- Oct 16, 2019
- 2,746
- 6,653
- 146
This is not a good picture Charlie's painting of whatever's left of Intel's roadmap.
For the record, this is also significantly worse than any expectations I ever had as well. Far, far worse.
This is not a good picture Charlie's painting of whatever's left of Intel's roadmap.
For the record, this is also significantly worse than any expectations I ever had as well. Far, far worse.
AMD will price it in a way that they can sell all they can produce for highest price.
If their production capacity would be unlimited, raising price would be dumb because that gives your competition the option of selling cheaper. However if you price it in a way the competing product doesn't make any sense to buy, you get all the business.
But supply is limited and hence above doesn't work. So you price as high as you can get away with while still producing as much as you can. If you price too low you run into a shortage (and retailers get the money), if too high, the competition can take that as an opportunity to lower prices and still make a profit.
Like cascade-lake x 18-core is even on performance/$ a viable option because TR3 is so expensive. (if it were available).
This shows also AMD is seriously supply constraint.
EDIT:
Really same reason GPUs are so expensive. There is only so many huge dies Nvidia can get produced. If they were cheaper they would sell more than they can supply. If you can only sell so many, you price it high even if you want to win market share. And I think that's why AMD plays along. They don't have the capacity to produce enough dies if 5000 series would have been much cheaper.
It's probably not even about margins. If NV wanted they could have made every single AMD product a stupid choice simply on performance/$.
If that will be a huge success, something is really wrong with the industry.
I think what Charlie really means is that physically there's 56 cores available but there would only be 48 core models available due to the power consumption. But since performance and perf/watt is embarrassing compared to Rome they decided to add more SKUs (that OEMs will ignore) with all the cores enabled and the TDP at 300 or more so they can claim they are at least competitive on performance.
- Oct 16, 2019
- 2,746
- 6,653
- 146
Intel specifically said up to 56 cores socketed. They announced it the day before Rome launched.
Charlie22911
Senior member
- Mar 19, 2005
- 614
- 231
- 116
I’d also be surprised if this was anything other than a functional equivalent to a on-package “dual socket“ setup (better terminology escapes me at the moment).
If so, I wonder if their implementation will also have some of the weaknesses first-gen Threadripper had with latency.
moinmoin
Diamond Member
- Jun 1, 2017
- 5,234
- 8,442
- 136
Talking about shortages: What honestly baffles me is that so far Intel did nothing to resolve their capacity problems through smaller dies.
"In December of 2018, they admitted that it was due to increasing die size, which dropped the number of chips that they can produce. Intel's sweet spot, with 60% of sales, used to be 80-90 mm^2. AMD core increases drove that up to 125 mm^2, so they lost one third of their production capacity. Intel's original plan was to have a four core max at 14 nm. That is now 6 to 8 cores, even 10."
They could remove the iGPU, after all they even offer F chips with the iGPU fused off already. One reason they don't could be that the majority of the chips go into laptops.
"Laptops account for 70% of Intel units."
But even reducing the size of the iGPU would be an improvement. Instead the iGPU gets bigger in Ice Lake.
"The GPU is better, but it takes up a LOT of die area."
"In December of 2018, they admitted that it was due to increasing die size, which dropped the number of chips that they can produce. Intel's sweet spot, with 60% of sales, used to be 80-90 mm^2. AMD core increases drove that up to 125 mm^2, so they lost one third of their production capacity. Intel's original plan was to have a four core max at 14 nm. That is now 6 to 8 cores, even 10."
They could remove the iGPU, after all they even offer F chips with the iGPU fused off already. One reason they don't could be that the majority of the chips go into laptops.
"Laptops account for 70% of Intel units."
But even reducing the size of the iGPU would be an improvement. Instead the iGPU gets bigger in Ice Lake.
"The GPU is better, but it takes up a LOT of die area."
DrMrLordX
Lifer
- Apr 27, 2000
- 22,702
- 12,652
- 136
Cascade Lake-AP all over again. Yay!
@jpiniero isn't specifically disagreeing with you. He's saying that nobody will buy that SKU. They'll opt for a 48c SKU instead.
Intel is still married to the idea of iGPUs. It makes sense for a few applications, and depending on how well they implement OneAPI, it could leave to more GPGPU possibilities than QuickSync. That aside, they could keep more customers happy if they would just sell them iGPU-less chips and let said customers use mobile dGPUs.
Having the IGP saves a ton of power in mobile. Bigger is probally because Marketing thinks they need to be competitive with AMD in IGP Gaming on the top end model.
They do actually sell an F model for mobile, but that's only because the alternative is the Trash Can and the laptop models it goes into are aimed at discrete GPU gaming. At some point they will split the CPU and GPU into chiplets.
They do actually sell an F model for mobile, but that's only because the alternative is the Trash Can and the laptop models it goes into are aimed at discrete GPU gaming. At some point they will split the CPU and GPU into chiplets.
- Oct 16, 2019
- 2,746
- 6,653
- 146
Oh. In which case, it depends on if there's a SKU under 250W (the cutoff point for needing water-cooling).
Agree. Said it for years. The iGPUs are way too big and AMDs APU strategy wasn't very clever back then. If you need that much GPU power you will want a dGPU anyway and a fast CPU. iGPU should focus more on dedicated hardware like smartphone SOCs. decoding/encoding and such stuff. The 6 core and up also wouldn't really need an iGPU. Iin laptops does will be paired with a dGPU by OEM anyway so it's not really useful.
