No idea. Most of the Intel users who used to "debate" in the AMD threads seem to have rage quit due to various reasonsAnyone tried it on Intel yet?
ARL is supposed to consume much less power. I think you’re underestimating this.At Computerbase the 9950X is ahead of the 14900K both in ST by 6% and MT by 10%, and at 275W for the 14900K, as for ARL it will be eventually slightly ahead only in Cinebench and that s by using 250W.
Same as the 241W 12900K wich was trounced by the 142W 5950X in almost any MT bench excepted in Cinebench, but this time ARL wont have any ST advantage, if early numbers are accurate it will be crushed in WebXPRT4, an Intel friendly browsing bench, and Speedometer and sligthly below in GB, wich is telling about its SC perf.
ARL is supposed to consume much less power. I think you’re underestimating this.
And it’ll likely be a good bit better in gaming.
I think you have it backwards. The high end Intel chips will be made with 20A silicon. I think the low end Intel offerings are any CPU's below what we know as i3 CPU's. Those could be made with TSMC silicon.I wish you would start using the modern node names. RPL is on "Intel 7". Rumor is last I checked ARL will use 3N for higer end and Intel 20A for lower end chips. There is no "intel 5" and 20A is not "5nm".
Zne 5+ has been a popular idea since Zen 5 came out pretty crappy. The early ARL numbers aren't looking great though at least for ST. In MT Skymont seems to do well. I expect the 9800X3D will be the most popular when it comes out assuming AMD doesn't delay those too.
First, the bad news:
View attachment 105567
View attachment 105568
View attachment 105569
View attachment 105570
View attachment 105571
View attachment 105572
View attachment 105573
Most of these regressions I think are due to the latency.
The real kicker is that the 7950X@5200 MT/s is beating the 9950X@5600 MT/s in some of these tests !!!
Keeping that bit of craziness in view, I think the disappointment over 9950X is justified.
Now the good news:
9950X generally does pretty well in the Pugetbench test suite.
AMD Ryzen 9000 Content Creation Review
Introduction Following some launch delays, AMD has now released their entire AMD Ryzen 9000 Series of processors. These CPUs are based on AMD’s new Zen 5www.pugetsystems.com
Long time i'm not watching results on Pudget, because their results always difference with others. Something off with pudgetPudget results should always be taken with some salt
I think you have it backwards. The high end Intel chips will be made with 20A silicon. I think the low end Intel offerings are any CPU's below what we know as i3 CPU's. Those could be made with TSMC silicon.
Not even people payed to do it are interested in doing it, and even if they did, the goalpost for Zen 5 gaming performance would shift somewhere else.Anyone tried it on Intel yet?
The compute unit/tile is what they are talking about. Fabricated on N3 3nm silicon. That is only part of the CPU. The rest of the CPU will be on Intel 20A silicon.This should be taken to the Intel thread, but all of the "leaks" have said that all of the high end SKUs and the majority of SKUs for ARL will be on TSMC 3B. The 20A SKUs will be mid-range and low volume. Maybe they are all wrong, but that has been the consistent leak thus far.
I thought that only applied if you did the testing while dancing under the moonlight of the full moon wearing only a tinfoil hat?Not even people payed to do it are interested in doing it, and even if they did, the goalpost for Zen 5 gaming performance would shift somewhere else.
For instance, there is increasing evidence that Zen 5 performs up to 10% better in games when testing during a full moon. We hope reviewers will test this claim thoroughly in the next few days, otherwise we'll to wait for another month to get proper testing conditions. If true, this find could have astronomical implications.
That's interesting. I know for at least Unreal Engine - Compile shaders, it's a very memory bandwidth sensitive task. Guessing same may go for many of theseFirst, the bad news:
View attachment 105567
View attachment 105568
View attachment 105569
View attachment 105570
View attachment 105571
View attachment 105572
View attachment 105573
Most of these regressions I think are due to the latency.
The real kicker is that the 7950X@5200 MT/s is beating the 9950X@5600 MT/s in some of these tests !!!
Keeping that bit of craziness in view, I think the disappointment over 9950X is justified.
Now the good news:
9950X generally does pretty well in the Pugetbench test suite.
