AlderLake anyone?

[JWNoctis]

Originally Posted by nordi

What do you mean? The y-cruncher results show Alder Lake in the lead, especially with DDR-5 RAM.

...I think I've been confused by the single-thread result. My bad.

Would be interesting if someone could try that again, but without E-core and with AVX-512 enabled.

[kriesel]

Originally Posted by chalsall

like the Spanish Inquisition. Only, different.

So, "does not include a free kitten" is a feature. What you described is why I prefer senior cats. Mellow. Content to watch TV or perch on lap or nap in a corner. Don't jump on top of the counter or table or cupboards. Don't chew live electrical cords or mouse cables. Etc.

[M344587487]

Originally Posted by Trilo

DDR5 is quad channel. I assume this helps with these benchmarks especially with y-cruncher being memory bound.

Technically each DDR5 DIMM has two channels, but each channel has half the width relative to DDR4 and all marketing will still call a single DIMM of DDR5 single channel, meaning the basic speeds are comparable between DDR4 and DDR5. Dual channel DDR5 at 3200 is twice the bandwidth of dual channel DDR4 at 1600. The win Alderlake with DDR5 naturally has for benchmarks is that they're specified to work with 4400/4800 speeds, compared to Alderlake and Ryzen specified to work with DDR4 at 3200. Most places will benchmark only at specified speeds, some benchmark with overclocked memory but this is mostly common for gaming which doesn't necessarily tax the memory like Prime95 would (the common overclock being around DDR5 5500, I've seen one that went with DDR5 6000 which looks to be near the fastest available).

[JWNoctis]

Originally Posted by M344587487

Most places will benchmark only at specified speeds, some benchmark with overclocked memory but this is mostly common for gaming which doesn't necessarily tax the memory like Prime95 would (the common overclock being around DDR5 5500, I've seen one that went with DDR5 6000 which looks to be near the fastest available).

Anything much faster than DDR5-6000 might not help much anyway, not with latency. Though I don't know whether that would actually do much, as far as Prime95 is concerned.

Next-gen processors with enough cache to run first-time tests without eviction couldn't come soon enough.

[VBCurtis]

Prime95 generally saturates memory bandwidth, so any increase in bandwidth via higher speeds is good for our use case. Latency isn't important to P95 speed.

You can test this by changing your own memory latencies in BIOS, if you have motherboard support to do so. Leave speed the same, slow down (higher number) latency, benchmark.

[Xyzzy]

We decided to put together a gaming computer so we could free up our YOLO computer for "real" work.

Several sources claim that DDR4 memory Z690 motherboards are (roughly) as performant as DDR5 motherboards in gaming, maybe because of the increased latency that DDR5 introduces. Given that DDR5 is also impossible to find/buy right now we went with a DDR4 board. We should have the parts within a week or so.

Here is what we have ordered so far.

CPU: Intel Core i5-12600K 3.7 GHz 10-Core Processor
CPU Cooler: Noctua NH-U12A chromax.black 60.09 CFM CPU Cooler
Memory: G.SKILL Ripjaws V Series 32GB (4 x 8GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Intel XMP 2.0 Desktop Memory Model F4-3600C14Q-32GVKA
Motherboard: MSI MAG Z690 TOMAHAWK WIFI DDR4 ATX LGA1700 Motherboard
Storage: Western Digital Black SN850 1 TB M.2-2280 NVME Solid State Drive
Case: Fractal Design Define 7 Compact ATX Mid Tower Case
Power Supply: SeaSonic FOCUS Plus Platinum 850 W 80+ Platinum Certified Fully Modular ATX Power Supply

OTOH, if someone here has a source for DDR5 memory we are willing to switch to a DDR5 build.

[Xyzzy]

[Xyzzy]

https://www.techpowerup.com/review/i...vs-ddr5/7.html

[ewmayer]

Originally Posted by Xyzzy

Several sources claim that DDR4 memory Z690 motherboards are (roughly) as performant as DDR5 motherboards in gaming, maybe because of the increased latency that DDR5 introduces. Given that DDR5 is also impossible to find/buy right now we went with a DDR4 board.

I continue to be pleasantly surprised by the performance of the budget refurb KNL system ($500) I bought a year ago ... the 16GB onboard MCDRAM is probably pretty close in terms of total bandwidth to a DDR5 module. Per AnandTech, the MCDRAM is capable of 400+ GB/sec ... how does that compare to high-end DDR5?

Best part I found out only in the past month ... when crunching on datasets much too large to fit into the MCDRAM (~200GB of p-1 stage 2 buffers in my case), the 192GB 2400MHz server-ddr4 I installed ($900) comes into play, and my initial timings showed a 2.5x performance hit vs MCDRAM-only. But fiddling with numactl flags I was able to get the OS to make much better use of the MCDRAM as a huge L3 cache, now down to just 1.5x slower than the throughput for the smaller MCDRAM-only dataset.

[Xyzzy]

Originally Posted by ewmayer

…how does that compare to high-end DDR5?

https://www.anandtech.com/show/16143...-and-latencies

Intel 12th Gen CPU → DDR5-4800 → 38.40GB/s × 2 (dual channel) = 76.8GB



Our YOLO computer → DDR4-3200 → 25.20GB/s × 4 (quad channel) = 100.8GB/s

[Xyzzy]

Originally Posted by Xyzzy

Memory: G.SKILL Ripjaws V Series 32GB (4 x 8GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Intel XMP 2.0 Desktop Memory Model F4-3600C14Q-32GVKA

This memory runs perfectly OOTB using its XMP profile. The only manual tweaking we did was to force it to a 1T command rate.

Overall, we ended up with DDR4-3600 14-14-14-34 1T Gear 1!