Haswell Xeon With Iris Pro Confirmed for Christmas!

[LukaP]

Should've thrown Iris pro on the G3258. That would've made almost every AMD APU they've made irrelevant. 

no need for the almost slap a 5300 in there, price it at 100$ and you have a budget gaming build for 150$ that beats the 7850k that costs (or did) 180$

[patrickjp93]

no need for the almost slap a 5300 in there, price it at 100$ and you have a budget gaming build for 150$ that beats the 7850k that costs (or did) 180$

With only 2 threads there are a number of games you wouldn't have been able to play. Also, it looks like Carrizo will kick some ass in gaming with onboard HBM. It will be interesting to see if AMD can get some HPC adoption of HSA for the iGPU compute.

[LukaP]

With only 2 threads there are a number of games you wouldn't have been able to play. Also, it looks like Carrizo will kick some ass in gaming with onboard HBM. It will be interesting to see if AMD can get some HPC adoption of HSA for the iGPU compute.

1. 2 intel cores are about the same as 2 kaveri modules, so no difference there, you just do interrupts when you get a task done
2. i havent looked into carrizo, so i cant comment on that, what i will say is, that GCN is a gaming architecture, while intel just focused on TFLOPS, not adding any real hardware engines, so im not sure brute force will be able to beat carrizo. kaveri sure

[patrickjp93]

 

1. 2 intel cores are about the same as 2 kaveri modules, so no difference there, you just do interrupts when you get a task done
2. i havent looked into carrizo, so i cant comment on that, what i will say is, that GCN is a gaming architecture, while intel just focused on TFLOPS, not adding any real hardware engines, so im not sure brute force will be able to beat carrizo. kaveri sure

 

There are games existing where if the available core count according to the OS is below 4 they will not open.

Intel is slowly changing on that. They added a bunch to pixel throughput on the Broadwell GPU architecture for instance. They also increased throughput per core by 40%.

[LukaP]

Given Iris Pro 6200 is coming to the Broadwell 5770k which is on LGA 1150, and also given the PCIe lane allocation for this Xeon is 1x16, 2x8, 1x8/2x4, I'd say it's actually pretty likely this is LGA1150 material.

With only DDR3L support? Naah...

There are games existing where if the available core count according to the OS is below 4 they will not open.

Intel is slowly changing on that. They added a bunch to pixel throughput on the Broadwell GPU architecture for instance. They also increased throughput per core by 40%.

That's a software limitation. If you have a core that is twice as strong as the other core, but you have 2 of those, with smart predictions and interrupts you can have exactly the same performance.

[patrickjp93]

With only DDR3L support? Naah...

That's a software limitation. If you have a core that is twice as strong as the other core, but you have 2 of those, with smart predictions and interrupts you can have exactly the same performance.

Eh, we can petition Intel

 

I know that's a software limitation, but 2 of Intel's threads do not have the full performance of 4 AMD threads even in the best case scenario.

[LukaP]

There are games existing where if the available core count according to the OS is below 4 they will not open.

Intel is slowly changing on that. They added a bunch to pixel throughput on the Broadwell GPU architecture for instance. They also increased throughput per core by 40%.

Pixel throughput means jack shit when you have polygons be processed by your shaders. And they have to process tessellation, occlusion, and other stuff that gaming focused arches have had forever.

[LukaP]

Eh, we can petition Intel

I know that's a software limitation, but 2 of Intel's threads do not have the full performance of 4 AMD threads even in the best case scenario.

It's extremely close. And Intel has (from my experience) quite great prediction. And they do awesome resource allocation.

[Opcode]

Can we see Iris pro on a quad core celeron?

 

That would be almost perfect for a budget gamer.

Broadwell-H and Broadwell-K will have GT3e. Celeron and Pentium will more than likely have GT1. While the mobile sector will have GT2 (excluding the U series which may have GT3e as well).

 

Should've thrown Iris pro on the G3258. That would've made almost every AMD APU they've made irrelevant. 

Iris Pro isn't exactly all that it's cracked up to be. It's twice as fast as GT2 (says Intel) tho it's still completely irrelevant compared to GCN.  GT2 alone is over three times as slow as current GCN offerings in gaming. So with what Intel is promising even if they put GT3e into a G3258 it would be still slugging behind what APU's that already exist on the market. Carrizo will bring GCN 1.2 and possibly HBM so Intel's doing good but they're still playing catch up. On the up side is GT3e will be all the better at compute than GT2. So Quicksync and whatever else will be all the faster.

