Will AMD win high end gaming now?; Ryzen 9 4950x specs leaked

[MageTank]

10 minutes ago, Sauron said:

There you go, ryzen just loses because of the significantly lower clocks as I mentioned.

I am a little confused. My post was to highlight the contrasting difference in single threaded scores when showing specific CPU's against one another.

 

@Moonzy said:

1 hour ago, Moonzy said:

zen 2 and intel scores about the same in cinebench single core scores, i think

but there's still clear difference in gaming somehow, i thought it was to do with latency? guess not?

Your response to this statement was:

1 hour ago, Sauron said:

Do they?

  Reveal hidden contents

 

Which served to refute Moonzy's original statement using a link to PCGamer's Cinebench results. I then posted my links to prove Moonzy somewhat correct in their original statement:

17 minutes ago, MageTank said:

Yes, if we want to be picky with our benchmarks and sources, lol.

 

https://www.pcgamer.com/amd-ryzen-7-3800xt-review/

 

For sake of transparency, you can see the 3900XT lose against the 10900K from the exact same source:

https://www.pcgamer.com/amd-ryzen-9-3900xt-review/

 

With a stronger IMC, AMD could easily catch up to or even exceed Intel's current gaming performance. Couple that with better core scaling over time and AMD is sitting in a pretty good spot right now.

 

As for Intel, they need to roll with the changes. Do away with trying to keep everything in-house on an aging fab and start outsourcing to other fabs. Design something that will produce higher yield rates to meet the supply demands and do away with the terrible marketing decisions plaguing their platform. Stop treating overclocking like it's a premium feature that must be paid for on both their CPU's and chipsets, and give consumers their options back. As much grief as people give Intel, they are still offering a good product, they are just shooting themselves in the foot with these artificial limitations and premature platform obsoletion. Plenty of people would buy a Core i5 10400 if it were overclockable on an H470 board, even if the quality of the silicon wasn't as great as a 10600k.

The point is, if we cherry pick specific AMD CPU's against specific Intel CPU's, we can absolutely say AMD has "caught up in single thread". Granted, this is still cherry picking, but it's not an entirely inaccurate statement either. If your original point in quoting Moonzy was to agree with them, then forgive me for misconstruing the intent of your post. 

 

As for AMD having "significantly lower clocks", I wouldn't consider this entirely accurate. Comparing single core turbo speeds, AMD isn't that far behind Intel. AMD's fastest turbo is 4.7ghz on the 3950X, 3900XT and 3800XT. Intel's fastest single core turbo speed is 5.3ghz on the 10900K. This 600mhz difference in clock speed might sound like a lot on paper, but it's really not. It's roughly a 13% difference in clock speeds and ignores various advantages and disadvantages with specific instruction sets. Couple this with the fact that game performance often does not scale linearly with clock speeds, and you'll see it's pretty insignificant in the grand scheme of things. I firmly believe something else is at play for the disparity in AMD's gaming performance at high frame rates relative to Intel's, and I strongly doubt that it's simply clock speed. This is why I lean towards the memory controller (and uncore in general), though I have no real concrete evidence to go with this claim other than some observations made when overclocking memory on Ryzen.

 

It scales so well with both frequency as well as adjustments to timings. This includes primary, secondary, as well as tertiary timings. The difference in performance between single vs multi-rank DIMM's is also far more pronounced on AMD than it is on Intel, so I am pretty sure a lot is left on the table when it comes to the IMC and uncore in general.

 

I am genuinely excited for Ryzen 4000, I just hope they really deliver on the IMC front.

[Sauron]

2 minutes ago, MageTank said:

The point is, if we cherry pick specific AMD CPU's against specific Intel CPU's, we can absolutely say AMD has "caught up in single thread". Granted, this is still cherry picking, but it's not an entirely inaccurate statement either. If your original point in quoting Moonzy was to agree with them, then forgive me for misconstruing the intent of your post. 

The conversation was more about whether the small advantage Intel has (or used to have when Zen 2 launched) in tasks that don't use a lot of cores despite the lower IPC was due to them just having faster clocks (which was my opinion) or to lower latency. Cinebench was mentioned as an example where AMD supposedly won out despite the much lower clock speeds (which would mean the reason is to be found elsewhere) but as far as I can tell that's not the case. XT chips are clocked higher so it stands to reason they'd close the already small gap.

[HenrySalayne]

33 minutes ago, MageTank said:

As for Intel, they need to roll with the changes. Do away with trying to keep everything in-house on an aging fab and start outsourcing to other fabs.

