Speculation: Ryzen 4000 series/Zen 3

[JoeRambo]

Well, that is what you get from 15% higher IPC.It is expected, if CPU core has several additional improvements.

Isn't IPC tangential to memory latency/BW ? ZEN2 already has IPC advantage over Skylake cores and very sizable deficit in games and AotS benchmark.
My point is, to get results these good, AMD must have done something to memory subsystem, like cutting quite some latency, core/L3 redesign comes as part of it and on top of that.

 

[eek2121]

It just occurred to me these chips are being announced in a bit over a week...

 

[Arzachel]

Isn't IPC tangential to memory latency/BW ? ZEN2 already has IPC advantage over Skylake cores and very sizable deficit in games and AotS benchmark.
My point is, to get results these good, AMD must have done something to memory subsystem, like cutting quite some latency, core/L3 redesign comes as part of it and on top of that.

Don't have to take the long way to memory if you hit the L3. It's harder to predict how much the unified L3 contributes compared to a straight up increase in cache size but I wouldn't be surprised if that's the single biggest upgrade in gaming benchmarks, even with the rumoured IF speed increase.

 

[JoeRambo]

Don't have to take the long way to memory if you hit the L3. It's harder to predict how much the unified L3 contributes compared to a straight up increase in cache size but I wouldn't be surprised if that's the single biggest upgrade in gaming benchmarks, even with the rumoured IF speed increase.

Some reddit guy ran VTune:

https://cdn.discordapp.com/attachments/727089095570096211/760421092124524564/unknown.png

Looks like straight and hard 30% memory bottleneck, for L3 cache to help, data has to get there somehow, either by evictions from L2 or prefetch from memory, and that prefetch must be correct too, or else it will compete for memory subsystem resources.
Unified L3 will surely help, the question was is that the only improvement they put in.

 

[Thunder 57]

Some reddit guy ran VTune:

https://cdn.discordapp.com/attachments/727089095570096211/760421092124524564/unknown.png

Looks like straight and hard 30% memory bottleneck, for L3 cache to help, data has to get there somehow, either by evictions from L2 or prefetch from memory, and that prefetch must be correct too, or else it will compete for memory subsystem resources.
Unified L3 will surely help, the question was is that the only improvement they put in.

I would hope not considering it's a "New architecture/family". Besides, it has to have SMT4, remember?

 

[cortexa99]

https://twitter.com/x/status/1310849002439663616

https://twitter.com/x/status/1310850653397413888

Seems a bit difficult to have a conslusion to say the least, but yeah if this leak is true, Zen3 would have a huge step forward at gaming performance, intel's lead at gaming is vanishing......

OTOH I agree this uplift would mostly due to latency optimization and not only just IPC increase.

 

[Thunder 57]

OTOH I agree this uplift would mostly due to latency optimization and not only just IPC increase.

I would agree with that. Coming soon, Intel says gaming is no longer a "real world" benchmark.

 

[TheGiant]

https://twitter.com/x/status/1310849002439663616

https://twitter.com/x/status/1310850653397413888

Seems a bit difficult to have a conslusion to say the least, but yeah if this leak is true, Zen3 would have a huge step forward at gaming performance, intel's lead at gaming is vanishing......

OTOH I agree this uplift would mostly due to latency optimization and not only just IPC increase.

to have such a difference between "std 10900K" and tuned 10700K the config of that 10900K must really suck

 

[Timorous]

I would agree with that. Coming soon, Intel says gaming is no longer a "real world" benchmark.

Intel will just say the only number that matters is TDP and Intel win there.

 

[exquisitechar]

to have such a difference between "std 10900K" and tuned 10700K the config of that 10900K must really suck

Don't think so, tuned memory is just huge in this benchmark.

 

[uzzi38]

Don't think so, tuned memory is just huge in this benchmark.

This. Tuned memory can account for a 50%+ improvement to performance when compared to stock.

 

[Nereus77]

Wow that 8 core looks like a beast! Beats the 10 core Intel right off the bat, amazing!
I wonder what a 5600 can do? Go toe-to-toe with an i7 9700K and beat it?

 

[Arzachel]

Some reddit guy ran VTune:

https://cdn.discordapp.com/attachments/727089095570096211/760421092124524564/unknown.png

Looks like straight and hard 30% memory bottleneck, for L3 cache to help, data has to get there somehow, either by evictions from L2 or prefetch from memory, and that prefetch must be correct too, or else it will compete for memory subsystem resources.
Unified L3 will surely help, the question was is that the only improvement they put in.

