@Zoltan: I'd be very interested in an investigation whether Haswell scales better with HT than previous chips. The reason I suspect this: it's got more execution units! They're not terribly relevant to FP crunching, but in mixed workloads they should definitely show. And BOINC is the best real world usage case for mixed workloads.
I know this is not easy to measure, but so far I haven't seen any meaningful data besides the usual benchmarks. I also suspect Haswell will show quite some teeth under heavy server load.. but still no benchies.

Now, that I've fixed my random restarts (it turned out that my old memory module kit (KVR1333D3N9K2/4G) is not compatible with my new motherboard), I can do some benchmarks. How can I run mixed workloads? (Should I install two BOINC managers on this host?). What other benchmark do you want me to run? (I've read on a local hardware test site, that the Apache web server is actually slower on i7-4770K than on the previous CPUs.)

@Haswell in general: I think people are looking at this chip in the wrong way, or were simply expecting too much. I think Haswell is great for BOINC, even if you can't use AVX2 yet. Great in the way "a significantly better choice than Ivy, for the same price".

I do agree with you.

Skgiven has a nice idea of getting a cheap i3 or i5 and make a rig for GPU crunching only. But I know myself then I will do more with the system and then it won´t work. Not that I am depended on 1 system. I have taken a lot of old stuff home from work...Won´t do that anymore, unless there is a T7400 :)

I went for an even cheaper G2020 (~£50), rather than an i3 for ~£120, an MSI board with 3 slots (2 immediately useful, but the 3rd has potential if I use a riser) and a reasonable 80+ PSU (30decibels at400W).
My i7-3770K @4.2GHz is in an identical 2/3 PCIE slot sub-€100 motherboard with two GPU’s (was 3 for a while with a riser).
While the Gigabyte GA-Z87X-OC is an excellent motherboard and something I might consider if I wanted 3 or 4 top GPU’s (GTX780s) for a new build, my ambitions are rather more reserved, and I like the challenge of doing it on the cheap.
Going by the specs, even the ‘GA-X58-OC Force’ motherboard won’t allow you to have 4 GPU running at PCIE3X16. Anyway, it's an extreme end board and requires water cooling - it's for rich plumbers :)

The two GTX660 have cost me €420, so €580 for the rest is not possible in the Netherlands.

You can buy the following from Amazon.de:

Two GTX 660’s (€182each) €364
Corsair TX650 €75
Motherboard ~€90
8GB DDR3 €65
CPU ~€60
128GB SSD €78
External USB2 DVDRW €30
Antec Three Hundred Two €66
- Total €828

Thanks for your suggestions skiven, appreciated.
But the GTX660 at Amazon.de where not able to be delivered to the Netherlands, some parts do and some don't even sold and shipped at Amazon. That's why I bought them in the Netherlands for more.
I will invest in an Intel CPU, therefor I want an six core with HT and will wait until they arrive as Zoltan mentioned.
For now I want a new rig to fit the two 660's nicely and that will be an AMD based one, as I already have the Sabertooth MOBO for it. I always search for the best or second best parts when they are in an action. Eventually they need to replace my old stuff, but I need those for working from home, so the old powers users keep around for a while, and it is nice that when they are powered on they will crunch a bit.
In the past I have looked and bought the cheaper parts, but I will never do that anymore due to bad experience.
Most of my rigs are Dell, not without reason, but I now lots of others think otherwise. Some are more than 10 years old and work still great, even all the fans. From brandless rigs, I had to replace fans within a year. Could absolutely be coincidence though.

---

Greetings from TJ

Some are more than 10 years old and work still great, even all the fans. From brandless rigs, I had to replace fans within a year. Could absolutely be coincidence though.

Totally depends on the fans you buy. For instance the Antec case I mentioned comes with 2 tri-speed fans. They generally last a long time. If you add more fans you can go cheap cheap and replace them more often or you can get quality fans and probably never replace them. BTW I've had to replace plenty of fans on Dell and other brand name systems too.

Some are more than 10 years old and work still great, even all the fans. From brandless rigs, I had to replace fans within a year. Could absolutely be coincidence though.

Totally depends on the fans you buy. For instance the Antec case I mentioned comes with 2 tri-speed fans. They generally last a long time. If you add more fans you can go cheap cheap and replace them more often or you can get quality fans and probably never replace them. BTW I've had to replace plenty of fans on Dell and other brand name systems too.