I personally don't think their current problems are related to old technology, but problems with their process, design and hence the yield. One of the reasons AMD switched to a chiplet design is to keep the size small enough. A faulty Intel die is a dead processor, a faulty AMD die is just a third of a processor and can be easily switched out. And AMD is to my knowledge still using an "older" but well optimized node for the IO chip.
 

[cj09beira]

5 hours ago, TheTechWizardThatNeedsHelp said:

It's supposed to be 4.8 all core boost. Anything is possible. 

Yeah. The main problem for current gen intel/and is that Intel kills on the IPC right now.

those are single core boosts, and the time of getting all core overclocks as high or higher than stock single core boosts are long gone mostly due to just how dense the heat becomes.

5 hours ago, Moonzy said:

all core boost? isnt that 1 or 2 core boost?

i thought AMD has better IPC but intel cpu has lower latency? i could be wrong though

intel has better latency thats true, but (and we will know for sure pretty soon how much of a problem it is) there is also the 4 cores per ccx problem, as communication between ccx's ins't fast, so workloads that need lots of communication between threads will suffer, how much this is a problem its still to be determined.

1 hour ago, Moonzy said:

zen 2 and intel scores about the same in cinebench single core scores, i think

but there's still clear difference in gaming somehow, i thought it was to do with latency? guess not?

latency might as well still be the biggest problem, as ryzen 2 does still gain a lot from fast memory and specially tight timings, and as i have been writing this i have noticed that there isn't much of a difference at all between 6 core chips and 8 core chips which points to the 4 core ccx not being too much of a problem (6 core chips use 3 core ccx's), only with the 3100x vs 3300x does it actually matter which is a extreme case where one has 2 2 core ccx's and the other 1 4 core ccx

[Vector0102]

35 minutes ago, HenrySalayne said:

And AMD is to my knowledge still using an "older" but well optimized node for the IO chip.

They're using GloFo 12nm (LP) for the I/O die in the 3000 series.

 

https://www.techpowerup.com/256511/amd-ryzen-3000-matisse-i-o-controller-die-12nm-not-14nm

[Taf the Ghost]

1 hour ago, porina said:

Certainly for my compute uses, the 4 core CCX was the biggest single hurdle stopping Zen 2 being perfect. Forces you to fragment workloads to get best throughput and limits job sizes.

 

I think I'll only get one sample of Zen 3 to play with, and will target the lowest bin of 8 core. The minor clock differences don't really matter for my purposes. It will most likely form the basis of my next gaming system along with a next gen GPU when they come out.

 

Going beyond 8 cores, by implication requiring more CCDs, will of course bring more performance in other use cases, but also reintroduces the fragmented nature of the design.

 

Got a link to that? 

It got taken down a while ago.

[Taf the Ghost]

47 minutes ago, Sauron said:

The conversation was more about whether the small advantage Intel has (or used to have when Zen 2 launched) in tasks that don't use a lot of cores despite the lower IPC was due to them just having faster clocks (which was my opinion) or to lower latency. Cinebench was mentioned as an example where AMD supposedly won out despite the much lower clock speeds (which would mean the reason is to be found elsewhere) but as far as I can tell that's not the case. XT chips are clocked higher so it stands to reason they'd close the already small gap.

The Lake-designs are narrower with faster caches. Zen is wider with larger caches. The places where things are purely latency bound, i.e. old single threaded games, you'll see the narrow/faster win out. Anything that requires heavy workloads will favor the wider/larger. This is basically IBM's still current PowerPC approach. 

 

Basically, AMD sells their Server CPU designs on desktop and they work very well. They don't scale quite as hard as the high-clocking, top-SKU Intel parts, but AMD doesn't really care about that. Neither does Intel, oddly enough. Everything is about that OEM market.

[Helpful Tech Witch]

Just now, comander said:

AMD has higher IPC on the desktop parts. 

 

ICL has the highest IPC of anything out but it's limited on core count and doesn't clock well. 

true. heres to even better IPCs

[Helpful Tech Witch]

10 minutes ago, comander said:

h Intel ahead on perf (though not by much) and AMD ahead on efficiency. 

True, but games evolving could make AMD better. Games now just use 4c usually, but in thw near future games may use 8c, maybe even could take advantage of a 64c

[bcredeur97]

I'm just waiting to replace my 3960X with something better. (the intel 3960X, not a threadripper)

I want something that makes my 3960X @ 4.6 feel like ancient technology. Nothing really does still honestly. it does bottleneck my 1080ti a little bit in some AAA titles @ 1440p but it's not too bad. I'm kinda happy I've been able to get away with this chip for so long tho.

