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.
Microcenter currently has the 5800X and 12600K priced at $300. Doesn't matter which one you like, that's an awful lot of compute for $300. Competition!!!
The point is that it didn't matter that Sweclockers used only 2.8 GHz to compare the cores in Cinebench R20 because GC had the same percentage higher score when Tweaker.nl tested it at 3.5 GHz or TweakTown did at 4 GHz.A benchmark that scales nearly perfectly with clock rate is terrible for IPC testing, as the "IPC" measured would be the same for every CPU! i.e. a CPU clocked twice as high gets numbers twice is high so dividing performance by clock rate ends up the same for both.
I dont see any other test sites that feel it is necessary to refrain from updating hardware in order to "keep variables to a minimum".We can count on Anandtech of all places to keep variables to a minimum between reviews.
If you don't get why that's important... 🤷♂️
I am certain that places like Tom's in upcoming DGPU reviews will show the 12900K in their scaling charts, and given they are looking at game performance at more commonly used presets vs using games as CPU benchmarks, that will likely prove way more useful for the type of data you are looking for.
All fair enough - but it seems like Steam Deck is going be pretty thermally/power constrained vs what we might see in a 65W type desktop chip. It sounds like DDR5 offers even greater potential than LPDDR5 on that front but I guess we'll find out in due time.
Yeah, my mistake. Didn't know UHD 770 was clocked higher on i9. 1450MHz is for i5. Another factor for the increased performance might be the increased single threaded performance making the driver give more work to do for the iGPU.11900k= 1300 Mhz
12900K= 1550 Mhz +19%
The difference is much higher in many of the tests there. 32% average, some games 50% faster with only 19% higher GPU clock, I wonder why. 32EU Xe isn't bandwidth limited with DDR4-3200, I don't think it's DDR5 related.
I dont see any other test sites that feel it is necessary to refrain from updating hardware in order to "keep variables to a minimum".
I worked in scientific research for over 30 years, so I *do* understand the concept of controls. If one wants to be a stickler for using appropriate controls, one should run the test using both the old dgpu and the latest one and compare results, not just assume they would yield the same results as you did. Or you could just use the latest generation of dgpus, which almost anyone who buys a 12900k will be using.
Anyway, enough off topic bashing of AT. I guess we will have to agree to disagree before the conversation degenerates into a series of personal insults.
Zen 2 has a DDR5 memory controller??! I never looked at Steam Deck specs before. So does this mean Zen3D might launch with DDR5 support?but the steam deck seems to match the bus width of a desktop dual channel DDR5 configuration (I think? "16 GB LPDDR5 on-board RAM (5500 MT/s quad 32-bit channels")
You sir, are wrong.
Zen 2 has a DDR5 memory controller??! I never looked at Steam Deck specs before. So does this mean Zen3D might launch with DDR5 support?
Anyone know/find out anything about this e-core only Handbrake score of 87fps? Suddenly 8 Gracemonts at 3.7GHz are better in this Handbrake test that 8 Rocket cores at 5GHz Obviously a mistake but also the one metric I'd really like to know the actual value.
View attachment 52417
Anyone know/find out anything about this e-core only Handbrake score of 87fps? Suddenly 8 Gracemonts at 3.7GHz are better in this Handbrake test that 8 Rocket cores at 5GHz Obviously a mistake but also the one metric I'd really like to know the actual value.
View attachment 52417
@guidryp - the Steamdeck is using slower memory that the M1. 5500 Mbps, I believe. I believe that total bus width can be higher depending on configuration of LPDDR5 vs DDR5, but the steam deck seems to match the bus width of a desktop dual channel DDR5 configuration (I think? "16 GB LPDDR5 on-board RAM (5500 MT/s quad 32-bit channels")
In any case, we should see slotted DDR5 in speeds massively above the standard speeds so long as the chipset and CPU nominally support them.
I mean, we'll never see that in an Anandtech review, but there are other sites, right? Haha, back to AT bashing.
Hopefully ADL will let us see what impact cranking up DDR5 speeds has as that technology matures/becomes available at retail at some of the uber frequencies we've heard about.
LPDDR is usually faster than desktop DDR slotted memory.
LPDDR is usually faster than desktop DDR slotted memory.
Current JEDEC, numbers are:
Desktop DDR5 - 4800 Mbps
Mobile LPDDR5 - 6400 Mbps (what the Apple M1 uses). This may be what Intel and AMD next generation APU are using in laptops with soldered in DDR.
You will need OC'd desktop DDR5 just to catch up with LPDDR5 in mobile.
I really think the most RDNA2 CU's you could expect will be 8, (Same as Steamdeck). It might even be less. AMDs only real concern here, is they want to make sure they beat Intels Laptop iGPU, and anything beyond that makes the dies larger than it needs to be, and the APU is a volume part.
If they go with 8 CU's then it should beat the Steam Deck APU because, as you say, less thermally constrained, but it won't put it in a new category or anything. It will be a bit better than Steam Deck.
To make a big jump in APU GPU performance requires a purpose built big GPU APU (See Apple M1 family), not the incremental updates of a general purpose laptop APU.
Probably the new AVX units. I see a ton of talk about Gracemont using a ton of power (is 48W power hungry? 🤣), but it seems to have outstanding AVX performance.
A lot of people simply dislike Intel.
EDIT: Intel needs to just keep iterating on atom, and avoid bloat. IPC increases, don’t push the frequency beyond what is sensible, keep the power usage at a moderate level, and things will work out.
On the "A lot of people simply dislike Intel." I would like to comment.Probably the new AVX units. I see a ton of talk about Gracemont using a ton of power (is 48W power hungry? 🤣), but it seems to have outstanding AVX performance.
A lot of people simply dislike Intel.
EDIT: Intel needs to just keep iterating on atom, and avoid bloat. IPC increases, don’t push the frequency beyond what is sensible, keep the power usage at a moderate level, and things will work out.
I DON'T think a lot of people hate Intel, just their recent (last several years) products, and some even these new ones.
Huh, if LPDDR5 vs DDR5 is like LPDDR4X vs DDR4 it won't help much if at all.
Bigger number is better right? Well, it's not.
They don't exactly make APUs for extreme memory OC'd DIY desktop users though.
Are you going to spend $400 on a high end memory kit, to squeeze 20% more out of iGPU. Even if you OC the heck out of everything, you would still fall short of a 1050 Ti.
Are you going to spend $400 on a high end memory kit, to squeeze 20% more out of iGPU. Even if you OC the heck out of everything, you would still fall short of a 1050 Ti.
Desktop Rembrandt, depending on how many CUs they leave enabled, should be faster than a 1050 Ti with fast enough DDR5.
If CU rumor counts are true, and you get very high speed DDR5 (6400+), you might beat a 1050 Ti with an APU that will likely only be shipping for desktops in 2023.
Signaling components for PCIe5 are more expensive, and it requires more board layers. We first saw this with PCIe4.For the folks who dont want to be part of the debate...
What is the reason of the expensive motherboard compared to a last gen board? In which order?
Update:
- PCIe5
- DDR5
- Increased power requirements?
- Increased socket pin out
Just BOM cost, no inflation related or scalping stuffs