[Deletive]

With only 2 threads there are a number of games you wouldn't have been able to play. Also, it looks like Carrizo will kick some ass in gaming with onboard HBM. It will be interesting to see if AMD can get some HPC adoption of HSA for the iGPU compute.

Lets hope Excavator can bring the CPU performance up too  

[patrickjp93]

Broadwell-H and Broadwell-K will have GT3e. Celeron and Pentium will more than likely have GT1. While the mobile sector will have GT2 (excluding the U series which may have GT3e as well).

 

Iris Pro isn't exactly all that it's cracked up to be. It's twice as fast as GT2 (says Intel) tho it's still completely irrelevant compared to GCN.  GT2 alone is over three times as slow as current GCN offerings in gaming. So with what Intel is promising even if they put GT3e into a G3258 it would be still slugging behind what APU's that already exist on the market. Carrizo will bring GCN 1.2 and possibly HBM so Intel's doing good but they're still playing catch up. On the up side is GT3e will be all the better at compute than GT2. So Quicksync and whatever else will be all the faster.

GT3e is also based on a new architecture which improved throughput per EU by 40% by cutting 2 SIMD blocks from each sub-slice and adding a 3rd. Intel was hamstringing its own performance by limiting bandwidth to its shaders despite the huge bandwidth advantage of Crystallwell L4 cache.

[patrickjp93]

Lets hope Excavator can bring the CPU performance up too

It's looking like it will from the benches of Carrizo I posted on here back in October, but the onboard HBM is doing the heavy lifting. IPC improvements are minuscule it looks like. Though having the HBM will make the iGPU for AMD much more powerful.

[ComradeHX]

There are games existing where if the available core count according to the OS is below 4 they will not open.

Intel is slowly changing on that. They added a bunch to pixel throughput on the Broadwell GPU architecture for instance. They also increased throughput per core by 40%.

I have never played a game that does that.

 

Which game was it?

[patrickjp93]

I have never played a game that does that.

 

Which game was it?

I believe BF4 and AC Unity both don't tolerate fewer than 4 cores. Bioshock 2 I believe is another one. GTA IV plays like shit without 4 cores.

[ComradeHX]

I believe BF4 and AC Unity both don't tolerate fewer than 4 cores.

I play BF4 just fine.

 

Too bad I will never touch ACU though.

[patrickjp93]

I play BF4 just fine.

 

Too bad I will never touch ACU though.

Are you on an I3 with Hyperthreading? 

[ComradeHX]

Are you on an I3 with Hyperthreading? 

Pentium

[That Norwegian Guy]

I have a severe lack of reaction from my pants from this news.

 

Up the core count and I might donate inches.

[Deletive]

It's looking like it will from the benches of Carrizo I posted on here back in October, but the onboard HBM is doing the heavy lifting. IPC improvements are minuscule it looks like. Though having the HBM will make the iGPU for AMD much more powerful.

Thats what I'm afraid of, the IPC improvements being little. Atleast they said a ~30% performance increase at 15w. I'm just looking for something that a module can carry it's own.

Like mantle having a chance to get the igpu in there as well to help improve performance. HBM will be very promising but I'd also like them really do something with Dgpu's higher then the crappy ones they gave us last year.

[ionbasa]

Now all we have to do is wait for SuperMicro and Tyan to come up with some E-ATX motherboards using the new chipset and call it a day.

 

This might actually go into a new Workstation I need...

[Opcode]

GT3e is also based on a new architecture which improved throughput per EU by 40% by cutting 2 SIMD blocks from each sub-slice and adding a 3rd. Intel was hamstringing its own performance by limiting bandwidth to its shaders despite the huge bandwidth advantage of Crystallwell L4 cache.

GT3e is already being used in a select few current Haswell microprocessors. It does sort of play catch up with the A10-7700k tho the A10-7850k edges in another level of performance. Broadwell will bring an extra 8 EU's to GT3e so Intel can achieve above A10-7700k iGPU performance with Broadwell. Tho I doubt it will stand fairly against even the past generation A10-7850k. Plus new APU's from AMD will come before Broadwell makes it to the market bringing even more competition. Broadwell will only be bringing GT3e to high end skews (H, K, and U). So it's hard for Intel to even compete with AMD when you have to spend $200+ just to get their flagship graphics. At the end of the day anyone who's buying a microprocessor that has GT3e in it will already be buying a discrete GPU anyways. The only place I see it having a selling point for Intel is possibly in devices like the Macbook Air.