[Derangel]

I wouldn't get too excited yet. There are going to be hundreds of various rumors and supposed "leaks", as well as half-assed hyped based off of alleged ES samples, and so on. Best to wait until there is official information lest you set yourself up for disappointment.

[HenrySalayne]

5 minutes ago, Derangel said:

I wouldn't get too excited yet. There are going to be hundreds of various rumors and supposed "leaks", as well as half-assed hyped based off of alleged ES samples, and so on. Best to wait until there is official information lest you set yourself up for disappointment.

AMD will most likely not pull an Intel and release a new generation without improvements. Maybe it will be 5% faster, maybe 20%, who knows. And the current generation of AMD processors is up to the task. If 16 cores (or up to 64 for HEDT) are not enough today, Zen 3 probably won't be the solution.

[The Benjamins]

1 hour ago, TheTechWizardThatNeedsHelp said:

True, but games evolving could make AMD better. Games now just use 4c usually, but in the near future games may use 8c, maybe even could take advantage of a 64c

Games been using 6c for a little while now, and there are already games that use 8c fairly well.

[Helpful Tech Witch]

1 minute ago, The Benjamins said:

Games been using 6c for a little while now, and there are already games that use 8c fairly well.

Yes, but the majority of games are still 4 core optimised.

[Derangel]

1 minute ago, TheTechWizardThatNeedsHelp said:

Yes, but the majority of games are still 4 core optimised.

Pretty much every modern game engine is threaded for more than 4 cores.

[The Benjamins]

 

4 minutes ago, Derangel said:

Pretty much every modern game engine is threaded for more than 4 cores.

 

6 minutes ago, TheTechWizardThatNeedsHelp said:

Yes, but the majority of games are still 4 core optimised.

I agree with Derangel, most games that has come out in the past 1-2 years has engines that can utilize more then 4c fairly well. lumping old games into it really skews teh numbers, you can just as much say most games still use 2 cores optimized.

 

but new games that just came out or coming out are already made with 8c in mind.

[DildorTheDecent]

4 hours ago, MageTank said:

With a stronger IMC

This is all I'd want tbh. The ability to push the FCLK would be great. The 1900MHz "wall" sucks. At least 2200MHz. Ideally topping out a 2500MHz since that would cover pretty much everybody.

[porina]

3 hours ago, Taf the Ghost said:

The Lake-designs are narrower with faster caches. Zen is wider with larger caches. The places where things are purely latency bound, i.e. old single threaded games, you'll see the narrow/faster win out. Anything that requires heavy workloads will favor the wider/larger. This is basically IBM's still current PowerPC approach. 

Intel are going in a similar direction. The Coves are wider and get more cache.

 

2 hours ago, comander said:

Tigerlake and RKL are out soon. I expect both to clock better than ICL and possibly to have better IPC.

That's what Intel essentially said for Tiger Lake at their Architecture Day 2020 reveal. Process allows more clock at a given voltage over ICL, and will take higher voltage also. IPC improvements from ram/cache upgrades.

 

2 hours ago, comander said:

Here and now AMD is getting close to 2x perf/watt.

At best that is a gross simplification and it is vastly more complicated than that. I'm sure you can find specific scenarios where that may be the case, but it certainly isn't the general case.

 

 

2 hours ago, TheTechWizardThatNeedsHelp said:

Games now just use 4c usually, but in thw near future games may use 8c, maybe even could take advantage of a 64c

Games have been downwardly scaleable for a long time. 4 cores are at a basic level of performance for today. They work, you can get an ok experience out of them, but already you need to have at least 6 cores to have a higher end experience. 

 

45 minutes ago, HenrySalayne said:

AMD will most likely not pull an Intel and release a new generation without improvements. Maybe it will be 5% faster, maybe 20%, who knows. And the current generation of AMD processors is up to the task. If 16 cores (or up to 64 for HEDT) are not enough today, Zen 3 probably won't be the solution.

As we've seen, the opposite is true. Not every task scales well to more cores. Sometimes you just want the fewer cores you have to do better.

 

3 minutes ago, DildorTheDecent said:

This is all I'd want tbh. The ability to push the FCLK would be great. The 1900MHz "wall" sucks. At least 2200MHz. Ideally topping out a 2500MHz since that would cover pretty much everybody.