Arkaign
Lifer
- Oct 27, 2006
- 20,736
- 1,379
- 126
Yeah it's pretty stupid tbqh. Imagine if AMD insisted on cramming a Radeon on every Ryzen. There's a reason their biggest APU is a Quad core.
Glo.
Diamond Member
- Apr 25, 2015
- 5,930
- 4,991
- 136
Which is why in Q1 2020 APUs from AMD will have 8C/16T design with at least 12 Vega CUs.
Silly guys dissing on Charlie while bumping the thread into the limelight.
He's not the only one on the net with a alleged agenda. If you don't like what you read it's probably better to just ignore it or just let it die.
He's not the only one on the net with a alleged agenda. If you don't like what you read it's probably better to just ignore it or just let it die.
Charlie22911
Senior member
- Mar 19, 2005
- 614
- 231
- 116
They really should have just made the iGPU a separate chiplet, seems like the perfect use for emib.
I hope not because that would be a waste of good dies. If AMD wants a big piece of the laptop market they will need one with a pretty small die and low power use, meaning 4-cores tops and relatively small iGPU.
How many laptops actually sell with more than 4 cores? Very few. A good business laptop doesn't need 8 cores or a buffed up iGPU. battery life is more important and a lot of stuff unrelated to the CPU/APU. And same really for home users. There price is an important factor. All things a large APU will fail at.
Going after the low end hasn't worked out for AMD in the past. Why not go after the higher end and just use the cut down defective die for lesser offerings. In the end the cut down die with dead silicone will probably have more performance do to better heat dissipation. I'm not really sure it would be economically sound producing multiple die tiers just to satisfy the lower end.
Charlie22911
Senior member
- Mar 19, 2005
- 614
- 231
- 116
If you’ll allow me to nitpick: it’s silicon, not silicone...
The problem comes when the demand for lower end parts outstrips your ability to supply defective dice (dies?). The flagship parts are rarely, if ever, the volume product for a company. Also, the smaller your die, the more good parts you can get from a wafer.
Glo.
Diamond Member
- Apr 25, 2015
- 5,930
- 4,991
- 136
8C/16T+12 CU Vega iGPU would be small die. Like really small die compare to... other offerings.
Markfw
Moderator Emeritus, Elite Member
- May 16, 2002
- 27,096
- 16,014
- 136
I take it you mean, a 7nm monolithic die, and compared to Intels dies....
it's a silly coneIf you’ll allow me to nitpick: it’s silicon, not silicone...
The problem comes when the demand for lower end parts outstrips your ability to supply defective dice (dies?). The flagship parts are rarely, if ever, the volume product for a company. Also, the smaller your die, the more good parts you can get from a wafer.
Glo.
Diamond Member
- Apr 25, 2015
- 5,930
- 4,991
- 136
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,787
- 136
GPU is the same size in Icelake. Actually Icelake's GPU takes up 41mm2 and Whiskeylake's GPU takes up 43mm2.
Yes, they'll reduce it with the GT1 32EU parts in desktops.
But yea the primary motivation of having a good iGPU is on laptops, and especially ultrabooks and convertibles.
When 80% of what they sell don't have a dGPU at all, of course you can't ignore the iGPU.
This makes sense on desktops where no one even bats an eye even when it uses 5-10W extra on idle. And I assume Intel will offer some chiplet/MCM based strategy just so they can better compete with ballooning core counts on the AMD side.
But on laptops, they'll not only have to fend off against AMD, but ARM competition too. And the ARM contenders not only have good battery life, but hefty GPUs as well. The talk about custom 8cx in Surface Pro X having 2TFlops(FP16) means the shaders are comparable to what's in Icelake, and only TDP differentiates between the two.
GPU is the same size in Icelake. Actually Icelake's GPU takes up 41mm2 and Whiskeylake's GPU takes up 43mm2.I think he meant relative size. Since intel claimed 2.7 times density for 10nm vs 14nm, it'd be horrendous if the net size wouldn't be smaller, even if it has much more shaders.
And I assume Intel will offer some chiplet/MCM based strategy just so they can better compete with ballooning core counts on the AMD side.Some time in the distant future, if you mean desktop here
Finally some real competition and innovation on all possible sides
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,787
- 136
@lobz They can't avoid making the GPU larger, because people do care about the graphics part. In the ultrabook form factor, there's no other option, and if you are going to buy an expensive laptop, of course some will opt for the best option, and that includes the iGPU. Manufacturers charged $200-300 extra for the Iris option, and some people went for it. You have to have a well-rounded product, otherwise some people will go elsewhere.
And I assure you if they stop on the iGPU development altogether, even Nvidia will find a way to increase use of their dGPU even for ultrabooks and convertibles.
As for the size reduction, I assume CPU core has shrunk 2x and GPU close to 2.7x. Their GPU was was large for its performance level, so it was a much needed focus. They still need at least 1 more generation of efficiency gains.
Their desktop/server is quite messy, but eventually they'll go in that road.
And I assure you if they stop on the iGPU development altogether, even Nvidia will find a way to increase use of their dGPU even for ultrabooks and convertibles.
As for the size reduction, I assume CPU core has shrunk 2x and GPU close to 2.7x. Their GPU was was large for its performance level, so it was a much needed focus. They still need at least 1 more generation of efficiency gains.
Their desktop/server is quite messy, but eventually they'll go in that road.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter