- Oct 9, 1999
- 4,230
- 2,017
- 136
With the release of Alder Lake less than a week away and the "Lakes" thread having turned into a nightmare to navigate I thought it might be a good time to start a discussion thread solely for Alder Lake.
They have what appears to be valid results in Alder Lake review arcticle. I guess numbers in the bench are due to running SPEC on Windows 10. Kinda valid as warning not to use Alder Lake with Windows 10 : )
Still, that makes them useless for IPC as discussed in this thread, as somehow Alder Lake seems to have brutal sized advantage with DDR4 in Win10 testing.
CPU 2021 Benchmarks - Compare Products on AnandTech
CPU 2021 benchmarks: Compare two products side-by-side or see a cascading list of product ratings along with our annotations.
Would look completely different.
If one works out the GHz ratios we get the same % ratios. Final nail, project finished.
This is what I meant as well to some degree - I think the response has already happened, and like @TheELF mentioned I am confused as to why the 5600x even continues to exist with the presence of the 5600G at ~$80 cheaper and a 5700G at roughly the same price. 5800X sitting somewhere above the 5700G might not makes sense to us but from a merchandising segment they fit into their slots of of cheapest (5600G) - up a step (5700G) - step into the "real" Zen 3 w/PCIe 4 etc - 5800x. I expect the prices available now will be held onto, but that is AMD stuff and the under $200 market? Another thread.
By lack of backlog I mean that there isn't stock sitting around creating a lot pressure - it's hard to imagine there is given the import situation. I would be really interested to hear what is going on with (negotiated) OEM prices but we don't get access to those generally
In terms of Alder Lake excitement - is the 12400 happening when again? End of Q1? Q2? Will that be close enough to the follow on product (Raptor Lake?) that we'll be slow walking at that point? Shopping at MC for a "minimal PC" right now it's hard to walk past the 10400 through the 11700K and maybe those products on discount might create some great entry level builds - obviously the 11700k should be cheaper than the 12400. It'll be "slower" but... value?For me the fact that it has the new gen GPU means that for my "normal people" builds it would be A-OK. I'll definitely consider *any* ADL SKU with 4C/8T as a minimum as well - will MC have the ADL i3 at $100 or less? If so - let's go! I am more excited about that. Even replacing 3770k through 4770K (what I will be targeting) should result in a nice uplift in performance and the modern chipset/firmware goodness.
I'd expect those 11500 through 11700K being on 14nm Intel will keep RKL in production for a bit? (tangential but they must have some plan for this, right? ha)
AMD will cut prices when they need to, not before. If sales over a longer period of time trend downward, you will see some price movement. Unless that happens, they won't bother. Their goal is to increase margins, and reducing prices conflicts with that goal. The reduced sales would have to eat significantly into their margins for it to happen. I suspect Zen3D will introduce a price realignment, but they won't take a margin hit. If anything, margins will increase.
Check out DDR5 Memory Performance Scaling with Alder Lake Core i9-12900K
DDR5-6000 CL36 is the fastest.
If you use DDR5-2400 CL36 you end up with only 1/2 of the performance.
The DDR5 memory controller looks good in my opinion.
DDR5-6000 CL36 is the fastest.
If you use DDR5-2400 CL36 you end up with only 1/2 of the performance.
The DDR5 memory controller looks good in my opinion.
Gotta love TPU for their extensive testing. The E-cores don't do any favors for DB performance.
- Oct 9, 1999
- 4,230
- 2,017
- 136
Their DDR4 3600 CL16 "sweet spot" is exactly what I went with. Even a blind squirrel gets lucky sometimes!
- Oct 9, 1999
- 4,230
- 2,017
- 136
With the understanding that the primary reason for Gracemont's inclusion on ADL is for increasing multithreaded performance while using a minimum die area, wouldn't the best use of resources for Raptor Lake to be adding HT to Gracemont? I have read that one of the major benefits of HT is that it can be added to a core without adding significant complexity (area) to the design.
Looking at the Gracemont block diagram it seems as though it is wide enough to support HT.
I'm thinking there could be a couple of reasons for Intel not including HT with Gracemont.
1. There simply was not enough area on the die and they didn't want to increase the size.
2. They reasoned that ADL performance would be more than competitive with AMD and decided to "save" HT for Gracemont for Raptor Lake.
3. ?
I'm curious as to some more informed opinions on this?
Looking at the Gracemont block diagram it seems as though it is wide enough to support HT.
I'm thinking there could be a couple of reasons for Intel not including HT with Gracemont.
1. There simply was not enough area on the die and they didn't want to increase the size.
2. They reasoned that ADL performance would be more than competitive with AMD and decided to "save" HT for Gracemont for Raptor Lake.
3. ?
I'm curious as to some more informed opinions on this?
Here is a very interesting graph made by Raichu.
Raichu compared power consumption in Cinebench R23 using different voltages.
Raichu tweet
So 8C at 3.6GHz should have 2*12.67W = ~25W power consumption, that's very positive for the mobile variant.
If It can work at only 0.8V and at least 2GHz, then It would mean => 25W / 3.6GHz * 2GHz * (0.8)^2 = 8.888W ~ 9W power consumption in R23 and you have 6W left for the 2 Golden Cove cores for a total of 15W.
Raichu compared power consumption in Cinebench R23 using different voltages.
Raichu tweet
So 8C at 3.6GHz should have 2*12.67W = ~25W power consumption, that's very positive for the mobile variant.
If It can work at only 0.8V and at least 2GHz, then It would mean => 25W / 3.6GHz * 2GHz * (0.8)^2 = 8.888W ~ 9W power consumption in R23 and you have 6W left for the 2 Golden Cove cores for a total of 15W.
Last edited:
biostud
Lifer
- Feb 27, 2003
- 18,251
- 4,765
- 136
Yeah, latency wise it is hard to improve the overall speed. But bandwidth is continuously being improved. And depending on which kind of software you run, you should decide whether the extra cost is worth it.
Makaveli
Diamond Member
- Feb 8, 2002
- 4,724
- 1,061
- 136
We will get there but will take another year +.
As we all know memory always starts slow and expensive.
By the time I go to a DDR5 platform in 2024/25 this should all be sorted.
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,785
- 136
No, probably not. It will greatly increase scheduler complexity. Also while SMT adds minimal die area, the validation required dramatically increases. The best way is to continue to increase the amount of the E cores, and improve interconnect and caches to minimize scaling losses.
That assumes a lot of things. It'll probably need to go to 1.6-1.8GHz to leave 6W for the Golden Cove cores. Remember the chipset and uncore also takes 2-3W, so you really only have 12-13W for the cores. Then either you cut the Golden Cove core portion to 3-4W, or you have 6-7W left for the 8 Gracemont cores.
Even if you assume increased efficiency from the process, Gracemont significantly improves perf/clock which will increase power.
By comparison the 6W Tremont runs at 2-2.1GHz in MT workloads. Based on some limited reviews, you need Tremont at 7W to keep the clocks identical to Goldmont Plus at 6W. Of course the chipset is less featured and lower power target as well.
Doesn't mean it's negative though. Cause they are very behind in U chips for MT. The e cores are basically Whiskeylake Core i7 bolted on top of Tigerlake.
Last edited:
Ok, you are right about Uncore, so I will modify my calculation.
4C at 3.6GHz 12.67W -> I will substract 2.67W for Uncore, that will leave 10W, for an 8 core It would mean 20W + 2.67W Uncore.
20W / 3.6GHz * 2GHz * (0.8)^2 = 7.1W(~0.9W per Gracemont core) + 2.67W for a total of 9.7W power consumption in R23.
That would leave 5W for the 2 Golden Cove cores or 2.5W per core.
Of course this is just an example, but I think 2GHz clockspeed should be possible for CB23 within 15W.
Regardless of how high which core clocks, we can expect a massive MT jump.
i7-1165G7 or i7-1185G7 15W TDP manages 3500-3679 points in CB R23.
R7 5800U manages 7480 points with 15W TDP.
Based on TPU graph of CB R23 a 2+8 ADL clocked at 2GHz should provide ~7166-7244 points or almost 2x MT performance compared to Tiger Lake.
If It could maintain only 1.6GHz, then that would mean only 5733-5795 points in CB R23, that's still >55% better score than Tiger Lake, but not really that great.
With such low clocks It's pointless to use both types of cores, It would be better to use only 8 Gracemont cores clocked at 2.4-2.7GHz instead, which should provide 6379-7176 points in CB R23 and still be within 15W.
So I have to wonder If Golden Cove cores will actually work during highly parallel work like Cinebench.
Last edited:
Arkaign
Lifer
- Oct 27, 2006
- 20,736
- 1,377
- 126
Lol I won't post the whole thing again, but yeah, I think it should be hardly surprising to anyone who has seen this play out so many times already. It's harder to find a memory gen leap that didn't at the very least have initial limited appeal, high cost, and eventually large improvement once it had time on market. I guess RDRAM counts as an odd one out due to collapsing on itself.
Sdram started weak, but reached nice levels. DDR1, ditto. Then DDR2, 3, and 4 all in turn started meh and eventually reached obvious advantage over time. Shock 😲🤔 Amazement 🙇🙇
DDR3 -> DDR4 transition was different tho. We had DDR4 on HEDT first and when SLK with DDR4 came out, the market was already there, there was no shortage of stock or SKU divercity of DDR4. I think noone really bothered to make DDR3 supporting motherboards for SKL, they were a rarity.