Of course that happens, even Dell uses Samsung stuff as an example. I never go cheap, I have my brands I like and have good experience with and will select from them. So I do with coffee and pens but also with hardware. Those days Samsung has improved but I will never buy it anymore as experience in the past where bad.
That is the reason I have two EVGA GTX660's, I could have got them cheaper from another brand.

---

Greetings from TJ

Going back a bit to the APU discussion, the benefit of the A10-6800K over the A10-6700 is limited to the CPU’s top performance and supposedly 2133MHz RAM vs 1866MHz for the lesser model (which might help at some GPU projects). It’s not a processor I would recommend for several reasons. The 100W TDP is poor vs the 65W TDP of the A10-6700, which has the same GPU with the same GPU clocks and shader count. It’s there for people that have inexpensive electric/want a slightly faster processor (performance at any cost), but don’t want/need a discrete GPU, and chumps.
At a basic level iGPU’s don’t yet have enough oomph for anything other than lightweight crunching. It’s still the case that a basic dual core CPU and a low-end to mid-range GPU will outperform a high end iGPU. Unfortunately most reviewer sites aren’t paid to demonstrate this,
but you can see the reality here!

While both AMD and Intel CPU’s with integrated GPU’s make for good desktop/office systems they are not really much use for crunching, relative to discrete GPUs. BTW. I found that AMD iGPU processors ran surprisingly cool and quiet with the box set heatsink – even when crunching.

By the end of the year Kaveri 3rd Generation A series APUs should change things a bit, maybe up to a 30% improvement in terms of CPU performance/Watt. The APUs will be 28nm Graphics Core Next and the CPUs are supposed to bring 15 to 20% improvement per cycle (IPC). While this is relatively better than Intel’s i7-3000 to i7-4000 (5 or 10%) move AMD have to catch up on several 22nm Intel revisions. I thought this might bring in 35 to 45W APU desktop processors, but the speculation is still 65W, which is actually a good thing. For GPU crunching and especially mixed CPU/GPU, hUMA will be very interesting. It should at least remove some bottlenecks and better facilitate the devs, but I suspect it might find itself a special place with some projects. It’s really the first glimpse of the next step forward in computing, and it’s far too early to say if it will only increase GPU performance by ~40% or more than double it…

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

What I have read and heard from other people the build in graphics from Sandy Bridge aren´t great either.
What I see is that a lot of computers (of the shelf) are build smaller and smaller. BOINC started as a project to use PC´s idle time on a CPU only. But now we have GPU´s to crunch really fast. And a lot of crunchers build
crunch-dedicated rigs.
Personally I think that build in features of two devices into one will always have a loss somewhere. The link that skgiven posted shows that clearly.
I thinks these APU´s are great for office use and browsing/mailing.

---

Greetings from TJ

@APUs: comparing HD4600 to GTX650Ti doesn't make much sense other than to show that discrete GPUs can be much faster - at significantly higher cost (in transistors and power consumption). Apart from that the two are in completely different leagues.. it's a bit like comparing GTX650 to HD7970. Of course the fatter card wins, but otherwise it doesn't tell us much.

And the current Intel GPU are obviously still catching up, but are slowly getting there. Starting from Ivy they can crunch BOINC OpenCL. My HD4000 could do ~45k RAC at Collatz, but is shifted to Einstein now for 9.5k RAC. That's not much.. but it only consumes ~20 W and I have it anyway.

Looking further into the future the discrete GPUs will still always be faster, as long as they don't require fast communicatiion with the CPU. The reason is simply power consumption (up to 130 W shared with the CPU vs. up to 300 W for GPUs) and memory bandwidth (memory soldered onto a PCB near the processor can always be faster than socketed memory further away). Die stacking DRAM could provide an APU with massive bandwidth, and we could even argue: that if a GPU is allowed to pull 300 W, why isn't a CPU? But we'd still be looking at "300 W shared with the CPU or all for the GPU".

However, I can also see some really neat things happening. Soon CPU and GPU will get unified adress space, both are slowly merging as evidenced by AMDs fusion roadmaps for some time. Where could this lead? Imagine a CPU module not unlike Bulldozer, maybe increase the number of integer cores per module to 4 or 8, and instead of letting them share 2 FPUs let them share a bunch of GPU shaders (e.g. one GCN compute cluster) for floating point and vector operations. That's the ultimate Fusion!

