Search results

Great article. Quick question: in the nginx CDN test on page 11, is the QAT accelerator being used on the Xeon 8490H?

It's crazy that they release these SKUs now and in a couple of months they are going to release Cooper lake (which is yet another refresh). The market will be awash with new SKUs which are barely distinguishable between each other

Yep. One thing I don't understand is this refresh getting higher clocks because of a node improvement (I mean like 14nm+++), or is it just because they increased the TDP?

BTW, Ice Lake SP HCC and XCC parts will not be available in 2020. This is according to internal Lenovo leak: Google Translate I suspect Intel will paper launch Ice Lake SP this year.

Isn't this the old one? Are HPe planning to release a PCIe4 version?

I think it's misleading to say that Intel Optane DC gives Intel an advantage over AMD. Memory prices have dropped so low that equipping a machine with Optane DC is now more expensive than using the equivalent DDR4 config (going by list prices). Also, Optane DC is incredibly slow compared to DDR4...

Oh right. They look small in the photo for some reason

The second last graphic shows what looks like workstation PCs. Are they launching EPYC workstations too? Also, it seems at least some of these have 160 PCIe lanes.

Ah yeah, I see now that all servers do the same thing. Cascade lake down clocks to 2666MHz with 2DPC. On another note, I was thinking EPYC Rome is meant to have a single PCIe root, that would theoretically mean that DMA and RDMA work a lot better than Cascade lake (which doesn't have single root)

You say it supports "8x DIMMs at DDR4-3200 speeds or 16x at DDR4-2933" Is that a limit of the motherboard, or is it a limit of the CPU? I mean do all EYPC systems have the same limitation?

Yeah, I wouldn't be surprised if AMD took over the virtualized server market. On top of the performance, there's security. It seems there is a new Intel side-channel vulnerability every month, which doesn't affect AMD.

But not all workloads. For instance, large databases won't fit into L3 cache. The latency is quite a lot more than for Xeons according to Anandtech: AMD Rome Second Generation EPYC Review: 2x 64-core Benchmarked

I was quite surprised that there is still latency differences depending on which DIMM is accessed by which core. I would have though that all that since all DIMMs are connected to the same I/O chip they would all be the same latency.

Yes, but the main advantage was that it was to be significantly cheaper. There still remains customers that need the persistence, but that is a niche use case.

"Platinum 9200 does not support Optane DCPMM" Not really a problem now, as the massive drop in DRAM prices have made Optane DIMMs mostly redundant

Really, Cascade lake is a price cut and a small clock increase. For instance, the old 4116 Silver has the same spec as the new 4214 (except small clock increase) and price is cut from $1000 to $700. This is as expected, because competition has better product (i.e Rome)

That's quite interesting. Thanks for the info.

EPYC market share is mostly Cloud providers and hyperscalers. I assumed that these customers buy their servers directly from ODMs (or am I mistaken?). My point is that they wouldn't be using Dell or HPE servers, so the theory that EPYC sales were held up by lack of servers from big OEMs is false.

Yes it was confirmed. Indeed RAM access will be more uniform, but that hardly matters. Really, you just want access to be fast as possible.

I seems the new architecture increases off-die bandwidth and reduced off-die latency. This will benefit non NUMA aware applications, like databases. However, since all DRAM access has to happen off-die, the latency for NUMA aware applications will certainly increase. Maybe, they are using...