Now, due to general shortage or whatever situation is completely different, not only there are plenty of DDR4 motherboards for ADL, they have higher mind share at launch as well. Very good performance of ADL with DDR4 also has to do with it, as something like DDR4 4000C16 takes tuned DDR5 above 6000 to even match it.
Now, due to general shortage or whatever situation is completely different, not only there are plenty of DDR4 motherboards for ADL, they have higher mind share at launch as well. Very good performance of ADL with DDR4 also has to do with it, as something like DDR4 4000C16 takes tuned DDR5 above 6000 to even match it.
From the rules of CPU forum.
New rules in CPU forum, and reminder of current rules: Rules Updated 5-4-20
OK, I will keep this short and sweet. You all know the rules, or better learn to read them. Here are cliff notes: No baiting, flaming, trolling, thread crapping, cussing or insulting. So if you have an opinion you can state without insulting people, do so, stating it as an opinion. If you have...
forums.anandtech.com
You must post your own commentary to the link/image/story you have posted. This is your one-time zero point warning.
esquared
Anandtech Forum Director
Last edited by a moderator:
Arkaign
Lifer
- Oct 27, 2006
- 20,736
- 1,377
- 126
True. The focus on 12900K is less interesting to me overall compared to the entire Alder stack and what it represents.
We were all impressed with Zen3, but less so for it's extremely incomplete product stack and sizable barrier to entry for the contemporary 7nm Zen3 derivative.
The promise led by the extremely rare 3100/3300X Zen2 value options, of which I'm still yet to see one in person as a bare CPU or retail package (seen one or two OEM boxes back when they were newish), well, that evaporated with the launch of Zen3. Suddenly $300+ was the bare minimum to participate, abandoning the market for $100-$250 range parts.
The full Alder lineup seems to indicate a full stack from i3-F all the way to i9-K, and everything in between. Given how most PCs are used, and the general competence of IGP/APU for even a pair of 4K displays for home/office usage, this really represents something I hope AMD is going to return to offering SKUs for.
For example, I recently built out a workstation for a guy who owns an independent auto dealership, and his primary focus is running a number of Mannheim auction windows. About a decade ago, I built him an FX-8320 rig with 16GB, and it did well for him for a good while, and it got him in the door really cheap, at i3 pricing. But a few weeks back, the best option I could find was pairing a $179 i5 10400 with a quite nice $120 B560 (2.5Gbe, all solid state cap) and a 16GB kit of 3200 for $65. At Micro Center, the cheapest AMD option in stock was north of $300 just for the CPU, and that level of price premium didn't make sense. I expect the Alder i3 to be equal or better than that 10th Gen i5 for general use, and that will be what, $140ish? I know the full 12400 i5 is slated for ~$200.
Not everyone needs even that level of performance (12400/5600X level), so it's sure nice to have decent stuff in the sub $200 range.
Of course we do need B660 boards of decent quality to hit the $110ish range and of course the cheapo sub $90ish H610 boards to follow the non K value SKUs onto the market to complete this picture, but I expect that will come in time.
All this $300-$700 stuff is great, and honestly overkill for most people, it does get the headlines and fan wars going, but I don't find it all that relevant or interesting as a long time industry soldier just trying to make people happy.
On voltage and power, are we even 100% sure that the 2+8 Alder Lake parts are being manufactured on the EXACT same process mix as the 8+8, 6+8 and 6+0 parts are? TSMC has stated publicly that, even inside of an established process node, there are various "levers and switches" that can be adjusted to achieve specific goals with respect to the speed-efficiency-density triangle of overall performance. Is it possible that Intel has made those adjustments for the "mobile-first" 2+8 part to bias it towards the efficiency point of the triangle as opposed to the other parts?
There's sure going to be binning going on. Desktop parts are going to be high leakage, highest clocking parts with voltage. Notebook parts are going to be binned for low leakage. low voltage, and highest efficiency. I expect the mobile parts to show even better efficiency than we're seeing with undervolted desktop parts at iso clocks.
uzzi38
Platinum Member
- Oct 16, 2019
- 2,637
- 5,990
- 146
Hm?
Tiger Lake used leakier silicon for the i7s than it did for the i5s. Just look at all those 1135G7 vs 1165G7 comparisons people made which showed the 1135G7 leading on power efficiency at 15-25W especially.
If there's going to be benefits to efficiency compared to these results it's much more likely going to come from the different configuration (8+4 vs 6+8, assuming this even makes a notable difference) and lowered SoC power.
This voltage here is about as low as you should expect for mobile parts as well. ADL-S ships with 0.85v min stock voltage.
TRENDING THREADS
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 10K
-
Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)- Started by Vattila
- Replies: 13K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 7K
-
Facebook
Twitter