As far as I know it's not yet on the roadmaps and you'd need to provide some low latency and special function units as well.. but the more sharing, the better the heavy number crunching will be.

@Zoltan: by "mixed" I meant a mix of integer and floating point operations. The easiest way to produce them would be to run x tasks of an integer-only project like ABC or Collatz and x tasks of some regular project. Measure overall throughput wiith HT on and off, then do the same on an Ivy (or Sandy). This would be enough to quantify if HT scaling became better under mixed loads.

Giving both the same ressource share could be enough to keep BOINC running x/x tasks.. but using 2 separate managers may be safer. This depends on how long you need to average WU runtimes to know how fast the setup is, which will depend on the projects choosen.

And since Einstein has responded well to HT in the past it would be nice to see HT scaling of Haswell vs. Ivy just for Einstein. I'm not sure if I would like to run such tests if I had a Haswell.. but these are the tests I think we're still missing :)

MrS

---

Scanning for our furry friends since Jan 2002

Something like this?

Going back a bit to the APU discussion, the benefit of the A10-6800K over the A10-6700 is limited to the CPU’s top performance and supposedly 2133MHz RAM vs 1866MHz for the lesser model (which might help at some GPU projects). It’s not a processor I would recommend for several reasons. The 100W TDP is poor vs the 65W TDP of the A10-6700, which has the same GPU with the same GPU clocks and shader count.

While both AMD and Intel CPU’s with integrated GPU’s make for good desktop/office systems they are not really much use for crunching, relative to discrete GPUs. BTW. I found that AMD iGPU processors ran surprisingly cool and quiet with the box set heatsink – even when crunching.

By the end of the year Kaveri 3rd Generation A series APUs should change things a bit, maybe up to a 30% improvement in terms of CPU performance/Watt. The APUs will be 28nm Graphics Core Next and the CPUs are supposed to bring 15 to 20% improvement per cycle (IPC). While this is relatively better than Intel’s i7-3000 to i7-4000 (5 or 10%) move AMD have to catch up on several 22nm Intel revisions. I thought this might bring in 35 to 45W APU desktop processors, but the speculation is still 65W, which is actually a good thing. For GPU crunching and especially mixed CPU/GPU, hUMA will be very interesting. It should at least remove some bottlenecks and better facilitate the devs, but I suspect it might find itself a special place with some projects. It’s really the first glimpse of the next step forward in computing, and it’s far too early to say if it will only increase GPU performance by ~40% or more than double it…

Interesting read, thanks skgiven. Haven't paid a lot of attention to the A10 but your comments made me take a look.

From the ITPRO review: "Intel Haswell vs AMD Richland head-to-head":

"The A10-6800K scored 0.81 in our tests – an improvement on the 0.76 scored by the A10-5800K. The A10-6700 wasn’t far behind, scoring 0.79 in the same tests. It’s progress, but it’s not enough to match Intel’s Haswell-based Core i5s – in terms of pure application performance, these top-end APUs match Ivy Bridge-based Core i3 chips.

It’s in gaming where Richland APUs make up ground. The A10-6800K’s Radeon HD 8670d graphics core scored 90fps in our 1,366 x 768 Low-quality Crysis test, and 43fps when we upped the resolution to 1,600 x 900 and quality settings to Medium. In both tests that’s five frames faster than Haswell’s HD Graphics 4600 could manage, and even further ahead of Ivy Bridge.

The gap was even more pronounced in Just Cause 2: the A10-6800K averaged 64fps in the game’s low-quality benchmark, but the Core i7-4770K scored just 44fps.

AMD’s latest chips perform well without consuming much power. Our test rig, comprising 8GB of RAM and an SSD alongside the A10-6800K, consumed a miniscule 37W when idle and 108W when stress-tested. Our high-end Haswell machine required just 38W when idling – but this figure rocketed to 185W at peak load."

http://www.itpro.co.uk/desktop-hardware/19975/intel-haswell-vs-amd-richland-head-head/page/0/3

Like you say, not a lot of compute difference between the 6800K and the 6800K. The 35 watt TDP difference should make A10-6700K even more efficient.

"Conclusion

Haswell generates plenty of column inches for Intel thanks to its dominance at the top of the processor market but, unless you’re going to make full use of an expensive Core i7-4770K or i5-4670K, it’s worth looking on the other side of the fence.