I'm not to worried about getting data into the caches, games loop over reasonably small data sets every frame with relatively small changes. You can see the impact cache has on memory bound games in the difference between Zen1 and Zen 2 in something like CSGO.

 

[Hans Gruber]

Remember, AMD said they solved the memory controller issues that plagued previous Zen CPU's with latency. I guess a single L3 cache is the way they are going with Zen3. The big question I have is regarding the coupled mode with the memory clock and fabric clock. Will performance be the same if unequal memory dividers are used. If coupled mode is still required. What will be the max memory frequency that Zen3 will support. They say Zen2 could do 3800mhz in coupled mode but I found 3600mhz worked best on my ancient B350. 3733mhz worked but memory ghost issues appeared which introduced stability issues for me.

Will Zen3 support memory @ 4000mhz or more? Or does Zen3 eliminate all the previous Zen limitations with regards to memory. It would be nice if Zen3 supports whatever speed your memory and motherboard can do.

 

[LightningZ71]

I feel like there are a few different things going on here. The unified L3 effectively doubles the L3 size as seen by any single process. So, if the game working set has high locality, but spills a bit from a 16MB L3, unifying the L3 to an effective 32MB will be a big help there. The second thing is that, from what rumors we've heard, the IF links between the CCD and the IOD retain the same number of pins. Before, there were two separate pathways between the two chips, one for one CCX, and one for the other. With a unified L3 and a single 8 core CCX, that means that there is a dual ported link between the CCD and the IOD, or, the single link has about twice the bus width. We also hear that the IF speed has increased. What the sum total of all of this means is that, in a situation where there are few heavy threads, such as in most games, there is a massive increase in available bandwidth between the individual core and the memory controller.

Granted, that's all based on rumors that have been passed around, but it certainly speaks to what we're seeing here.

I'd be interested in seeing the memory setup for that 5800x...

 

[cytg111]

Looking good looking good *reaching for wallet*

 

[Thibsie]

Checking how many weeks to save the little cents.... 5600/5600X should be good enough to replace my 2600.

 

[Panino Manino]

Looking good looking good *reaching for wallet*

Oh, right time for that price increase?

 

[cytg111]

Oh, right time for that price increase?

Its gonna last 5 years at least ... 200 bucks, meh.

 

[Thunder 57]

Checking how many weeks to save the little cents.... 5600/5600X should be good enough to replace my 2600.

Sort of on the same boat. Except I'm thinking I'll go to an 8 core version to upgrade my 2600X. Figure it will last awhile. RAM would probably be become a problem before that with a mere 16GB.

 

[swilli89]

I'm eager to see benchmarks and I hope the 5800x is tested with multiple RAM speeds from some reputable source.

I have DDR-4-2933 RAM and I'd like to not be forced to upgrade my sticks if I choose to upgrade my Ryzen 2700 to whatever the low-end 8-core Zen 3 ends up as.

 

[Thibsie]

Yep, more or less the same.
I have acceptable 3200 sticks, 2x16GB. And I really don't intend to replace them.

 

[dr1337]

https://cdn.discordapp.com/attachments/727089095570096211/760421092124524564/unknown.png

Where did this come from? Is this supposed to be zen 3? If so then they have made some great leaps and strides. Even if this is comet lake I think this is still interesting, and incredible if zen 3 hits 5.1ghz. Here's zen 2 for reference.

 

[Kenmitch]

Looks interesting so far. The 8th can't come soon enough. Of course some more leaks would be welcomed.

I have a 3700x currently and debating on upgrading it. I don't really need more power, but the tinkerer in me is interested. I guess it depends on the pricing in the end. I can always kick down the 3700x to my son's rig and sell off his 3600 to offset the cost.

It would be nice to see how far my b-die kit will go. It was a pointless adventure with the 3700x once the 1:2 divider kicked in. I figure somewhere around 4000-4200MHz @ C16 is doable.

 

[Justinus]

I'm just going to drop this in here for some evidence of the staying power of good old Haswell-E.

GPU Benchmark - Ashes of the Singularity

Video Card GPU Benchmarks, High quality testing of your video card and gaming system.

www.ashesofthesingularity.com

I've been waiting for Zen to eliminate the various real world performance deficits it's had vs. my 5960x. I think this might actually be the time I've been waiting for.