AMD Ryzen 9000 Content Creation Review
Introduction Following some launch delays, AMD has now released their entire AMD Ryzen 9000 Series of processors. These CPUs are based on AMD’s new Zen 5www.pugetsystems.com
Do you realize the above is using DDR4? Regardless it is not whether there will be memory bottleneck or not, it is more like 24/32 core Zen4/5 processor will beat 16 core Zen4/5 processor in multi threaded tasks. It is a better approach than trying to go 6Ghz 16 core CPU. There is no reason for AMD to concede multi threaded performance to Intel.But don't worry, I'm sure all the "AMD should release a 32 core desktop CPU with 2 channel memory people" don't care
Very true. AMD themselves made a reputation of creating MT monsters and getting defeated by Arrow Lake in MT tests would be kinda shameful.There is no reason for AMD to concede multi threaded performance to Intel.
I wrote clearly that you should support your claim as best you can.
That or AMD is appealing directly to their largest/most lucrative markets instead of attempting to appeal to everyone with their latest design. Months ago I was flabbergasted when @adroc_thurston suggested that Zen5 would be the last gen to share CCDs with server/workstation, but maybe now it makes more sense.To AMD employees who are also reading this thread.
I have seen various tests of the 9950x. After 2 years of work on ZEN 5, hundreds of changes and architecture improvements, it is obvious that in a hurry the engineers made a mistake somewhere and SOMETHING is seriously blocking the performance of this CPU. There is a bottleneck somewhere/something that effectively negates the potential of all these improvements in ZEN 5.
If we're considering DDR4 bandwidths here, and the 24c chip is still faster than the bandwidth constrained 64c, why would a 32c on dual channel DDR5 be so much worse? If anything it would be a lot faster with these newer architectures being (relatively) more bandwidth efficient. Or, at least, they'd be able to field 24c/48t on standard DDR5 just fine.But don't worry, I'm sure all the "AMD should release a 32 core desktop CPU with 2 channel memory people" don't care
View attachment 105584
The people saying “more coars because DDR5 is faster” are neglecting to remember that the new cores are more powerful, have more throughput, and thus require more bandwidth per-core. That seems pretty common sense?If we're considering DDR4 bandwidths here, and the 24c chip is still faster than the bandwidth constrained 64c, why would a 32c on dual channel DDR5 be so much worse? If anything it would be a lot faster with these newer architectures being (relatively) more bandwidth efficient. Or, at least, they'd be able to field 24c/48t on standard DDR5 just fine.
Everything has its gives and takes, Zen 5 doesn't do much for gaming yet phoronix finds it to be the fastest on the block for literally everything else.
View attachment 105602
A lot of people are calling it a flop because in select workloads its a marginal improvement, yet in others it makes a great deal of improvement. This same idea applies to every aspect of CPU design including absolute core counts. Some workloads need memory bandwidth and other will prefer more logic units in total. Its just the way she goes.
Of course if MLID calls his bottom 'The manufacturer' it would explain a lot.This "drivel" comes directly from AMD, supposedly. I would not call it so badly.
Sorry... DEI ?Somehow "DEI hires" are at the root of every corporate failure, but successful companies with DEI policies like Apple and Nvidia are ignored. Unless their stock falls, then it will be because of DEI!
Sorry... DEI ?
Oh yeah Bulldozer was a marketing failure. Calling their modules dual core was a failure and it cost them. And rightly so.AMD has failed before with products - e.g. Vega, RDNA3, Bulldozer. They weren't a failing of marketing, but a failing of engineering. Marketing could have saved at least RDNA3, but no amount of time would've fixed them. Ryzen 9000 isn't this huge failure like them, although from those mentioned it is most similar to RDNA3. It's just a disappointment.
DDR5 is roughly twice as fast as DDR4. Are the new cores on DDR5 twice as fast as the latest cores on DDR4? Do they require twice the memory bandwidth?The people saying “more coars because DDR5 is faster” are neglecting to remember that the new cores are more powerful, have more throughput, and thus require more bandwidth per-core. That seems pretty common sense?
Depends on what kind of workload and data you're using AVX512 for. If the amount data being processed is limited, it'll be kept in cache which fast and has very high memory bandwidth. Same as for other operations BTW. Not sure why AVX512 should be special in this regard.But clearly Zen5, especially when running in AVX 512 is a memory bandwidth hog.
Source for this claim?I think Zen5 highlights more than any previous generation the insufficiency of memory bandwidth for these 16 cores.
Because for ideal MT workloads, you can double performance by doubling number of cores. This is technically quite easily achievable by adding more CCX dies. So it's a relatively easy way to improve MT performance a lot. And even if the MT workloads are not ideal, you'll still get a massive MT performance increase by increasing number of cores.Why would you want to keep pushing those diminishing returns?