After all, there’s plenty to like about AMD’s top-end A10-6800K APU. It’s got more graphical power than anything Intel can muster, it’s more frugal, and it’s much cheaper than Intel’s top-end Haswell chips. It can’t quite match Haswell when it comes to pure processing power, but that’s its only weakness – and, for many, it’ll be more than good enough.

It’s not as one-side as you might think. If you want pure processing power and don’t mind the cost then Haswell is for you. If it’s a balanced experience you’re after and graphics and budget are the priority then a top-end APU is the chip of choice."

http://www.itpro.co.uk/desktop-hardware/19975/intel-haswell-vs-amd-richland-head-head/page/0/4

If you look at the iGPU/APU reviews, several things stand out. CPU bound processing is generally better on the Intel processors, but gaming is better on AMD’s APU’s. At the same price range you need to compare the APU’s with i3 processors. In this situation the AMD processors are better for gaming and the Intel processors are better at most other things. However, AMD's competitive edge actually extends to the i7-4000 processors – it’s still the case that gaming on the integrated AMD GPU is better,
http://hexus.net/tech/reviews/cpu/57197-amd-a10-6700-32nm-richland/?page=6

The alternative to an iGPU/APU is a discrete GPU, so it’s reasonable enough to compare them. A discrete GPU not only consumes more power because it has it's own board, but because it’s more powerful (does more work). It’s the very lack of GPU transistors that makes the Intel hd4600 so uncompetitive against discrete GPU’s. The hd5200 compares better simply because it’s twice as big, but the increase in transistors requires a CPU downclock, suggesting Intel would struggle to add more graphics performance as is. I would be inclined to ignore the transistor costs, or argue that AMD have the better balance, and concern myself with purchase costs. The fact still remains that you can build a better system with a discrete GPU for less than an iGPU system.

If we are talking about crunching it’s better to think of performance per Watt of the system. The 15W pulled by an hd4600 at Einstein still has to be supported by the rest of the system. Even if you OC this iGPU it’s not going to compete well with a mid-range discrete GPU in terms of System performance/Watt.

While I don’t think this situation will continue, entry level GPU’s are set to disappear as Intel and AMD eat into that market on the laptop and desktop front, it will take time (years). This has actually been developing for a long time - A6 and A8. The hd4000 and hd5000 just introduce some tangible competition by Intel. Unfortunately a lot of ‘Release’ reviews showed the hd4600 to be a good iGPU against the latest APU’s, but a broader look has shown the APU’s can outperform the hd4600 in non CPU dependent situations. AMD's APU’s are better priced and the running cost/performance are better, depending on what you are doing.

When it comes to crunching there are only two projects that can presently utilize the iGPU’s for OpenCL, but with time this should change. More projects have used the APU’s and in my experience the AMD’s are a cooler while crunching flat out (CPU+iGPU). IIRC the GPU of an A6 got about 30K at POEM. While my hd4000 can get 6.5 to 8.5K at Einstein (depending on what else I’m doing) last year’s Trinity appears to be able to get 36K (but possibly only 11.3K/day). If it is 36K/day that would suggest Richland would be able to get >40K/day.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

@Flashawk: no, I'd replace the FPU in the CPU modules by a compute unit array. Or, on second thought, I'd augment the FPU by this, as you'd still need a low-latency unit for special functions and serial code (anything not handled well by the CU). In the theoreticla APU you linked the GPU and CPU are still separate and communicate through the north bridge.

@"ITPRO-Review": that seems to be rubbish. Under load +150 W for the Haswell? How is the CPU going to do this while staying under 83 W TDP? Sounds more like they included a fat discrete GPU and forgot to tell us (no configuration given, except calling the Intel "high end" and the AMD not).

But they're trying to answer the wrong question anyway: A10-6800K and i7-4770K are simply not alternatives on the desktop. If you want/need maximum CPU power without going straight to socket 2011 or higher, you choose the 4770(K). If you just want/need "enough" CPU performance, you choose the A10-6700 or an i3 or i5. If you want serious gaming you need a discrete GPU anyway.. and should go for the i5. For casual gaming you'd be stupid not to choose the A10 (either one).

In essence: when buying a 4770(K) the iGPU is just a nice bonus which might help little or at least increase the resell value, even for us crunchers. Nothing more and nothing less. This doesn't really change when you consider the 4770R with the eDRAM cache and the biggest iGPU. It outperforms Richland, but you don't buy it because of its iGPU on the desktop!

This all changes in latop environments where both chips are power-limited and you can't just stick a discrete GPU in there.. but that's not what we're concerned with here.

MrS

---

Scanning for our furry friends since Jan 2002

It wasn't just the CPU's power, it was the entire system/machine that used 185W (at the wall).
Several bench tests will push a processor over it's TDP and it is a K model, so perhaps they had an OC.

"Our high-end Haswell machine required just 38W when idling – but this figure rocketed to 185W at peak load".
Ref: http://www.itpro.co.uk/desktop-hardware/19975/intel-haswell-vs-amd-richland-head-head/page/0/3

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

From Tom's Hardware:

"Here's the bombshell we figured out from today's testing: for overclocking, a quad-core Haswell-based processor at 22 nm requires more cooling than a six-core Sandy Bridge-E CPU at 32 nm, even though its lower power consumption produces less heat. The back-up for this is that two of the coolers from our Sandy Bridge-E cooling round-up re-appeared today with far worse apparent performance. Most overclockers blame Intel’s newer integrated heat spreader and transfer material for this discrepancy. These days, cheap paste replaces solder for connecting the CPU die to the spreader.

Cross-compatibility between LGA 1150, 1155, and 1156 sinks theoretically makes it possible for us to test dozens of heat sinks and fans. Unfortunately, most solutions are too small to cope with the heat issues an overclocked Haswell-based CPU suffers."

http://www.tomshardware.com/reviews/best-heat-sink-haswell,3554-25.html

It also makes one wonder if all this heat buildup is only due to the poor transfer material or if Intel is being less than honest about the power draw of Haswell. Low power ratings sell processors...

It also makes one wonder if all this heat buildup is only due to the poor transfer material or if Intel is being less than honest about the power draw of Haswell. Low power ratings sell processors...

They use a poor (cheap) TIM because of the low TDP there's no need for better at 3.6GHz (considering everyday usage - no, crunching is not an everyday use). It's a shame however, that they don't use a better TIM (e.g. soldering) in the K series, which definitely made for overclockers (and costs more while some features are disabled). So the way to heavily overclock a K series CPU leads through IHS removal and TIM replacement - voiding warranty. In my opinion this is intentional from Intel's part, they don't want to give warranty for heavy overclockers.
BTW my tests with my new motherboard and CPU are going so well that I've replaced them in my old host (ASUS P7P55 WS Supercomputer + Core i7-870). The new configuration consumes only 384W (from the outlet) while crunching 2 GPUGrid (NATHAN_KIDKIXc22 + NOELIA_1MG_RUN1) and 4 Rosetta@home workunits simultaneosly. The CPU and the GPUs are about 60-65°C.
The configuration is:
PSU: Enermax Platimax 600W
CPU: Core i7-4770K (@3.5GHz - no overclock yet - factory made cooler)
MB.: Gigabyte GA-Z87X-OC
RAM: Kingston KVR16N11S8K2/8 2x4GB 1600MHz
SSD: Kingston V100 96GB
GPU: ASUS GTX-670-DC2OG-2GD5 x2

Heat, I had a lot of problems with heat, the liquid cooler is back in the alienware and does okay when the CPU is doing nothing, 38-40°C, When running Rosetta then fast to 68-72°C.

If I could choose I would buy a CPU that stays cool when active, even if it would use 200Watt, then one with a low TDP and a lot of heat.
Anyway next project is to build the AMD driven rig with the two GTX660's in a big case with a lot of fans. I have checked a lot of them and the CM storm trooper will be it. Plenty of room, lots of fans, dust filters and both panels can be removed for easy building.

---

Greetings from TJ

Heat, I had a lot of problems with heat, the liquid cooler is back in the alienware and does okay when the CPU is doing nothing, 38-40°C, When running Rosetta then fast to 68-72°C.

If I could choose I would buy a CPU that stays cool when active, even if it would use 200Watt, then one with a low TDP and a lot of heat.
Anyway next project is to build the AMD driven rig with the two GTX660's in a big case with a lot of fans. I have checked a lot of them and the CM storm trooper will be it. Plenty of room, lots of fans, dust filters and both panels can be removed for easy building.

That doesn't make sense; if it used 200 watts, where would all the watts go? It can't disappear. It gets converted into heat.

---

XtremeSystems.org - #1 Team in GPUGrid

Heat, I had a lot of problems with heat, the liquid cooler is back in the alienware and does okay when the CPU is doing nothing, 38-40°C, When running Rosetta then fast to 68-72°C.

If I could choose I would buy a CPU that stays cool when active, even if it would use 200Watt, then one with a low TDP and a lot of heat.
Anyway next project is to build the AMD driven rig with the two GTX660's in a big case with a lot of fans. I have checked a lot of them and the CM storm trooper will be it. Plenty of room, lots of fans, dust filters and both panels can be removed for easy building.

That doesn't make sense; if it used 200 watts, where would all the watts go? It can't disappear. It gets converted into heat.

Off course it doesn't ;-) that's why I wrote; when I could choose...

---

Greetings from TJ

My understanding is that Haswell Iris Pro is actually 2 chips, a separate CPU and a GPU on the one 1150 processor board,
http://media.pcgamer.com/files/2013/06/Haswell-Iris-Pro.jpg

Are the other Haswell's the same?
I guess there might be cooling issues with that. Just having one Heat Spreader is dubious.

Perhaps they will be on the one chip when they drop to 14nm?

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

SK wrote:

It wasn't just the CPU's power, it was the entire system/machine that used 185W (at the wall).
Several bench tests will push a processor over it's TDP and it is a K model, so perhaps they had an OC.

"Our high-end Haswell machine required just 38W when idling – but this figure rocketed to 185W at peak load".
Ref: http://www.itpro.co.uk/desktop-hardware/19975/intel-haswell-vs-amd-richland-head-head/page/0/3

I know - that's why I brought this up. They can't just include some other hardware (or OC) in the Intel, not tell us about this, not include it in the AMD system and then attribute the difference in power consumption to the CPU. Well, I'm assuming that there is some other hardware not included in the AMD system, but that's because they don't tell us otherwise and because otherwise the numbers don't make any sense at all.

Beyond wrote:

It also makes one wonder if all this heat buildup is only due to the poor transfer material or if Intel is being less than honest about the power draw of Haswell. Low power ratings sell processors...

It's been measured: power consumption goes up roughly proportional to the performance increase (at the same clock speed), or significantly less than that in AVX2-heavy code (higher performane increase).

The problem with cooling is the insufficient TIM and the smaller area, over which the heat is released compared to e.g. Sandy-E. The Haswell cores got a bit larger compared to Ivy, but they've become almost tiny. Conducting the heat away from such small areas is harder and increases temperatures for a given cooling system.

SK wrote:

My understanding is that Haswell Iris Pro is actually 2 chips, a separate CPU and a GPU on the one 1150 processor board

No, the extra chip is the 128 MB eDRAM working as a L4$ in the top models. The i7 4770R as it too - I'm really curious to see what the cache could do for BOINC, server and professional apps. Intel promised gains in the high double digit percentages for some apps, which I don't doubt can be found in BOINC-land (POEM, Einstein, CPDN come to mind). Sadly it's as present as a ghost, even after being launched.

Edit: having read Tom's article.. they use 1.25 V and call it moderate. Yet that blows power efficiency completely out of the window. For 24/7 crunching I'm using 1.03 V on my 22 nm CPU and wouldn't recommend going any higher than 1.10 V. They're also using LinX, which gains ~70% performance per clock over Ivy thanks to AVX2. but that performance has to come from somewhere, i.e. it's about the most stressful piece of software for Haswell we currently have. Real world BOINC will produce less power consumption & heat.

Cooling Haswell is still not fun.. but not as dramatic as such articles might make you believe. Stay with moderate clocks & voltages and it will be fine. Except if you're TJ ;)

MrS

---

Scanning for our furry friends since Jan 2002

SK wrote:

My understanding is that Haswell Iris Pro is actually 2 chips, a separate CPU and a GPU on the one 1150 processor board
http://media.pcgamer.com/files/2013/06/Haswell-Iris-Pro.jpg

No, the extra chip is the 128 MB eDRAM working as a L4$ in the top models

Note that the larger (almost square) chip (on the left) is the 128MB eDRAM, and the smaller (rectangular) chip (on the right) is the Haswell CPU itself.