 

It's looking like it will from the benches of Carrizo I posted on here back in October, but the onboard HBM is doing the heavy lifting. IPC improvements are minuscule it looks like. Though having the HBM will make the iGPU for AMD much more powerful.

IPC increases for Carrizo is looking to be around 5% according to SiSoft benchmarks. Tho like always they are to be taken with a grain of salt as we will have to wait for more leaks to get a base line understanding of what to expect. Hopefully we get a real leak from Carrizo-L soon.

[TheRetiredSlave]

Given Iris Pro 6200 is coming to the Broadwell 5770k which is on LGA 1150, and also given the PCIe lane allocation for this Xeon is 1x16, 2x8, 1x8/2x4, I'd say it's actually pretty likely this is LGA1150 material.

 

Broadwell, yes.. Given that this is Haswell, I'd expect expect this to be the same as the other Haswell CPUs with Iris Pro.

 

If you look at a Comparison between this and the i7-4770R, they're pretty much exactly the same other than the normal spec differences between i7s and Xeons.. I would expect that this is just a differently binned variant of the same package as the 4770R. Since the i7-4770R is a physically larger package, I would doubt that they'd be able to easily fit everything into the much smaller 1150 package.. I'm not saying that they can't (it is Intel we're talking about), but I would have serious skepticism towards any rumors of this being on an LGA1150 package.

[patrickjp93]

GT3e is already being used in a select few current Haswell microprocessors. It does sort of play catch up with the A10-7700k tho the A10-7850k edges in another level of performance. Broadwell will bring an extra 8 EU's to GT3e so Intel can achieve above A10-7700k iGPU performance with Broadwell. Tho I doubt it will stand fairly against even the past generation A10-7850k. Plus new APU's from AMD will come before Broadwell makes it to the market bringing even more competition. Broadwell will only be bringing GT3e to high end skews (H, K, and U). So it's hard for Intel to even compete with AMD when you have to spend $200+ just to get their flagship graphics. At the end of the day anyone who's buying a microprocessor that has GT3e in it will already be buying a discrete GPU anyways. The only place I see it having a selling point for Intel is possibly in devices like the Macbook Air.

 

IPC increases for Carrizo is looking to be around 30% according to SiSoft benchmarks. Tho like always they are to be taken with a grain of salt as we will have to wait for more leaks to get a base line understanding of what to expect. Hopefully we get a real leak from Carrizo-L soon.

GT3e using generation 8 architecture is not being used in any Haswell product currently. GT3e based on Generation 7.5 is. 

[patrickjp93]

 

Broadwell, yes.. Given that this is Haswell, I'd expect expect this to be the same as the other Haswell CPUs with Iris Pro.

 

If you look at a Comparison between this and the i7-4770R, they're pretty much exactly the same other than the normal spec differences between i7s and Xeons.. I would expect that this is just a differently binned variant of the same package as the 4770R. Since the i7-4770R is a physically larger package, I would doubt that they'd be able to easily fit everything into the much smaller 1150 package.. I'm not saying that they can't (it is Intel we're talking about), but I would have serious skepticism towards any rumors of this being on an LGA1150 package.

Yes, Haswell is still Haswell. Though, you also need to remember how much die area is still left under the hood of those Quad-Cores. There's plenty of room for a 40-core iGPU.

[patrickjp93]

GT3e is already being used in a select few current Haswell microprocessors. It does sort of play catch up with the A10-7700k tho the A10-7850k edges in another level of performance. Broadwell will bring an extra 8 EU's to GT3e so Intel can achieve above A10-7700k iGPU performance with Broadwell. Tho I doubt it will stand fairly against even the past generation A10-7850k. Plus new APU's from AMD will come before Broadwell makes it to the market bringing even more competition. Broadwell will only be bringing GT3e to high end skews (H, K, and U). So it's hard for Intel to even compete with AMD when you have to spend $200+ just to get their flagship graphics. At the end of the day anyone who's buying a microprocessor that has GT3e in it will already be buying a discrete GPU anyways. The only place I see it having a selling point for Intel is possibly in devices like the Macbook Air.

 

IPC increases for Carrizo is looking to be around 30% according to SiSoft benchmarks. Tho like always they are to be taken with a grain of salt as we will have to wait for more leaks to get a base line understanding of what to expect. Hopefully we get a real leak from Carrizo-L soon.

I chalk most of that up to having HBM, which can be used as LLC.