I have to ask, why do you feel this need? Is it for better IF scaling, or do you want the higher ram bandwidth? Or both? I'd love to see far more ram bandwidth, but faster DDR4 isn't the solution. DDR5 can't come soon enough, and the other alternative I'd love to see is more channels on higher end consumer platforms. The line between consumer and HEDT is messier than ever now so it would be an ideal time to redefine it. 

[DildorTheDecent]

7 minutes ago, porina said:

I have to ask, why do you feel this need?

I just want to go fast.

[MageTank]

26 minutes ago, DildorTheDecent said:

This is all I'd want tbh. The ability to push the FCLK would be great. The 1900MHz "wall" sucks. At least 2200MHz. Ideally topping out a 2500MHz since that would cover pretty much everybody.

I want to be able to push big dual rank 4000mhz DIMM's in a 2DPC configuration without suffering from the performance penalty of unstrapping the FCLK ratio. We know Ryzen loves higher clocked memory. We know Ryzen loves dual rank DIMM's. We know Ryzen loves operating in a 2DPC configuration. I want it all, lol.

[Athan Immortal]

5 minutes ago, CircleTech said:

Then there's me still using a Dual x79 Thinkstation D30 from 2012 as their daily driver.

Different strokes for different folks matey, your Dual Xeons are a bit more unique especially for that time period So I bet they still maul through most workloads.

 

For me an increase in render speed of videos for example can make hours of productive difference a week, so in an ideal world I'd love a 3900x for a bump in cores too, but I'll settle for a 3700x if the price comes down after the 4000 series.

[hollyh88]

so.. ryzen 4000 (3000) will have the same core count as last ryzen? and before that? thats kinda lame. would have liked to see increased core count... and still no 5ghz? come on the 3600xt can get 4.5/4.6 all core the 4700g can do 4.6/4.7 all core. still no 5ghz with more power?? 
bit of a let down for me since i wanted to upgrade my cpu this time around.. 

[RejZoR]

1 hour ago, hollyh88 said:

so.. ryzen 4000 (3000) will have the same core count as last ryzen? and before that? thats kinda lame. would have liked to see increased core count... and still no 5ghz? come on the 3600xt can get 4.5/4.6 all core the 4700g can do 4.6/4.7 all core. still no 5ghz with more power?? 
bit of a let down for me since i wanted to upgrade my cpu this time around.. 

People fell into the same Gigahertz craze as with Pentium 4's back in the day. Clock is NOT everything. And Athlon XP proved that well back in the day running dramatically lower clocks and still kicking Intel's butt. Hell, I remember my AMD Thunderbird 1GHz being as fast as 1.2GHz Pentium 3's. And later Intel's Core CPU's proved that too. They came with very tame clocks yet they beat everything in their path with ease. You just don't need to push 5GHz if your CPU logic is designed in such a way that it can reach needed performance at only 4.7GHz. And Ryzen 3000 series was already there except serious edge cases. New series hitting same or even higher clocks by a bit and extra IPC on top, they'll be great performers.

[Helpful Tech Witch]

6 minutes ago, RejZoR said:

People fell into the same Gigahertz craze as with Pentium 4's back in the day. Clock is NOT everything. And Athlon XP proved that well back in the day running dramatically lower clocks and still kicking Intel's butt. Hell, I remember my AMD Thunderbird 1GHz being as fast as 1.2GHz Pentium 3's. And later Intel's Core CPU's proved that too. They came with very tame clocks yet they beat everything in their path with ease. You just don't need to push 5GHz if your CPU logic is designed in such a way that it can reach needed performance at only 4.7GHz. And Ryzen 3000 series was already there except serious edge cases. New series hitting same or even higher clocks by a bit and extra IPC on top, they'll be great performers.

Yeah. On photoshop a 4.3ghz 3600 beats a 5ghz 10600k. 700mhz slower, but still better. 

https://www.gamersnexus.net/hwreviews/3592-intel-i5-10600k-cpu-review-benchmarks-ryzen-5-3600-et-al

[Derangel]

2 hours ago, hollyh88 said:

so.. ryzen 4000 (3000) will have the same core count as last ryzen? and before that? thats kinda lame. would have liked to see increased core count... and still no 5ghz? come on the 3600xt can get 4.5/4.6 all core the 4700g can do 4.6/4.7 all core. still no 5ghz with more power?? 
bit of a let down for me since i wanted to upgrade my cpu this time around.. 

 

5Ghz doesn't matter and there really isn't need for core increases on mainstream processors right now.