU with 20x the rendering power of Xenos (680 level) handles the graphics.